{"dataset":{"version":"effort-benchmarks-1","collectedOn":"2026-10-18","id":"sha256-49305f3d640f7387a764d95d62f10bbce44aeff4a4012ddba973eddf1e57a707+sha256-443256461e035ccdee585c01fdbd68b4b26545d45e2bd6b2bd93e1e913a2d41a+sha256-32c6374f8a0cdb9cd64f17244f1a697dd35d31cff071002599f3e1226d1e027d+sha256-9b7bfe52f0cba33df0362a330cb12551ea2f5853bf843931b08dad98e5d8d15c+sha256-c4f243913106bee769c57849d3cb4662fee3097206d30d3c8d4d37d687f8e556+sha256-c6f9cb8b67b4289e79e621b65d529e5f57eee55d89d86915f397859092051dad+sha256-78cb5ce46bffe96a698a9a7117551eafb90b2c3d05066722f4a703be2680aa55+sha256-e34a2de55476481fa69ea734e4027620fbc14df27f896c3bcec6c07275b0da42","benchmarks":[{"id":"aa-aa-analyst-agent","title":"AA-AnalystAgent","sourceUrl":"https://artificialanalysis.ai/evaluations/aa-analyst-agent","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Share of tasks solved on all 5 of 5 attempts · 80 tasks, 5 attempts per task · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-aa-briefcase","title":"AA-Briefcase Elo","sourceUrl":"https://artificialanalysis.ai/evaluations/aa-briefcase","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"AA-Briefcase v1.1 is an agentic knowledge work benchmark developed by Artificial Analysis. AA-Briefcase Elo is a combined metric that aggregates rubric pass rate, analytical quality Elo and presentation Elo · Higher is better · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"elo"},{"id":"aa-aime-2025","title":"AIME 2025","sourceUrl":"https://artificialanalysis.ai/evaluations/aime-2025","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-apex-agents-aa","title":"APEX-Agents-AA","sourceUrl":"https://artificialanalysis.ai/evaluations/apex-agents-aa","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-artificial-analysis-long-context-reasoning","title":"AA-LCR v1.1","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-automationbench-aa","title":"AutomationBench-AA","sourceUrl":"https://artificialanalysis.ai/evaluations/automationbench-aa","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Share of task objectives completed with no guardrail violations · Higher is better · Benchmark developed by Zapier · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-critpt","title":"CritPt","sourceUrl":"https://artificialanalysis.ai/evaluations/critpt","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Benchmark developed by Argonne and UIUC, with contributions from 60+ researchers globally. CritPt is under review by its authors following external feedback · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-enterprise-ops-gym-aa","title":"EnterpriseOps-Gym-AA","sourceUrl":"https://artificialanalysis.ai/evaluations/enterprise-ops-gym-aa","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Overall task success rate across all eight enterprise domains (oracle tool mode) · Benchmark developed by ServiceNow Research · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-gdp-pdf","title":"GDP.pdf: All-pass Rate","sourceUrl":"https://artificialanalysis.ai/evaluations/gdp-pdf","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Share of attempts where every atomic criterion passed · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-gdpval-aa","title":"GDPval-AA v2.1 Leaderboard","sourceUrl":"https://artificialanalysis.ai/evaluations/gdpval-aa","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Elo rating for performance on real-world work tasks · Anchored to DeepSeek V4.1 Flash (max) at 1600 · Higher is better · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"elo"},{"id":"aa-global-mmlu-lite","title":"Global-MMLU-Lite","sourceUrl":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-gpqa-diamond","title":"GPQA Diamond","sourceUrl":"https://artificialanalysis.ai/evaluations/gpqa-diamond","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-harvey-lab-aa-v1.1","title":"Harvey LAB-AA v1.1: Hallucination-Gated All-Pass Rate","sourceUrl":"https://artificialanalysis.ai/evaluations/harvey-lab-aa","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Average over tasks of the share of the judge panel finding every rubric criterion passed, with a task zeroed on any material hallucination · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Stirrup; Harvey private dataset v1.1.0; 120 tasks across 24 practice areas; up to 200 turns; internet disabled; GPT-6 Sol, Grok 4.7 and Claude Opus 5.5 rubric judges; GPT-6 Sol High two-pass hallucination check; hallucination-gated all-pass averaged across judges","unit":"percent"},{"id":"aa-humanitys-last-exam","title":"Humanity's Last Exam","sourceUrl":"https://artificialanalysis.ai/evaluations/humanitys-last-exam","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-ifbench","title":"IFBench","sourceUrl":"https://artificialanalysis.ai/evaluations/ifbench","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-itbench-aa","title":"ITBench-AA","sourceUrl":"https://artificialanalysis.ai/evaluations/itbench-aa","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Combined public and private Site Reliability Engineering (SRE) tasks · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-livecodebench","title":"LiveCodeBench","sourceUrl":"https://artificialanalysis.ai/evaluations/livecodebench","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-math-500","title":"MATH-500","sourceUrl":"https://artificialanalysis.ai/evaluations/math-500","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-mlcr-aa","title":"MLCR-AA","sourceUrl":"https://artificialanalysis.ai/evaluations/mlcr-aa","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Share of tasks judged accurate, complete, and concise · Expert + compound (hard) sets · Benchmark developed by Wisedocs · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-mmlu-pro","title":"MMLU-Pro","sourceUrl":"https://artificialanalysis.ai/evaluations/mmlu-pro","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-mmmu-pro","title":"MMMU-Pro","sourceUrl":"https://artificialanalysis.ai/evaluations/mmmu-pro","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-omniscience","title":"AA-Omniscience Index","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"index"},{"id":"aa-omniscience-accuracy","title":"AA-Omniscience Accuracy","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"AA-Omniscience Accuracy (higher is better) measures the proportion of correctly answered questions out of all questions, regardless of whether the model chooses to answer · Evaluation results measured independently by Artificial Analysis","harness":"6000 questions; 1 attempt; correct-answer share; GPT-5.6 Luna medium grader","unit":"percent"},{"id":"aa-scicode","title":"SciCode","sourceUrl":"https://artificialanalysis.ai/evaluations/scicode","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Benchmark curated by scientists across 16 disciplines. SciCode is under review following an independent audit of the dataset · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-tau2-bench","title":"𝜏²-Bench Telecom","sourceUrl":"https://artificialanalysis.ai/evaluations/tau2-bench","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-tau3-banking","title":"𝜏³-Banking","sourceUrl":"https://artificialanalysis.ai/evaluations/tau3-banking","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Benchmark developed by Sierra Research · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-terminal-bench-science","title":"Terminal-Bench-Science 0.1","sourceUrl":"https://artificialanalysis.ai/evaluations/terminal-bench-science","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Share of tasks passed (pass@1), averaged over three repeats · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"mini-swe-agent; Terminal-Bench-Science 0.1.0; 70 tasks; pass@1 averaged over 3 repeats","unit":"percent"},{"id":"aa-terminalbench-hard","title":"Terminal-Bench Hard","sourceUrl":"https://artificialanalysis.ai/evaluations/terminalbench-hard","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-terminalbench-v2-1","title":"Terminal-Bench v2.1","sourceUrl":"https://artificialanalysis.ai/evaluations/terminalbench-v2-1","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Benchmark developed by the Laude Institute and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"Artificial Analysis published evaluation; see source for benchmark-specific settings","unit":"percent"},{"id":"aa-terminalbench-v4-0","title":"Terminal-Bench v4.0","sourceUrl":"https://artificialanalysis.ai/evaluations/terminalbench-v4-0","sourceName":"Artificial Analysis","sourceHash":"986d1c3496ceaf0462edf1bb33a977999f4a9b1de9423b9ee6ad2578f4e421b6","description":"Benchmark developed by the Laude Institute, Stanford, and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis","harness":"mini-SWE-agent v2.4.6; 66 tasks; pass@1 averaged over 3 repeats; 500 steps; task timeout up to 8 hours","unit":"percent"},{"id":"arc-v2semiprivate","title":"ARC-AGI-2","sourceUrl":"https://arcprize.org/media/data/leaderboard/v2.json","sourceName":"ARC Prize","sourceHash":"be40992860f9689f109759eaca49bf4e62cfad12c692b9ae238d3942b64ccf73","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"v2 Semi-Private","unit":"percent"},{"id":"deepswe-minisweagent","title":"DeepSWE v1.1","sourceUrl":"https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json","sourceName":"Datacurve","sourceHash":"a7c15d66288fd249c020b9931c017b92d1a3b90e480b3ff34974b752bd030019","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"mini-swe-agent","unit":"percent"},{"id":"livecode-generationpass1","title":"LiveCodeBench","sourceUrl":"https://livecodebench.github.io/performances_generation.json","sourceName":"LiveCodeBench","sourceHash":"17f9a1d69e3f6290fc8b967a4dd13fbb46e34dd05716fdee793780a1cfadd0c1","description":"Evaluation window 2024-08-01 to 2025-05-01; 454 shared tasks.","harness":"generation pass@1","unit":"percent"},{"id":"swepro-minisweagentstarredrows","title":"SWE-bench Pro public","sourceUrl":"https://labs.scale.com/leaderboard/swe_bench_pro","sourceName":"Scale","sourceHash":"7726aaa6ea59090a16cb24697d17d342223fc9cc33a15fafba9e3497e540133c","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"mini-swe-agent starred rows","unit":"percent"},{"id":"terminal-claudecode","title":"Terminal-Bench 2.1","sourceUrl":"https://www.tbench.ai/?version=2.1","sourceName":"Terminal-Bench","sourceHash":"a03472372fd4cbf34bd7f344ec2d593e27c2147c00cb3170d898afb07711c1c2","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"Claude Code","unit":"percent"},{"id":"terminal-codex","title":"Terminal-Bench 2.1","sourceUrl":"https://www.tbench.ai/?version=2.1","sourceName":"Terminal-Bench","sourceHash":"a03472372fd4cbf34bd7f344ec2d593e27c2147c00cb3170d898afb07711c1c2","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"Codex","unit":"percent"},{"id":"terminal-cursorcli","title":"Terminal-Bench 2.1","sourceUrl":"https://www.tbench.ai/?version=2.1","sourceName":"Terminal-Bench","sourceHash":"a03472372fd4cbf34bd7f344ec2d593e27c2147c00cb3170d898afb07711c1c2","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"Cursor CLI","unit":"percent"},{"id":"terminal-geminicli","title":"Terminal-Bench 2.1","sourceUrl":"https://www.tbench.ai/?version=2.1","sourceName":"Terminal-Bench","sourceHash":"a03472372fd4cbf34bd7f344ec2d593e27c2147c00cb3170d898afb07711c1c2","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"Gemini CLI","unit":"percent"},{"id":"terminal-minisweagent","title":"Terminal-Bench 2.1","sourceUrl":"https://www.tbench.ai/?version=2.1","sourceName":"Terminal-Bench","sourceHash":"a03472372fd4cbf34bd7f344ec2d593e27c2147c00cb3170d898afb07711c1c2","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"mini-SWE-agent","unit":"percent"},{"id":"terminal-terminus2","title":"Terminal-Bench 2.1","sourceUrl":"https://www.tbench.ai/?version=2.1","sourceName":"Terminal-Bench","sourceHash":"a03472372fd4cbf34bd7f344ec2d593e27c2147c00cb3170d898afb07711c1c2","description":"Original public board. Configuration and task coverage may differ; inspect source.","harness":"Terminus 2","unit":"percent"},{"id":"vulcanbench-v3-07-grok-fable-sol","title":"VulcanBench v3 · Report 07","sourceUrl":"https://vulcanbench.com/benchmarks/07-grok-fable-sol.html","sourceName":"VulcanBench","sourceHash":"23ab20a29d4cb43306916bf4fc0e930b9879ff349d9f75db81848048ee2ab8ff","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"vulcanbench-v3-08-kimi-k3","title":"VulcanBench v3 · Report 08","sourceUrl":"https://vulcanbench.com/benchmarks/08-kimi-k3.html","sourceName":"VulcanBench","sourceHash":"e3ed4e7d0814c697a5a023f99ee5e83d12171c90dfe9871a81e7268f67a3c573","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"vulcanbench-v3-10-opus5-effort","title":"VulcanBench v3 · Report 10","sourceUrl":"https://vulcanbench.com/benchmarks/10-opus5-effort.html","sourceName":"VulcanBench","sourceHash":"33329962c53eacbf23a6e70013c1bf2f6293ba59c223bdec66f139680300fb8b","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"vulcanbench-v3-12-qwen38-max","title":"VulcanBench v3 · Report 12","sourceUrl":"https://vulcanbench.com/benchmarks/12-qwen38-max.html","sourceName":"VulcanBench","sourceHash":"2542eee9f673faa4efa452a0eb95def2121e100c8751cc52f333adf72ce313d5","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"vulcanbench-v3-14-grok-46-effort","title":"VulcanBench v3 · Report 14","sourceUrl":"https://vulcanbench.com/benchmarks/14-grok-46-effort.html","sourceName":"VulcanBench","sourceHash":"11691f07bfe90116fefadd41f984adf0a9f5037048161f611e531fb102471742","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"vulcanbench-v3-17-qwen38-27b-effort","title":"VulcanBench v3 · Report 17","sourceUrl":"https://vulcanbench.com/benchmarks/17-qwen38-27b-effort.html","sourceName":"VulcanBench","sourceHash":"245a270a98c14e43b3fe014174b1a4c8f861fffd9950da801de1f24eef01a9db","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"vulcanbench-v3-18-glm53-zcode-harness","title":"VulcanBench v3 · Report 18","sourceUrl":"https://vulcanbench.com/benchmarks/18-glm53-zcode-harness.html","sourceName":"VulcanBench","sourceHash":"6ee71f445a2a123032ddae49765021be4f48c5b3583a1fa17481f77c78da025c","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"vulcanbench-v3-19-musespark-effort","title":"VulcanBench v3 · Report 19","sourceUrl":"https://vulcanbench.com/benchmarks/19-musespark-effort.html","sourceName":"VulcanBench","sourceHash":"bc3e4d5b8698329ebf0d3c5f3870231eb55c470551216336b3ea9b29eb8387a8","description":"Frozen v3, 23 post-cutoff repository tasks. Published pass@1; repeat counts and time budgets vary by report. Display only pending comparability review.","harness":"VulcanBench bare-bones API loop, deterministic hidden tests in Docker","unit":"percent"},{"id":"arc3-provider-adapter","title":"ARC-AGI-3 · provider-adapter","sourceUrl":"https://arcprize.org/results/openai-gpt-6-astra","sourceName":"ARC Prize","sourceHash":"56782a69fd47c4c218086427fe8eeef8929c8729ebb1876708b4b6b7ecfd5a5e","description":"ARC-AGI-3; Provider Adapter. Open each result for its exact effort, task subset, metric and protocol details.","harness":"Provider Adapter","unit":"percent"},{"id":"arc3-standard","title":"ARC-AGI-3 · standard","sourceUrl":"https://arcprize.org/results/openai-gpt-6-astra","sourceName":"ARC Prize","sourceHash":"56782a69fd47c4c218086427fe8eeef8929c8729ebb1876708b4b6b7ecfd5a5e","description":"ARC-AGI-3; Standard. Open each result for its exact effort, task subset, metric and protocol details.","harness":"Standard","unit":"percent"},{"id":"frontiercode-1.1-extended-chisel","title":"FrontierCode 1.1 · extended-chisel","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; chisel. Open each result for its exact effort, task subset, metric and protocol details.","harness":"chisel","unit":"percent"},{"id":"frontiercode-1.1-extended-claude-code","title":"FrontierCode 1.1 · extended-claude-code","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; claude-code. Open each result for its exact effort, task subset, metric and protocol details.","harness":"claude-code","unit":"percent"},{"id":"frontiercode-1.1-extended-codex","title":"FrontierCode 1.1 · extended-codex","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; codex. Open each result for its exact effort, task subset, metric and protocol details.","harness":"codex","unit":"percent"},{"id":"frontiercode-1.1-extended-grok-build","title":"FrontierCode 1.1 · extended-grok-build","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; grok-build. Open each result for its exact effort, task subset, metric and protocol details.","harness":"grok-build","unit":"percent"},{"id":"frontiercode-1.1-extended-mini-swe-agent","title":"FrontierCode 1.1 · extended-mini-swe-agent","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; mini-swe-agent. Open each result for its exact effort, task subset, metric and protocol details.","harness":"mini-swe-agent","unit":"percent"},{"id":"frontiercode-1.1-main-chisel","title":"FrontierCode 1.1 · main-chisel","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; chisel. Open each result for its exact effort, task subset, metric and protocol details.","harness":"chisel","unit":"percent"},{"id":"frontiercode-1.1-main-claude-code","title":"FrontierCode 1.1 · main-claude-code","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; claude-code. Open each result for its exact effort, task subset, metric and protocol details.","harness":"claude-code","unit":"percent"},{"id":"frontiercode-1.1-main-codex","title":"FrontierCode 1.1 · main-codex","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; codex. Open each result for its exact effort, task subset, metric and protocol details.","harness":"codex","unit":"percent"},{"id":"frontiercode-1.1-main-grok-build","title":"FrontierCode 1.1 · main-grok-build","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; grok-build. Open each result for its exact effort, task subset, metric and protocol details.","harness":"grok-build","unit":"percent"},{"id":"frontiercode-1.1-main-mini-swe-agent","title":"FrontierCode 1.1 · main-mini-swe-agent","sourceUrl":"https://cognition.com/frontiercode","sourceName":"Cognition","sourceHash":"ff0040778b7bdf69c705ad7474f771fb5e53cfaf5cf8573b3a4614153543d75f","description":"FrontierCode 1.1; mini-swe-agent. Open each result for its exact effort, task subset, metric and protocol details.","harness":"mini-swe-agent","unit":"percent"},{"id":"terminal4-claudecode","title":"Terminal-Bench 4.0 · claudecode","sourceUrl":"https://snorkel.ai/leaderboard/terminal-bench-4-0/","sourceName":"Terminal-Bench","sourceHash":"7874d750adfc0b71defe97865a383e3e2323ce493bf4af64b9549858770edd6d","description":"Terminal-Bench 4.0; Claude Code. Open each result for its exact effort, task subset, metric and protocol details.","harness":"Claude Code","unit":"percent"},{"id":"terminal4-codex","title":"Terminal-Bench 4.0 · codex","sourceUrl":"https://snorkel.ai/leaderboard/terminal-bench-4-0/","sourceName":"Terminal-Bench","sourceHash":"7874d750adfc0b71defe97865a383e3e2323ce493bf4af64b9549858770edd6d","description":"Terminal-Bench 4.0; Codex. Open each result for its exact effort, task subset, metric and protocol details.","harness":"Codex","unit":"percent"},{"id":"terminal4-grokbuild","title":"Terminal-Bench 4.0 · grokbuild","sourceUrl":"https://snorkel.ai/leaderboard/terminal-bench-4-0/","sourceName":"Terminal-Bench","sourceHash":"7874d750adfc0b71defe97865a383e3e2323ce493bf4af64b9549858770edd6d","description":"Terminal-Bench 4.0; Grok Build. Open each result for its exact effort, task subset, metric and protocol details.","harness":"Grok Build","unit":"percent"},{"id":"terminal4-minisweagent","title":"Terminal-Bench 4.0 · minisweagent","sourceUrl":"https://snorkel.ai/leaderboard/terminal-bench-4-0/","sourceName":"Terminal-Bench","sourceHash":"7874d750adfc0b71defe97865a383e3e2323ce493bf4af64b9549858770edd6d","description":"Terminal-Bench 4.0; mini-SWE-agent. Open each result for its exact effort, task subset, metric and protocol details.","harness":"mini-SWE-agent","unit":"percent"},{"id":"terminal4-musecode","title":"Terminal-Bench 4.0 · musecode","sourceUrl":"https://snorkel.ai/leaderboard/terminal-bench-4-0/","sourceName":"Terminal-Bench","sourceHash":"7874d750adfc0b71defe97865a383e3e2323ce493bf4af64b9549858770edd6d","description":"Terminal-Bench 4.0; Muse Code. Open each result for its exact effort, task subset, metric and protocol details.","harness":"Muse Code","unit":"percent"},{"id":"scale-enigma_eval-scaleofficialevaluation","title":"EnigmaEval · Scale official evaluation","sourceUrl":"https://labs.scale.com/leaderboard/enigma_eval","sourceName":"Scale","sourceHash":"d7fc5c40ebb3b1c7d28d1e6173d15cdd3c7914f9b7542181181db11571ad918c","description":"Scale EnigmaEval puzzle-hunt accuracy; source-reported reasoning effort.","harness":"Scale official evaluation","unit":"percent"},{"id":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","title":"HLE · text only · Scale official evaluation","sourceUrl":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","sourceName":"Scale","sourceHash":"a30fb83890f802bde2bc1faa14a89df4ab9aa95c8717b01b6e94bd57dc950275","description":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.","harness":"Scale official evaluation","unit":"percent"},{"id":"scale-humanitys_last_exam-scaleofficialevaluation","title":"HLE · full · Scale official evaluation","sourceUrl":"https://labs.scale.com/leaderboard/humanitys_last_exam","sourceName":"Scale","sourceHash":"49c299d11030ae562b49f836734f1d410972689a77191a9c4ca509601bff122c","description":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.","harness":"Scale official evaluation","unit":"percent"},{"id":"scale-sweatlas-qna-claudecode","title":"SWE Atlas · QnA · claude code","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-qna","sourceName":"Scale","sourceHash":"4f6b3900aa968330fd791a184508ec0cc4695681120933868a032ad45029ef33","description":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.","harness":"claude code","unit":"percent"},{"id":"scale-sweatlas-qna-codex","title":"SWE Atlas · QnA · codex","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-qna","sourceName":"Scale","sourceHash":"4f6b3900aa968330fd791a184508ec0cc4695681120933868a032ad45029ef33","description":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.","harness":"codex","unit":"percent"},{"id":"scale-sweatlas-qna-miniswe","title":"SWE Atlas · QnA · mini-swe","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-qna","sourceName":"Scale","sourceHash":"4f6b3900aa968330fd791a184508ec0cc4695681120933868a032ad45029ef33","description":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.","harness":"mini-swe","unit":"percent"},{"id":"scale-sweatlas-refactoring-claudecode","title":"SWE Atlas · Refactoring · claude code","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-refactoring","sourceName":"Scale","sourceHash":"b1173bebe189116f0afed5ace35e0225335e152564950fffeea56f85c53b24c2","description":"70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.","harness":"claude code","unit":"percent"},{"id":"scale-sweatlas-refactoring-codex","title":"SWE Atlas · Refactoring · codex","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-refactoring","sourceName":"Scale","sourceHash":"b1173bebe189116f0afed5ace35e0225335e152564950fffeea56f85c53b24c2","description":"70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.","harness":"codex","unit":"percent"},{"id":"scale-sweatlas-tw-claudecode","title":"SWE Atlas · Test Writing · claude code","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-tw","sourceName":"Scale","sourceHash":"eb7c285dc6bae02aed6afc9477cb860e6b9daf23a0272303bdd5b4c7cf0a5768","description":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.","harness":"claude code","unit":"percent"},{"id":"scale-sweatlas-tw-codex","title":"SWE Atlas · Test Writing · codex","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-tw","sourceName":"Scale","sourceHash":"eb7c285dc6bae02aed6afc9477cb860e6b9daf23a0272303bdd5b4c7cf0a5768","description":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.","harness":"codex","unit":"percent"},{"id":"scale-sweatlas-tw-miniswe","title":"SWE Atlas · Test Writing · mini-swe","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-tw","sourceName":"Scale","sourceHash":"eb7c285dc6bae02aed6afc9477cb860e6b9daf23a0272303bdd5b4c7cf0a5768","description":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.","harness":"mini-swe","unit":"percent"},{"id":"livebench-2026-06-25-AMPS_Hard","title":"LiveBench · AMPS Hard · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-code_completion","title":"LiveBench · code completion · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-code_generation","title":"LiveBench · code generation · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-connections","title":"LiveBench · connections · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-consecutive_events","title":"LiveBench · consecutive events · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-integrals_with_game","title":"LiveBench · integrals with game · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-javascript","title":"LiveBench · javascript · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench Mini-SWE-Agent, 250-step limit","unit":"percent"},{"id":"livebench-2026-06-25-logic_with_navigation","title":"LiveBench · logic with navigation · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-math_comp","title":"LiveBench · math comp · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-olympiad","title":"LiveBench · olympiad · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-paraphrase","title":"LiveBench · paraphrase · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-plot_unscrambling","title":"LiveBench · plot unscrambling · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-python","title":"LiveBench · python · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench Mini-SWE-Agent, 250-step limit","unit":"percent"},{"id":"livebench-2026-06-25-simplify","title":"LiveBench · simplify · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-spatial","title":"LiveBench · spatial · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-story_generation","title":"LiveBench · story generation · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-summarize","title":"LiveBench · summarize · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-tablejoin","title":"LiveBench · tablejoin · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-tablereformat","title":"LiveBench · tablereformat · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-theory_of_mind","title":"LiveBench · theory of mind · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-typescript","title":"LiveBench · typescript · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench Mini-SWE-Agent, 250-step limit","unit":"percent"},{"id":"livebench-2026-06-25-typos","title":"LiveBench · typos · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"livebench-2026-06-25-zebra_puzzle","title":"LiveBench · zebra puzzle · 2026-06-25","sourceUrl":"https://livebench.ai/","sourceName":"LiveBench","sourceHash":"bf9ce6130134893d61157c214865de13d265a5e30790ff6cd4e3142a25113799","description":"LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.","harness":"LiveBench official task evaluation","unit":"percent"},{"id":"swe-rebench-2025-08-tools","title":"SWE-rebench · 2025-08 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2025-09-tools","title":"SWE-rebench · 2025-09 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2025-10-tools","title":"SWE-rebench · 2025-10 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2025-11-tools","title":"SWE-rebench · 2025-11 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2025-12-tools","title":"SWE-rebench · 2025-12 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2026-01-tools","title":"SWE-rebench · 2026-01 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2026-02-tools","title":"SWE-rebench · 2026-02 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2026-03-tools","title":"SWE-rebench · 2026-03 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2026-04-tools","title":"SWE-rebench · 2026-04 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2026-05-tools","title":"SWE-rebench · 2026-05 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"swe-rebench-2026-06-tools","title":"SWE-rebench · 2026-06 · tools","sourceUrl":"https://swe-rebench.com/","sourceName":"SWE-rebench","sourceHash":"fbb1b659562843da266aa20ec2295253d727ccd9b9469c78ab32df6fad1fffed","description":"Average resolved rate across repeated runs, not best-of-N success. Disjoint complete calendar months within each model’s tested range; standardized harness and all-language task set.","harness":"SWE-rebench standardized ReAct tools","unit":"percent"},{"id":"epoch-chess-puzzles-2-0-0-exact","title":"Epoch · Chess Puzzles · v2.0.0","sourceUrl":"https://epoch.ai/benchmarks/chess-puzzles","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 2.0.0; scorer exact","unit":"percent"},{"id":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","title":"Epoch · FrontierMath Tier 4 · v2.0.0","sourceUrl":"https://epoch.ai/benchmarks/frontiermath-tier-4-v2","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY. FrontierMath was developed with OpenAI funding and unequal access to part of the question set; this result is from Epoch's evaluation.","harness":"Epoch AI Inspect; task 2.0.0; scorer verification_code","unit":"percent"},{"id":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","title":"Epoch · FrontierMath Tiers 1–3 · v2.0.0","sourceUrl":"https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY. FrontierMath was developed with OpenAI funding and unequal access to part of the question set; this result is from Epoch's evaluation.","harness":"Epoch AI Inspect; task 2.0.0; scorer verification_code","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-0-choice","title":"Epoch · GPQA diamond · v1.0.0","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.0; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-1-choice","title":"Epoch · GPQA diamond · v1.0.1","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.1; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-10-choice","title":"Epoch · GPQA diamond · v1.0.10","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.10; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-11-choice","title":"Epoch · GPQA diamond · v1.0.11","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.11; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-2-choice","title":"Epoch · GPQA diamond · v1.0.2","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.2; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-3-choice","title":"Epoch · GPQA diamond · v1.0.3","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.3; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-4-choice","title":"Epoch · GPQA diamond · v1.0.4","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.4; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-5-choice","title":"Epoch · GPQA diamond · v1.0.5","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.5; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-6-choice","title":"Epoch · GPQA diamond · v1.0.6","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.6; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-7-choice","title":"Epoch · GPQA diamond · v1.0.7","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.7; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-8-choice","title":"Epoch · GPQA diamond · v1.0.8","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.8; scorer choice","unit":"percent"},{"id":"epoch-gpqa-diamond-1-0-9-choice","title":"Epoch · GPQA diamond · v1.0.9","sourceUrl":"https://epoch.ai/benchmarks/gpqa-diamond","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.9; scorer choice","unit":"percent"},{"id":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","title":"Epoch · Mystery Game Puzzles · v1.0.4 · Tool submission","sourceUrl":"https://epoch.ai/benchmarks/mystery-game-puzzles","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer","unit":"percent"},{"id":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","title":"Epoch · Mystery Game Puzzles · v1.0.4 · Exact move","sourceUrl":"https://epoch.ai/benchmarks/mystery-game-puzzles","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-0-0-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.0.0","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.0; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.0.0","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.0; scorer model_graded","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-0-2-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.0.2","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.2; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-0-3-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.0.3","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.3; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-0-4-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.0.4","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.4; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-0-5-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.0.5","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.5; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-1-2-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.1.2","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.1.2; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-1-3-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.1.3","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.1.3; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-1-4-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.1.4","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.1.4; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-1-5-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.1.5","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.1.5; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","title":"Epoch · OTIS Mock AIME 2024-2025 · v1.1.6","sourceUrl":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match","unit":"percent"},{"id":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","title":"Epoch · SimpleQA Verified · v1.0.0","sourceUrl":"https://epoch.ai/benchmarks/simpleqa-verified","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer","unit":"percent"},{"id":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","title":"Epoch · SimpleQA Verified · v1.2.0","sourceUrl":"https://epoch.ai/benchmarks/simpleqa-verified","sourceName":"Epoch AI","sourceHash":"7a50ffc96c8241360d7ff5b9c20e3d8b4b07d0c07f2f0dfe86f8c690d91fb91d","description":"Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer","unit":"percent"},{"id":"vals-finance-agent-v2","title":"Vals AI · Finance Agent v2","sourceUrl":"https://www.vals.ai/benchmarks/fabv2","sourceName":"Vals AI","sourceHash":"1194892ace4dd4b2360481348c579dfec40b305238340fb3082b652e7cd0387f","description":"Overall weighted rubric accuracy, averaged over three runs on 450 tasks. Dealbreaker failures and timeouts score zero. Related categories and all-pass scores are not additional votes. Three-model judge panel; private test set.","harness":"Vals FAB v2 shared six-tool harness; two-hour task limit","unit":"percent"},{"id":"vals-code-migration-v1","title":"Vals AI · Code Migration v1","sourceUrl":"https://www.vals.ai/benchmarks/code-migration","sourceName":"Vals AI","sourceHash":"b0c97d8038f01e6effdf134d79947787ff033e1ed43c587649001d97d17e4fea","description":"Hidden behavior-test pass rate, equally weighted across 40 repositories (30 CLI and 10 COBOL-to-Java). Anti-cheat failures score zero. Code-quality ratings and language splits are not additional votes.","harness":"Vals Code Migration shared agent harness","unit":"percent"},{"id":"vals-legal-research-v1","title":"Vals AI · Legal Research Bench v1","sourceUrl":"https://www.vals.ai/benchmarks/legal_research","sourceName":"Vals AI","sourceHash":"217d3801458cb32c305acd244c979cf25fd8fbce4c578f9fcb80c07b9a64abc3","description":"All-pass accuracy on 208 private legal research tasks. Every expert rubric check must pass; GPT-5.4 judges responses. Partial-credit scores and practice-area splits are not additional votes.","harness":"Vals Legal Research shared five-tool harness; three-hour task limit","unit":"percent"},{"id":"vals-emb-v1","title":"Vals AI · Excel Modeling Benchmark v1","sourceUrl":"https://www.vals.ai/benchmarks/emb","sourceName":"Vals AI","sourceHash":"99dae985e8543131005d7251ce7b44fe34347c1a85f5aa47b552208c58263e35","description":"Mean of seven equally weighted task categories. Template submissions use recalculated Excel cell accuracy; scratch submissions use rubric checks judged by GPT-5.4 Mini XHigh. Partial credit does not mean task completion.","harness":"Vals EMB shared workbook agent harness; 3.5-hour task limit","unit":"percent"},{"id":"vals-mmlu-pro-v1","title":"Vals AI · MMLU-Pro v1","sourceUrl":"https://www.vals.ai/benchmarks/mmlu_pro","sourceName":"Vals AI","sourceHash":"3c5bdffa902fbf90b8fff9c6785dfb54d0a2b6ce0bf788c4f11b1356ac8c6c2c","description":"Mean accuracy across 14 subjects under Vals five-shot prompting. Subject splits share one family and are not additional votes. Vals labels this benchmark archived; it is historical corroboration, not ongoing frontier evaluation.","harness":"Vals MMLU-Pro five-shot chain-of-thought; regex answer grading","unit":"percent"},{"id":"vals-terminal-bench-2-1-v2.1","title":"Vals AI · Terminal-Bench 2.1","sourceUrl":"https://www.vals.ai/benchmarks/terminal-bench-2-1","sourceName":"Vals AI","sourceHash":"1c7e1c245f8246c31fc0e34fb9f4de75dacd97f730c41b85722ac04664b3a89e","description":"Overall task pass rate under the shared Terminus 2 configuration. All task tests must pass. Difficulty splits share one family. Fable 5 and Opus 5 mixed-checkpoint fallback runs are excluded.","harness":"Vals Terminus 2 harness; pass@1 on 89 tasks","unit":"percent"},{"id":"vals-vibe-code-bench-v1.1","title":"Vals AI · Vibe Code Bench v1.1","sourceUrl":"https://www.vals.ai/benchmarks/vibe-code","sourceName":"Vals AI","sourceHash":"6282538768cb400b0e99e22f0d1f22c98a2447c82d33bc73e38254f51f6c1896","description":"Mean app test pass rate on the canonical v1.1 test set. Browser Use executes workflow tests; each test requires at least 90% of its substeps. Automated judging can introduce error. Alternate agents are excluded.","harness":"Vals modified OpenHands harness; five hours or 1,000 turns per app","unit":"percent"},{"id":"vals-legal-agent-benchmark-v1","title":"Vals AI · Harvey Legal Agent Benchmark v1","sourceUrl":"https://www.vals.ai/benchmarks/hlab","sourceName":"Vals AI","sourceHash":"208b955fcb15b58c24847fbe0cbae4c17675c96a9b6355deca3742b0bf1d99e1","description":"Mean task resolution across two judges (GPT-5.5 Medium and Sonnet 4.6), requiring every criterion to pass. Criteria pass rates and practice-area splits do not add votes. Shared Harvey family; Fable 5 fallback run excluded.","harness":"Vals Harvey held-out legal file-work protocol; internet disabled","unit":"percent"},{"id":"vals-tax-agent-bench-v1","title":"Vals AI · Tax Agent Bench v1: Weighted Pass","sourceUrl":"https://www.vals.ai/benchmarks/tax_agent_bench","sourceName":"Vals AI","sourceHash":"8169902770e7b0625af8bfda6e114e2bd185e7322dffad882a854f3f88656f64","description":"Weighted Pass on 193 private US corporate-tax questions. Each task score is the weighted rubric share, set to zero when a must-pass check fails, then multiplied by (0.7 + 0.3 times citation quality). Timeouts score zero. Category splits and all-pass rates are not additional votes. GPT-5.4 judges expert rubrics.","harness":"Vals Tax Agent Bench shared five-tool harness; three-hour task limit","unit":"percent"},{"id":"real-swe-gpt-6-astra-codex-cli","title":"Real-SWE · GPT-6 Astra · Codex CLI","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Codex CLI; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"real-swe-fable-5-1-claude-code","title":"Real-SWE · Fable 5.1 · Claude Code","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Claude Code; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"real-swe-gemini-3-8-flash-gemini-cli","title":"Real-SWE · Gemini 3.8 Flash · Gemini CLI","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Gemini CLI; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"real-swe-glm-5-3-claude-code","title":"Real-SWE · GLM 5.3 · Claude Code","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Claude Code; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"real-swe-muse-spark-1-3-muse-code","title":"Real-SWE · Muse Spark 1.3 · Muse Code","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Muse Code; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"real-swe-grok-4-6-grok-build","title":"Real-SWE · Grok 4.6 · Grok Build","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Grok Build; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"real-swe-gpt-5-6-sol-codex-cli","title":"Real-SWE · GPT-5.6 Sol · Codex CLI","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Codex CLI; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"real-swe-kimi-k3-kimi-code","title":"Real-SWE · Kimi K3 · Kimi Code","sourceUrl":"https://realswe.withspecific.com/","sourceName":"Specific Labs","sourceHash":"5fe0e104a345457a9235bd82250f5fba599836599cf0f1cc4d049dfd0236bb28","description":"September 2026 Specific Labs board. Published resolution rate, pass@1 averaged over 8 runs on 10 private tasks. Display only: this row is one model with its native harness, not a bare-model score, and it is excluded from weighted Overall.","harness":"Kimi Code; high reasoning; native product harness; Harbor verifier injected at grade time","unit":"percent"},{"id":"aa-cyber-index","title":"Artificial Analysis Cyber Index","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-cyber-index","sourceName":"Artificial Analysis","sourceHash":"cd84bd8f32e126648354ae64336e344fbc756d4269f9e341d6a3a8edaaa62883","description":"v1.0. Equal mean of CWE-Bench-AA, DeepsecBench-AA and CyberGym-E2E-AA. Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall. https://artificialanalysis.ai/methodology/cyber-index","harness":"Artificial Analysis Stirrup; defensive code audit, patch and end-to-end vulnerability tasks","unit":"index"},{"id":"autoresearchexam","title":"AutoResearchExam","sourceUrl":"https://benchmarks.bespokelabs.ai/autoresearchexam/","sourceName":"Bespoke Labs","sourceHash":"3d089595005ac25718305d4daa79c9ba8461875c0247af563f7282fa8b43f72b","description":"Hidden-test AUARC on 29 Harbor tasks. Terminus 2, 24-hour window, sealed tests. Display only. Excluded from weighted Overall.","harness":"Terminus 2 autoresearch harness in Harbor. Agents see validation rewards only.","unit":"index"},{"id":"bughunt-bench","title":"Bug Hunt Bench","sourceUrl":"https://bughunt.productcompass.pm/?preset=featured","sourceName":"Bug Hunt Bench","sourceHash":"7bd7acb9be79b118afaf94978253aff67c36530b2b07a8d0b6d5ecc0f5d84678","description":"105 planted bugs across two production repositories (45 and 60). Score is (fixed / 105) * 100, counting planted bugs fixed only. Primary rows are featured and not superseded. Display only pending comparability review. Many rows are single runs.","harness":"Native agentic CLI per row, blind judge against a withheld key. The CLI is named in each configuration.","unit":"percent"},{"id":"dayjob-healthcare","title":"DAYJOB Healthcare","sourceUrl":"https://surgehq.ai/benchmarks/dayjob-healthcare","sourceName":"Surge AI","sourceHash":"878a19ba735f5dda2774f324e3dc6b48a44eed20d6a57eb7314e465ce0efb5d3","description":"DAYJOB Healthcare strict pass@1 percent on the public Surge board. tau = 1, five attempts per task, 50 tasks in the paper and the Hugging Face dataset card. The page meta description says 65 assignments. That count is not on the scoreboard and is not used as n. An attempt passes only when every binary rubric criterion is met. Finance is a separate benchmark. Display only. Not an average and not GDPval-AA.","harness":"OpenHands SDK 1.43.1 inside Harbor 0.22.0. Judge is Claude Opus 4.8. Score is the board's strict pass@1 percent.","unit":"percent"},{"id":"dayjob-finance","title":"DAYJOB Finance","sourceUrl":"https://surgehq.ai/benchmarks/dayjob-finance","sourceName":"Surge AI","sourceHash":"f8eb6bb9e55a713b5b56aec2b53a233a161ade7b1a5ba55ca90d4c6015fbc0ea","description":"DAYJOB Finance strict pass@1 percent on the public Surge board. tau = 1, five attempts per task, 80 tasks in the paper and the Hugging Face dataset card. An attempt passes only when every binary rubric criterion is met. Healthcare is a separate benchmark. Display only. Not an average and not GDPval-AA.","harness":"OpenHands SDK 1.43.1 inside Harbor 0.22.0. Judge is Claude Opus 4.8. Score is the board's strict pass@1 percent.","unit":"percent"},{"id":"sudo-l7","title":"sudo L7","sourceUrl":"https://surgehq.ai/benchmarks/sudo-l7","sourceName":"Surge AI","sourceHash":"bf8adfadf9f6c1f5b17ee50dac34fc09389952454035c5737b2e1a16aa95d231","description":"Published percent on the sudo L7 board. The hero counts 60 tasks. Display only. Not in the weighted ranking.","harness":"The board names Claude Code, Codex, Qwen Code, Antigravity, Grok Build, Muse Code, OpenCode.","unit":"percent"},{"id":"toolathlon-verified","title":"Toolathlon-Verified","sourceUrl":"https://toolathlon.xyz/docs/leaderboard","sourceName":"HKUST NLP","sourceHash":"7dd93cf445f84b794c48931efddfbfa69f7be3b0ba4cacc1726f3f8356fe10a2","description":"Official Toolathlon-Verified board, released June 30, 2026. Same 108 tasks, new score series, not comparable with earlier Toolathlon. Headline is printed Pass@1 on the shared Default agent. Display only. The ± is recorded in the configuration and is not the score. Historical rows are excluded.","harness":"Shared Default agent. One harness for the board so Pass@1 rank matches the published table. Effort stays on the result.","unit":"percent"},{"id":"thinkingbox-bench","title":"ThinkingBox-Bench","sourceUrl":"https://huggingface.co/blog/microsoft/thinkingbox","sourceName":"Microsoft and Hugging Face","sourceHash":"08e9153724b0f297cafbf975d3dee118ca8cb03be61e2a2608fef7f4d3b3118a","description":"Task-weighted pass@1 on 507 stateful business workflows, 20 trials each. The stored score is the published overall from the 3 October 2026 blog. Display only. Excluded from weighted Overall.","harness":"ThinkingBox sandbox. Shared prompts, MCP tools, simulated user, and executable checks over terminal backend state.","unit":"percent"},{"id":"frontiermath-erdos","title":"FrontierMath Erdős","sourceUrl":"https://epoch.ai/benchmarks/frontiermath-erdos","sourceName":"Epoch AI","sourceHash":"7b27f45bae79ae4ac325b83ee436fbea5837d1b66be92f07527c1013db5d2337","description":"68 Lean conjectures. One attempt, $300 and 72 hours, Comparator-checked proof or disproof. Stored percent is the published proof_scorer figure; the CSV mean must agree within 0.00501. Epoch publishes no task version. Display only, separate from the weighted FrontierMath tier units. Data by Epoch AI, CC BY.","harness":"Epoch AI Inspect deepagent; one Lean attempt; $300 and 72h; Comparator proof check","unit":"percent"},{"id":"spireclimb","title":"SpireClimb","sourceUrl":"https://epoch.ai/benchmarks/spireclimb","sourceName":"Epoch AI","sourceHash":"f888097ed3ed930a9ab709f4e00687073dbf162343b5c7a6751045f78382bc77","description":"Slay the Spire 2 Ascension 10 is 75 runs. Slay the Spire 1 Ascension 20 Heart is 60 runs. Stored percent is the published win_rate/all figure for that setting. The CSV mean must agree within 0.00501. The equal-weight average and the composite SpireClimb row stay out. Display only. Excluded from weighted Overall. Data by Epoch AI, CC BY.","harness":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all","unit":"percent"},{"id":"swe-sweep","title":"SWE-sweep","sourceUrl":"https://swesweep.com/results/","sourceName":"SWE-sweep","sourceHash":"e6ba2a84b492a7f96405ecc132d35ba4c4b9cfc07ce7b2763719db79e8ff142c","description":"Whole-repository find-and-fix. 100 repositories and 4068 bugs. Score is the micro-average of bugs resolved, stored as a percent. Board updated 2026-10-08. Display only. 12 integer board rows. Astra 6 stays unscored because its published fraction is not an integer count of 4068 bugs. Fable 5.1 stays unmapped because its slug names max and the public label does not. Excluded from weighted Overall.","harness":"mini-SWE-agent on every published row. Shared scaffold, not a native product harness.","unit":"percent"},{"id":"agents-last-exam","title":"Agents' Last Exam","sourceUrl":"https://agents-last-exam.org/leaderboard","sourceName":"Agents' Last Exam","sourceHash":"20e5ddeed34fab2ccdfe4db0a24f55064323763871c52a88bebe9085ce0b4a5b","description":"Full/Overall pass rate on 152 tasks. One row per model and harness at the board's default effort, the best published pass rate among reported efforts. Display only. Not a weighted Overall input.","harness":"Per-row agent harness, named in each source label. The reference harness is Claude Code.","unit":"percent"},{"id":"lmarena-agent","title":"LMArena Agent Arena","sourceUrl":"https://arena.ai/leaderboard/agent","sourceName":"LMArena","sourceHash":"69295d5d1a85726489757988153331565b38dbaaa26e1e60dfb12fe5a2f18f2d","description":"Aggregate Agent Arena IPS net improvement, stored as the board's printed percentage points. Arena Agent Mode does not name a per-model scaffold. Display only. Excluded from weighted Overall.","harness":"Arena Agent Mode. The board does not name a per-model scaffold.","unit":"index"},{"id":"frontierswe-v2","title":"FrontierSWE v2","sourceUrl":"https://www.frontierswe.com/","sourceName":"Proximal","sourceHash":"a36f57ab0792d4214e2f55d4a30a41ae91568c33f93ef090807339e4949b7000","description":"Independent Proximal FrontierSWE v2 mean@5 on all 34 long-horizon coding tasks, nominal five trials per task and a 20-hour task budget. Actual task trial counts and any headline/task-summary differences remain in the source capture. Published headline scores retained. Independent verifiers. Methodology https://www.frontierswe.com/blog/v2. Collection date, not evaluation date. Display only. Excluded from weighted Overall.","harness":"Proximus. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.","unit":"percent"},{"id":"hss","title":"Humanity's Sixth Sense","sourceUrl":"https://labs.scale.com/leaderboard/hss","sourceName":"Scale","sourceHash":"4e84747aa47d930fb23a234254c60312026eeeca08c6f0e5e9eb8d4512462268","description":"Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall.","harness":"Scale AI + Elorian full-board evaluation; vendor-default temperature and top-p.","unit":"percent"}],"results":[{"id":"0ccbbf90d4c9c4326afce18d","modelId":"amazon-nova-2-lite","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":94.3333333333333,"warning":null},{"id":"e671611646e62c871981291c","modelId":"amazon-nova-2-lite","benchmarkId":"aa-aime-2025","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":46.6666666666667,"warning":null},{"id":"66cc84650427058f2772930d","modelId":"amazon-nova-2-lite","benchmarkId":"aa-aime-2025","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":88.6666666666667,"warning":null},{"id":"64756b2e556ca7f8a17d562d","modelId":"amazon-nova-2-lite","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":33.6666666666667,"warning":null},{"id":"03dfabe94473ec1383c87727","modelId":"amazon-nova-2-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":60.33333333333331,"warning":null},{"id":"b262e2499a8a04c4953ef1a8","modelId":"amazon-nova-2-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":54.3333333333333,"warning":null},{"id":"45da956013ea10df7c402504","modelId":"amazon-nova-2-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":60,"warning":null},{"id":"0751248f0bb7ed65861e7389","modelId":"amazon-nova-2-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":18.6666666666667,"warning":null},{"id":"47bd8869e46560faebc901e5","modelId":"amazon-nova-2-lite","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":0.28571428571428603,"warning":null},{"id":"d8ed88d0d23b97effed3af47","modelId":"amazon-nova-2-lite","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":0,"warning":null},{"id":"dccf044655a2e5816270d0b0","modelId":"amazon-nova-2-lite","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":0,"warning":null},{"id":"ee6d3dc313301dfa2d7580f1","modelId":"amazon-nova-2-lite","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":0,"warning":null},{"id":"f8d141d6b22ee6ed27856912","modelId":"amazon-nova-2-lite","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":366.46,"warning":null},{"id":"e0cbb3c39f8958e73c8b93c4","modelId":"amazon-nova-2-lite","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":78.17424242424241,"warning":null},{"id":"e6cbe2b537987c96f375520c","modelId":"amazon-nova-2-lite","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":81.1111111111111,"warning":null},{"id":"770e68112ea3ddae0c86295b","modelId":"amazon-nova-2-lite","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":69.7979797979798,"warning":null},{"id":"e36fe12f9d28aabea60a4cb8","modelId":"amazon-nova-2-lite","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":76.76767676767679,"warning":null},{"id":"341be8170b3f5df517a7b9f0","modelId":"amazon-nova-2-lite","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":60.303030303030305,"warning":null},{"id":"b603f7ce306c1e1e8f54c5fb","modelId":"amazon-nova-2-lite","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":11.631139944393,"warning":null},{"id":"d004f2656a21b9524e06f41e","modelId":"amazon-nova-2-lite","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":4.03151065801668,"warning":null},{"id":"9cafab22153f3adf2b760686","modelId":"amazon-nova-2-lite","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":8.98980537534754,"warning":null},{"id":"bb07aadd4166403d7629e25e","modelId":"amazon-nova-2-lite","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":2.87303058387396,"warning":null},{"id":"0b9be11de127b6a5dd38a758","modelId":"amazon-nova-2-lite","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":70.7482993197279,"warning":null},{"id":"af3ff7c1292c5e5f1a387250","modelId":"amazon-nova-2-lite","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":61.2244897959184,"warning":null},{"id":"96d13ee5ad91343cd323c116","modelId":"amazon-nova-2-lite","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":68.5034013605442,"warning":null},{"id":"49a7a5a737f5b970e9b48dc4","modelId":"amazon-nova-2-lite","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":40.5442176870748,"warning":null},{"id":"6c149c7a7286fecb5ccc6bcd","modelId":"amazon-nova-2-lite","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":71.1111111111111,"warning":null},{"id":"b58d55fa308463aa77c4d1ec","modelId":"amazon-nova-2-lite","benchmarkId":"aa-livecodebench","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":46.8783068783069,"warning":null},{"id":"e87e1f5c4d9b0f8147178b4e","modelId":"amazon-nova-2-lite","benchmarkId":"aa-livecodebench","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":66.3492063492063,"warning":null},{"id":"81d6342d15ab0e2f8e535ccd","modelId":"amazon-nova-2-lite","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":34.6031746031746,"warning":null},{"id":"2fcb2fb2385859105cfa5f69","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":81.8234707446808,"warning":null},{"id":"e2fde275d55c273e8760334d","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmlu-pro","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":78.7815824468085,"warning":null},{"id":"3c1f70d598ec189688264287","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmlu-pro","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":81.31648936170211,"warning":null},{"id":"3af2c688d4c527bc96bdfee4","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":74.3351063829787,"warning":null},{"id":"a5e24613d846e5bc4a3a5165","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":63.757225433526,"warning":null},{"id":"447ed76ffd85e1c68737a764","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":57.976878612716796,"warning":null},{"id":"c2dd08a9a597e44931beaddc","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":62.5433526011561,"warning":null},{"id":"971a463f2747f1886585e328","modelId":"amazon-nova-2-lite","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":49.0173410404624,"warning":null},{"id":"98da7bf72c754781eeedbe0f","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":-54.9833333333333,"warning":null},{"id":"af52f4d0de0e4af1d272cb6f","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":-54.1,"warning":null},{"id":"938024499b56b32b14c80f96","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":-56.9666666666667,"warning":null},{"id":"5b5db7acabdd817b3e7dd73e","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":-59.8166666666667,"warning":null},{"id":"7891c2dfbc96d674276156a4","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":19.03333333333333,"warning":null},{"id":"cec10762e7912f6f75ed3380","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":17.65,"warning":null},{"id":"9a368f392640f981cf550273","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":18.166666666666668,"warning":null},{"id":"3e84abbf7f996ec789cd2885","modelId":"amazon-nova-2-lite","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":14.016666666666666,"warning":null},{"id":"94a6a7a5352e148b739ba8d7","modelId":"amazon-nova-2-lite","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":72.8070175438596,"warning":null},{"id":"b7cc1ec4d45d91363e7e3394","modelId":"amazon-nova-2-lite","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":71.92982456140349,"warning":null},{"id":"c51971bddebaecbe524bf126","modelId":"amazon-nova-2-lite","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":75.7309941520468,"warning":null},{"id":"04fb036f055cd49b327a6ded","modelId":"amazon-nova-2-lite","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":61.9883040935672,"warning":null},{"id":"a6de7f73786737c3c3bd6da2","modelId":"amazon-nova-2-lite","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":16.6666666666667,"warning":null},{"id":"4b6c03512a0cc2c0532a836b","modelId":"amazon-nova-2-lite","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"Nova 2.0 Lite (Low)","configuration":"Nova 2.0 Lite (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-low","score":3.78787878787879,"warning":null},{"id":"987a4f6f97fbdd423c2df698","modelId":"amazon-nova-2-lite","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"Nova 2.0 Lite (Medium)","configuration":"Nova 2.0 Lite (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning-medium","score":17.4242424242424,"warning":null},{"id":"eb49445c621170b904099ffd","modelId":"amazon-nova-2-lite","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Nova 2.0 Lite (Non-reasoning)","configuration":"Nova 2.0 Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-lite","score":6.81818181818182,"warning":null},{"id":"cce657bb0f29d3c180b85e13","modelId":"amazon-nova-2-lite","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Nova 2.0 Lite (High)","configuration":"Nova 2.0 Lite (High); Model source: https://artificialanalysis.ai/models/nova-2-0-lite-reasoning","score":16.1048689138577,"warning":null},{"id":"5b5b9452eae1f755f1af5b09","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-aime-2025","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":63.3333333333333,"warning":null},{"id":"13346152511e09d30d1cfdc6","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-aime-2025","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":89,"warning":null},{"id":"6175ce4adc218cc2822fcbf9","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":30.666666666666696,"warning":null},{"id":"dfe9842c0735434bc21b3d96","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":64.6666666666667,"warning":null},{"id":"bd1063328a8fc5dc84164450","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":64,"warning":null},{"id":"27dd6af97bc82a3f33e9bbb4","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":30,"warning":null},{"id":"191638571a326d89f2d172a1","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":0,"warning":null},{"id":"2a2810cac886211d50260322","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":0,"warning":null},{"id":"0dde0536564f2839a2ce52d8","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":0,"warning":null},{"id":"61902e4111c8b5f16c58123d","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":436.45,"warning":null},{"id":"4e2a978cf1b25349ba8c9eb7","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":469.74,"warning":null},{"id":"5959fc580b7dd735b87780d8","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":331.66,"warning":null},{"id":"3527b43b12093b60f4c4417a","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-global-mmlu-lite","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":87.6,"warning":null},{"id":"7292759b8cb168b50cfc12a5","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":78.625,"warning":null},{"id":"7017becc8361a9c23a723ce0","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":75.0505050505051,"warning":null},{"id":"8232c24c070f4b3f1f77a27d","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":78.4848484848485,"warning":null},{"id":"38aa2e7ea58f399abf54de70","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":63.636363636363605,"warning":null},{"id":"79ecc009f3dbdbe3fb1dfe9d","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":5.18999073215941,"warning":null},{"id":"204bd1ee31406c92d27f6e68","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":9.40685820203892,"warning":null},{"id":"b14ffa0e6e2c2eaae9e72551","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":3.93883225208526,"warning":null},{"id":"902dc4fc6cc23bf4a62c8a17","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":79.5918367346939,"warning":null},{"id":"a9beabe09fa100c99468262b","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":79.0476190476191,"warning":null},{"id":"9fe32683ff453994e752cf0b","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":52.0408163265306,"warning":null},{"id":"cdcb51ed9f3f020082583698","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-livecodebench","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":63.8095238095238,"warning":null},{"id":"6429d609ec0f11dd8ffa179e","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-livecodebench","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":73.015873015873,"warning":null},{"id":"bd2de967d5e4ceefd8d21f5f","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":47.3015873015873,"warning":null},{"id":"32e569e98eea052f1f9fb1cc","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-mlcr-aa","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":3.88888888888889,"warning":null},{"id":"a02cbe3ba053252df487ca88","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-mmlu-pro","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":82.16422872340429,"warning":null},{"id":"4fced4b66edf60bfb2a4e60a","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-mmlu-pro","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":82.953789893617,"warning":null},{"id":"b491e7d678f7eb03aa97fe4f","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":77.1609042553192,"warning":null},{"id":"634fe80a6b864fb121b391cc","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":62.71676300578039,"warning":null},{"id":"da88117edeb51c2ac2876562","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":64.5086705202312,"warning":null},{"id":"8bcf6fb487bd4f7034945f07","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":-46.5333333333333,"warning":null},{"id":"9abe3df4334c51aad2aec0ad","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":-49.05,"warning":null},{"id":"83c3679cc2593a1416e8ece0","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":-49.5,"warning":null},{"id":"6713a85a37a3a382e22bdeaf","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":22,"warning":null},{"id":"1c0183c6de1a7b2084c29ccf","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":21.96666666666667,"warning":null},{"id":"e507780d68e9a88b3fe816c3","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":16.916666666666664,"warning":null},{"id":"9eb0dd38a4a360bfd08d573e","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":90.6432748538012,"warning":null},{"id":"b1e617855171778b680b1c86","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":92.6900584795322,"warning":null},{"id":"073593c114d374a6b5814324","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":71.6374269005848,"warning":null},{"id":"062bbb61636c7fa8ab91586f","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":17.4242424242424,"warning":null},{"id":"4c49b2990782fce5521733d9","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":24.2424242424242,"warning":null},{"id":"77e985c0ed8b2995e6b8a80d","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":16.6666666666667,"warning":null},{"id":"bf82530ea81bebcb3dd5a931","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Nova 2.0 Pro Preview (Low)","configuration":"Nova 2.0 Pro Preview (Low); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-low","score":19.4756554307116,"warning":null},{"id":"052225ad74be41cc4b6cb25f","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"Nova 2.0 Pro Preview (Medium)","configuration":"Nova 2.0 Pro Preview (Medium); Model source: https://artificialanalysis.ai/models/nova-2-0-pro-reasoning-medium","score":29.588014981273396,"warning":null},{"id":"f56f639149bec0565169758c","modelId":"amazon-nova-2-pro-preview","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Nova 2.0 Pro Preview (Non-reasoning)","configuration":"Nova 2.0 Pro Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/nova-2-0-pro","score":17.2284644194757,"warning":null},{"id":"50e4adca8e5e54dc5a129468","modelId":"amazon-nova-lite","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.000000000000001,"warning":null},{"id":"adf5df0b30d27f83e52f2b51","modelId":"amazon-nova-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":19,"warning":null},{"id":"ff486d9ba46adad27a1b6fe4","modelId":"amazon-nova-lite","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"dbd1d83f7580735392e6f6a7","modelId":"amazon-nova-lite","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.3333333333333,"warning":null},{"id":"981168cc7575baa0277b03a4","modelId":"amazon-nova-lite","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.29,"warning":null},{"id":"6feffecc4d281a47d9a8c63c","modelId":"amazon-nova-lite","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.149659863945594,"warning":null},{"id":"dbdd0586ea1c53079fda7843","modelId":"amazon-nova-lite","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.7195767195767,"warning":null},{"id":"7dd1a3d260f548765e07a7db","modelId":"amazon-nova-lite","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.46666666666671,"warning":null},{"id":"c8e86b20efeee800f08898bd","modelId":"amazon-nova-lite","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"c3a7675bf193929b296e1238","modelId":"amazon-nova-lite","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.959441489361694,"warning":null},{"id":"fd3e68f24b42e3314808e348","modelId":"amazon-nova-lite","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.8034682080925,"warning":null},{"id":"ed17e3be2f9a297198964c70","modelId":"amazon-nova-lite","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":-42.1666666666667,"warning":null},{"id":"c810f78cb4cfe1a60c148b93","modelId":"amazon-nova-lite","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.8,"warning":null},{"id":"02860b6478118aafcbe8cda2","modelId":"amazon-nova-lite","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.5438596491228,"warning":null},{"id":"69ddc6a403ed9b335c7c7e51","modelId":"amazon-nova-lite","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Nova Lite","configuration":"Nova Lite; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.757575757575758,"warning":null},{"id":"c395cbf5898af86d26300d73","modelId":"amazon-nova-micro","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":6,"warning":null},{"id":"cfcec14d811a655ca4726508","modelId":"amazon-nova-micro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":13,"warning":null},{"id":"db9b99c98dd617c9ca05910d","modelId":"amazon-nova-micro","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"696c084e4d6c62ab15b53492","modelId":"amazon-nova-micro","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.7575757575758,"warning":null},{"id":"f8a60f4c48d2cf35f26ad74c","modelId":"amazon-nova-micro","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.6339202965709,"warning":null},{"id":"7e8455203b39b886b2967732","modelId":"amazon-nova-micro","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.387755102040803,"warning":null},{"id":"aef838f6a1d2eca33b5a1e33","modelId":"amazon-nova-micro","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.968253968253999,"warning":null},{"id":"ffa04be81f3227c6f19350f1","modelId":"amazon-nova-micro","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.3333333333333,"warning":null},{"id":"35e37d3045a2a291273b84f1","modelId":"amazon-nova-micro","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.083444148936195,"warning":null},{"id":"03faabf6a483486fa063a636","modelId":"amazon-nova-micro","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":-48.5166666666667,"warning":null},{"id":"abebfd3cf4beb6117af462cc","modelId":"amazon-nova-micro","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.033333333333333,"warning":null},{"id":"88ec995dfbbdc4b5240ddad1","modelId":"amazon-nova-micro","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.0350877192982,"warning":null},{"id":"19d78b7b4a3613704624e4c9","modelId":"amazon-nova-micro","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Nova Micro","configuration":"Nova Micro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-micro; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.51515151515152,"warning":null},{"id":"ce9a2089796d737603a17e0b","modelId":"amazon-nova-premier","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.3333333333333,"warning":null},{"id":"f2f040f4479911816988b68b","modelId":"amazon-nova-premier","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.3333333333333,"warning":null},{"id":"56f117757019799897341310","modelId":"amazon-nova-premier","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"0e3b171515df1dcd1001b493","modelId":"amazon-nova-premier","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.64583333333334,"warning":null},{"id":"6a6eaae484ab8fcb709d1c0f","modelId":"amazon-nova-premier","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":56.8686868686869,"warning":null},{"id":"5f0a56bbcc02e28b010b7360","modelId":"amazon-nova-premier","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.17052826691381,"warning":null},{"id":"8be501c4015a8228e5bf59d1","modelId":"amazon-nova-premier","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.190476190476204,"warning":null},{"id":"63e379977a372098a782ec00","modelId":"amazon-nova-premier","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.7460317460317,"warning":null},{"id":"9c279ec3f04cf3fdc739c887","modelId":"amazon-nova-premier","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.93333333333331,"warning":null},{"id":"354ff394a1c479d16d1ff096","modelId":"amazon-nova-premier","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.3460771276596,"warning":null},{"id":"9ba3c2f0a8977ad288a8f0ec","modelId":"amazon-nova-premier","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":-35.3833333333333,"warning":null},{"id":"fe566bdc8b0fe8917d76032e","modelId":"amazon-nova-premier","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.25,"warning":null},{"id":"0bfbe8f57f24704b45f141ff","modelId":"amazon-nova-premier","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.3040935672515,"warning":null},{"id":"3abff7933c0a42c95c4f53c9","modelId":"amazon-nova-premier","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Nova Premier","configuration":"Nova Premier; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-premier; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.81818181818182,"warning":null},{"id":"541b9cd8656670591c6a1d18","modelId":"amazon-nova-pro","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.000000000000001,"warning":null},{"id":"9a068df3721adea27be27ecf","modelId":"amazon-nova-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":21,"warning":null},{"id":"d4896e593ff5512235c65933","modelId":"amazon-nova-pro","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"fca9c963afc237552313f230","modelId":"amazon-nova-pro","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.898989898989896,"warning":null},{"id":"3b7c848ef0c5b4072692c53a","modelId":"amazon-nova-pro","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.19,"warning":null},{"id":"1718757623befb2b04cdffa0","modelId":"amazon-nova-pro","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.0952380952381,"warning":null},{"id":"7a72aadc62751bdabe32c810","modelId":"amazon-nova-pro","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.2804232804233,"warning":null},{"id":"fc2fc962c57c7a95ec0f5ab6","modelId":"amazon-nova-pro","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.60000000000001,"warning":null},{"id":"2b21dffdfeaeb4feaa7217c7","modelId":"amazon-nova-pro","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.0990691489362,"warning":null},{"id":"99044be57784aba47c270823","modelId":"amazon-nova-pro","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.3352601156069,"warning":null},{"id":"ffd340540d54b4c2be0c4e29","modelId":"amazon-nova-pro","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":-47.7333333333333,"warning":null},{"id":"0050677da9a66bbd582ceef2","modelId":"amazon-nova-pro","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.866666666666667,"warning":null},{"id":"180bc2e6c945ab26a8631548","modelId":"amazon-nova-pro","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.0350877192982,"warning":null},{"id":"d8eefcfd98bc646739e26568","modelId":"amazon-nova-pro","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Nova Pro","configuration":"Nova Pro; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/nova-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.06060606060606,"warning":null},{"id":"9435766350d781bffea43ed6","modelId":"claude-fable-5","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":48.75,"warning":null},{"id":"76bcaf7e88c080b8ac04af34","modelId":"claude-fable-5","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":1539.25,"warning":null},{"id":"5ce598acfdcc6ca7457ec729","modelId":"claude-fable-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":82.3333333333333,"warning":null},{"id":"531a87c1984669bbce83e282","modelId":"claude-fable-5","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":54.071301140653816,"warning":null},{"id":"356477721c30ff51b6f282a7","modelId":"claude-fable-5","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":28.571428570000002,"warning":null},{"id":"d266a9f5ee053218775a4c59","modelId":"claude-fable-5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":51.11906893464637,"warning":null},{"id":"20d7159b0c6324c10ac262c7","modelId":"claude-fable-5","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":24,"warning":null},{"id":"1daae1ad1868c537e442fdc0","modelId":"claude-fable-5","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":1609.83,"warning":null},{"id":"86fa7444c37a3b98e534d304","modelId":"claude-fable-5","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":92.6262626262626,"warning":null},{"id":"4ed48500acd1cc582df42aca","modelId":"claude-fable-5","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":55.468025949953706,"warning":null},{"id":"c137a95a12dfae0d4072697c","modelId":"claude-fable-5","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":63.469387755102005,"warning":null},{"id":"a2756a1696ca10149dd776bd","modelId":"claude-fable-5","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":64.4444444444444,"warning":null},{"id":"47c17bcc7ee4f167db21f164","modelId":"claude-fable-5","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":43.3,"warning":null},{"id":"64e3aaa3b5101e5c4de66d15","modelId":"claude-fable-5","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":65.35,"warning":null},{"id":"75d2e4404944756c0cba8a37","modelId":"claude-fable-5","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":60.9953703703704,"warning":null},{"id":"1922a9b01edb48592b3c0b68","modelId":"claude-fable-5","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":98.5380116959064,"warning":null},{"id":"96bef260af3c6a14d7cf0897","modelId":"claude-fable-5","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":38.144329896907195,"warning":null},{"id":"eb937c63111f6acf387a7a14","modelId":"claude-fable-5","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":62.878787878787904,"warning":null},{"id":"3a5be26e225198df0b89d1e1","modelId":"claude-fable-5","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":84.6441947565543,"warning":null},{"id":"487463c73f1067c565b2bd7b","modelId":"claude-fable-5","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Fable 5 (Max, Opus 4.8 Fallback)","configuration":"Claude Fable 5 (Max, Opus 4.8 Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5","score":42.4242424242424,"warning":null},{"id":"28d5bc6a86675295d42cb6c0","modelId":"claude-fable-5","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Fable 5 (High)","configuration":"Claude Fable 5 (High); Source model ID: anthropic-claude-fable-5-high; 120 tasks","score":87.5,"warning":null},{"id":"6f9372a183ee68e6b48bb18c","modelId":"claude-fable-5","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Claude Fable 5 (Low)","configuration":"Claude Fable 5 (Low); Source model ID: anthropic-claude-fable-5-low; 120 tasks","score":76.80555555555554,"warning":null},{"id":"332952f68252290ee6a5a920","modelId":"claude-fable-5","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"Claude Fable 5 (Max)","configuration":"Claude Fable 5 (Max); Source model ID: anthropic-claude-fable-5-max; 120 tasks","score":89.16666666666667,"warning":null},{"id":"47f74dad473d960addbaf5cd","modelId":"claude-fable-5","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Claude Fable 5 (Medium)","configuration":"Claude Fable 5 (Medium); Source model ID: anthropic-claude-fable-5-medium; 120 tasks","score":82.5,"warning":null},{"id":"fd5f5a2e4f7c743ac93db432","modelId":"claude-fable-5","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"Claude Fable 5 (XHigh)","configuration":"Claude Fable 5 (XHigh); Source model ID: anthropic-claude-fable-5-xhigh; 120 tasks","score":88.33333333333333,"warning":null},{"id":"557dcb6d136782df1d6e3170","modelId":"claude-fable-5","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"claude-fable-5","configuration":"claude-fable-5; Reported effort: high; 113 unique tasks; 430 attempts; 4 runs","score":68.6,"warning":null},{"id":"59377a8e5a165f1e29a9f170","modelId":"claude-fable-5","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"claude-fable-5","configuration":"claude-fable-5; Reported effort: low; 113 unique tasks; 433 attempts; 4 runs","score":59.6,"warning":null},{"id":"3efb0703151365e0a5c57ad7","modelId":"claude-fable-5","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"claude-fable-5","configuration":"claude-fable-5; Reported effort: max; 113 unique tasks; 436 attempts; 4 runs","score":69.7,"warning":null},{"id":"9eade32ea719c5f0ba716c1d","modelId":"claude-fable-5","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"claude-fable-5","configuration":"claude-fable-5; Reported effort: medium; 113 unique tasks; 436 attempts; 4 runs","score":65.4,"warning":null},{"id":"2c326836891084fe24c0ed23","modelId":"claude-fable-5","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"claude-fable-5","configuration":"claude-fable-5; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":69.9,"warning":null},{"id":"c44bf1486b2f7dadbe84989a","modelId":"claude-fable-5","benchmarkId":"terminal-claudecode","level":"xhigh","sourceLabel":"Fable 5","configuration":"Fable 5; Reported effort: xhigh; 445 trials; recorded 2026-06-09","score":83.82,"warning":null},{"id":"667b208122296e70c88bf453","modelId":"claude-fable-5","benchmarkId":"terminal-terminus2","level":"high","sourceLabel":"Fable 5","configuration":"Fable 5; Reported effort: high; 445 trials; recorded 2026-06-09","score":80.45,"warning":null},{"id":"0d2241928ed5ecf21cc47ac8","modelId":"claude-fable-5-1","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":57.49999999999999,"warning":null},{"id":"79eb81a1926985df9fec881c","modelId":"claude-fable-5-1","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":1580.33,"warning":null},{"id":"16c61930c045c3ff73df0f76","modelId":"claude-fable-5-1","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":1481.82,"warning":null},{"id":"930522073edfaee595f3e2dc","modelId":"claude-fable-5-1","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":1672.26,"warning":null},{"id":"5564459db5e99d2ef9caab87","modelId":"claude-fable-5-1","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":1527.66,"warning":null},{"id":"e8f0298bfd7e6051d342b992","modelId":"claude-fable-5-1","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":1654.53,"warning":null},{"id":"e17b194d047e1fb07d6c7653","modelId":"claude-fable-5-1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":83.6666666666667,"warning":null},{"id":"209fa09996d72667664b84a1","modelId":"claude-fable-5-1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":82.3333333333333,"warning":null},{"id":"73886cb4960156696582a138","modelId":"claude-fable-5-1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":85.3333333333333,"warning":null},{"id":"66130fcc82d249a56cc72588","modelId":"claude-fable-5-1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":84.6666666666667,"warning":null},{"id":"4ed45207f55bfa9da7305754","modelId":"claude-fable-5-1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":83,"warning":null},{"id":"1b64b92573689bcaf597d7a4","modelId":"claude-fable-5-1","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":55.323097464639794,"warning":null},{"id":"ac9978895045dfb1f8f5a872","modelId":"claude-fable-5-1","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":52.2404059498186,"warning":null},{"id":"b7a6b6623b923e932e876ecc","modelId":"claude-fable-5-1","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":59.37591715646424,"warning":null},{"id":"b1ad55aeda82ff163c08fea5","modelId":"claude-fable-5-1","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":54.661034656728845,"warning":null},{"id":"cb8ccf39661f77db24de15c8","modelId":"claude-fable-5-1","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":57.77242508489599,"warning":null},{"id":"0c0d4e982b854745e5be73d6","modelId":"claude-fable-5-1","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":30.2857142857143,"warning":null},{"id":"3cc1c8cdc0fa9982f3a37060","modelId":"claude-fable-5-1","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":27.7142857142857,"warning":null},{"id":"1105d4c80e058bfc472b39e9","modelId":"claude-fable-5-1","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":29.714285714285698,"warning":null},{"id":"96c3e1daa9d8abb62ce6f2d0","modelId":"claude-fable-5-1","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":29.1428571428571,"warning":null},{"id":"5c39537b6d0a3b8567a30a33","modelId":"claude-fable-5-1","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":31.1428571428571,"warning":null},{"id":"9451f74c875d482256b056d2","modelId":"claude-fable-5-1","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":26.8,"warning":null},{"id":"cc78db35b4dd1022d87988d0","modelId":"claude-fable-5-1","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":28.000000000000004,"warning":null},{"id":"38c81d52c523e528ea529039","modelId":"claude-fable-5-1","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":26.200000000000003,"warning":null},{"id":"cc54b16b003b7e3f08a55612","modelId":"claude-fable-5-1","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":26.8,"warning":null},{"id":"71ccbbe1aa4031f1e0971630","modelId":"claude-fable-5-1","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":26.200000000000003,"warning":null},{"id":"9e3b37c8803a1c740544759a","modelId":"claude-fable-5-1","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":1634.67,"warning":null},{"id":"8bc44708303e26e32456d0da","modelId":"claude-fable-5-1","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":1468.77,"warning":null},{"id":"6bc006318c45f239804088dd","modelId":"claude-fable-5-1","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":1754.07,"warning":null},{"id":"c3ed0a2daacec6206bbaa8a0","modelId":"claude-fable-5-1","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":1545.43,"warning":null},{"id":"fbd8c9e79011ea244b27f50d","modelId":"claude-fable-5-1","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":1732.14,"warning":null},{"id":"a1a11f141688016a46e226b4","modelId":"claude-fable-5-1","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":90.6060606060606,"warning":null},{"id":"d230e7a8ed11054c123bb0d4","modelId":"claude-fable-5-1","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":88.0808080808081,"warning":null},{"id":"f0cb666d8b397509d01fcda7","modelId":"claude-fable-5-1","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":93.7373737373737,"warning":null},{"id":"5cb88ee62098016c06f6762c","modelId":"claude-fable-5-1","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":88.5858585858586,"warning":null},{"id":"792cbba759e6a457e3ce95d9","modelId":"claude-fable-5-1","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":93.4343434343434,"warning":null},{"id":"01e6485687b6c9b6d3ee29e3","modelId":"claude-fable-5-1","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":6.388888888888888,"warning":null},{"id":"fc7797b51089a31d390d12de","modelId":"claude-fable-5-1","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":55.9314179796108,"warning":null},{"id":"9b504eb1f3e662889517197b","modelId":"claude-fable-5-1","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":48.887859128823,"warning":null},{"id":"91ca5a285a815cc577861a90","modelId":"claude-fable-5-1","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":59.1288229842447,"warning":null},{"id":"26f130cc5f3afec42cfc4dd9","modelId":"claude-fable-5-1","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":53.79981464318811,"warning":null},{"id":"fb6fc67b57f0322e404e6415","modelId":"claude-fable-5-1","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":58.7117701575533,"warning":null},{"id":"f5ad1ec69a32540d6cf88829","modelId":"claude-fable-5-1","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":49.5291902071563,"warning":null},{"id":"5aa58ff856aa3368e5fc5250","modelId":"claude-fable-5-1","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":71.1111111111111,"warning":null},{"id":"aa0c7951aafc82ce0085649b","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":40.8,"warning":null},{"id":"0057ecc817d74d83239c4424","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":34.1333333333333,"warning":null},{"id":"f92c8f985db06aef5dfde1a6","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":43.45,"warning":null},{"id":"eb186484745af7365d75fb3b","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":37.6,"warning":null},{"id":"85f939b76eb0784c35e2e46b","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":42.3833333333333,"warning":null},{"id":"fc49aad1b1380e34338be308","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":64.93333333333334,"warning":null},{"id":"3f29825fbd260545d9e83952","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":60.23333333333334,"warning":null},{"id":"a4cae71c999c47df4f9393e6","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":67.23333333333333,"warning":null},{"id":"119a7c91a6eb069c611815ec","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":63.1,"warning":null},{"id":"102d33bbaf7adbb8853357a4","modelId":"claude-fable-5-1","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":66.21666666666667,"warning":null},{"id":"fe65b606410257a83ae591b0","modelId":"claude-fable-5-1","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":58.6805555555556,"warning":null},{"id":"915c42fed9c168b0cd0d4ce3","modelId":"claude-fable-5-1","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":56.712962962963,"warning":null},{"id":"24c940fd705993c6af78cc1f","modelId":"claude-fable-5-1","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":63.078703703703695,"warning":null},{"id":"63edf23165ae63ad675dd53b","modelId":"claude-fable-5-1","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":56.3657407407407,"warning":null},{"id":"b65d0a7245c9b5e0c5392552","modelId":"claude-fable-5-1","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":60.8796296296296,"warning":null},{"id":"3cedda1b84b7744fa92fd1a2","modelId":"claude-fable-5-1","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":43.0927835051546,"warning":null},{"id":"0b6f6d0f61173573dafa22b9","modelId":"claude-fable-5-1","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":38.9690721649485,"warning":null},{"id":"3251ff6e17e7d3a64776d445","modelId":"claude-fable-5-1","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":47.2164948453608,"warning":null},{"id":"f5ce17b2e3a7060bbaf0cadf","modelId":"claude-fable-5-1","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":41.0309278350515,"warning":null},{"id":"c5defbb2f0795fe80153019a","modelId":"claude-fable-5-1","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":45.7731958762887,"warning":null},{"id":"8d3ac42947ff081d75c9bfb5","modelId":"claude-fable-5-1","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":43.3333333333333,"warning":null},{"id":"e16e790adb38cab0330ff7c0","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":89.8876404494382,"warning":null},{"id":"4b54956296ee8159beebf70d","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":85.0187265917603,"warning":null},{"id":"2eec269bfe424fa9f99685ab","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":91.3857677902622,"warning":null},{"id":"3c549a2e779d3d457109065a","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":88.0149812734082,"warning":null},{"id":"a6fc2443743364aa88cce1af","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":91.01123595505621,"warning":null},{"id":"e5e803fa2a589225aa8c8419","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Claude Fable 5.1 (High, Default Fallback)","configuration":"Claude Fable 5.1 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-high","score":52.020202020202,"warning":null},{"id":"2632e3793bf5971caf4889c9","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Claude Fable 5.1 (Low, Default Fallback)","configuration":"Claude Fable 5.1 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-low","score":40.4040404040404,"warning":null},{"id":"ac87b1460549fabf982f7790","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Claude Fable 5.1 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1","score":52.020202020202,"warning":null},{"id":"6b835294bb69e85cf6d63b7c","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium, Default Fallback)","configuration":"Claude Fable 5.1 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-medium","score":44.949494949495,"warning":null},{"id":"6dbeca9772cc299a553e7919","modelId":"claude-fable-5-1","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Claude Fable 5.1 (Xhigh, Default Fallback)","configuration":"Claude Fable 5.1 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-fable-5-1-xhigh","score":55.050505050505,"warning":null},{"id":"ed1a32448622df1b6e6a2c76","modelId":"claude-fable-5-1","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Fable 5.1 (High)","configuration":"Claude Fable 5.1 (High); Source model ID: anthropic-claude-fable-5-1-high; 120 tasks","score":88.75,"warning":null},{"id":"e32cbb2b2e540c705e2682f4","modelId":"claude-fable-5-1","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Claude Fable 5.1 (Low)","configuration":"Claude Fable 5.1 (Low); Source model ID: anthropic-claude-fable-5-1-low; 120 tasks","score":78.33333333333333,"warning":null},{"id":"29140dd5416b42ea788cab53","modelId":"claude-fable-5-1","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"Claude Fable 5.1 (Max)","configuration":"Claude Fable 5.1 (Max); Source model ID: anthropic-claude-fable-5-1-max; 120 tasks","score":90,"warning":null},{"id":"07f22f183aad2883655252ba","modelId":"claude-fable-5-1","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Claude Fable 5.1 (Medium)","configuration":"Claude Fable 5.1 (Medium); Source model ID: anthropic-claude-fable-5-1-medium; 120 tasks","score":86.25,"warning":null},{"id":"01e39ca976d82e99ab3b9be8","modelId":"claude-fable-5-1","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"Claude Fable 5.1 (XHigh)","configuration":"Claude Fable 5.1 (XHigh); Source model ID: anthropic-claude-fable-5-1-xhigh; 120 tasks","score":90,"warning":null},{"id":"12613d0c4a993819d4dbdce2","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":15,"warning":null},{"id":"6c8d28d7b244e06c11c2edc9","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":616.35,"warning":null},{"id":"3c47d4da68b5bf084c0e9f76","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":39,"warning":null},{"id":"d6f6b975bd6c103d42d44df7","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":83.6666666666667,"warning":null},{"id":"2705174dc70afd80971d2f06","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":49.6666666666667,"warning":null},{"id":"468ce6503360afbf169e36d2","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":74.3333333333333,"warning":null},{"id":"bea6cc2230d015678cd7aa2a","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":3.248631462547438,"warning":null},{"id":"5791f8e4e8fde45e7fb66886","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":0,"warning":null},{"id":"b99ba0207b1b453a50cc06f6","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":0,"warning":null},{"id":"db86f964ce75806a4ed189c1","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":29.3942106833781,"warning":null},{"id":"2806b54519e581523aedfb18","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":3.8,"warning":null},{"id":"6ee054e5b60e917b043fe6e8","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":732.34,"warning":null},{"id":"d65a5ce38f7183b304600ef1","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":83.40625,"warning":null},{"id":"4d49a8d283248db5fd5e0e34","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":64.6464646464646,"warning":null},{"id":"509191a186cf539ad56147b5","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":67.1717171717172,"warning":null},{"id":"ffb8c6eaab9ce5c0d4156a57","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":4.21686746987952,"warning":null},{"id":"f4ba8ec0f816cf38c66148d7","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":10.3799814643188,"warning":null},{"id":"cf094d38468fec8c0e27adb0","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":42.0408163265306,"warning":null},{"id":"4261bc632ec7f2d41fba238e","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":54.285714285714306,"warning":null},{"id":"b93459d07cc3fd00f317e844","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":27.3069679849341,"warning":null},{"id":"77e2abafe21efe47759564a5","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":51.11111111111109,"warning":null},{"id":"ad5e1bd45ba3b1854e2f521b","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":61.4814814814815,"warning":null},{"id":"6f10d5c0ef1a592623e7cd20","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":4.444444444444439,"warning":null},{"id":"0905c2c3dc408b46b86803cb","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":6.11111111111111,"warning":null},{"id":"fc24d97f80cf49b75640cc90","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":80.01163563829789,"warning":null},{"id":"acf88c56fae018e5c7c5ee48","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":75.9890292553192,"warning":null},{"id":"23ecbd3ca8a4df7f15ba1758","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":55.144508670520196,"warning":null},{"id":"aacbc7330857d739a0430710","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":58.55491329479769,"warning":null},{"id":"6cbf4db0f85b1a9026199f02","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":-7.56666666666667,"warning":null},{"id":"2da4c6c978b2c7fcd202e37f","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":-4.36666666666667,"warning":null},{"id":"e9c6fcae19f0b875efc5d431","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":14.433333333333334,"warning":null},{"id":"8862c4640b472096ac1a95b1","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":18.016666666666666,"warning":null},{"id":"a4ab94d87626afc466c301f6","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":42.2453703703704,"warning":null},{"id":"cb4ff6598c0f6e58acbeb5e6","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":32.4561403508772,"warning":null},{"id":"5067c956bfdd13d3de44ff4f","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":54.678362573099406,"warning":null},{"id":"a14c2426307d122868a12c50","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":9.27835051546392,"warning":null},{"id":"d76d253d766085ae9b980404","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Claude 4.5 Haiku (Non-reasoning)","configuration":"Claude 4.5 Haiku (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku","score":27.2727272727273,"warning":null},{"id":"d18cf2172a3d45ea811b2610","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":27.2727272727273,"warning":null},{"id":"2c85fcef255e2897eb9b9056","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":44.194756554307105,"warning":null},{"id":"64b4f035f94e078ac148dbd9","modelId":"claude-haiku-4-5-20251001","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Claude 4.5 Haiku (Reasoning)","configuration":"Claude 4.5 Haiku (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-haiku-reasoning","score":0,"warning":null},{"id":"95e92e71b76ea62d02ca6cd4","modelId":"claude-haiku-4-5-20251001","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Haiku 4.5 (Thinking 32K)","configuration":"Claude Haiku 4.5 (Thinking 32K); Source model ID: claude-haiku-4-5-20251001-thinking-32k; 120 tasks","score":4.03,"warning":null},{"id":"ba3f20c5292fdd48682b2099","modelId":"claude-haiku-4-5-20251001","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Haiku 4.5 (Thinking 1K)","configuration":"Claude Haiku 4.5 (Thinking 1K); Source model ID: claude-haiku-4-5-20251001-thinking-1k; 120 tasks","score":1.25,"warning":null},{"id":"e2c60a47ba43609d5f7802f0","modelId":"claude-haiku-4-5-20251001","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Haiku 4.5 (Thinking 8K)","configuration":"Claude Haiku 4.5 (Thinking 8K); Source model ID: claude-haiku-4-5-20251001-thinking-8k; 120 tasks","score":1.67,"warning":null},{"id":"f4fbbe2e4e16767233d36d08","modelId":"claude-haiku-4-5-20251001","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Haiku 4.5 (Thinking 16K)","configuration":"Claude Haiku 4.5 (Thinking 16K); Source model ID: claude-haiku-4-5-20251001-thinking-16k; 120 tasks","score":2.78,"warning":null},{"id":"1ea2fedae0dfb86a5dd01bf1","modelId":"claude-haiku-5-5","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":53.75,"warning":null},{"id":"901530bc97d64833a4184bd6","modelId":"claude-haiku-5-5","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":1442.67,"warning":null},{"id":"e0a7906551f3c65040e20a8a","modelId":"claude-haiku-5-5","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":1111.68,"warning":null},{"id":"a49c9d96fc79463708ffe2b1","modelId":"claude-haiku-5-5","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":1576.85,"warning":null},{"id":"a77bcd02e14f4b65c326decd","modelId":"claude-haiku-5-5","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":1371.54,"warning":null},{"id":"bde02ab730bb204920c8be9e","modelId":"claude-haiku-5-5","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":1531.52,"warning":null},{"id":"c41a56974171e44af0708a92","modelId":"claude-haiku-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":77.3333333333333,"warning":null},{"id":"da7e71d81005998e0f0d9888","modelId":"claude-haiku-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":72,"warning":null},{"id":"412901d2797666a44c38da72","modelId":"claude-haiku-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":82.6666666666667,"warning":null},{"id":"25dd297e17e51fbde9fc87dc","modelId":"claude-haiku-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":77.3333333333333,"warning":null},{"id":"59b18fce3369e2928eda1801","modelId":"claude-haiku-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":78.3333333333333,"warning":null},{"id":"659011efc6ea7b5200f9236b","modelId":"claude-haiku-5-5","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":33.72685311767238,"warning":null},{"id":"f1bf9aab5e911aba6fed0673","modelId":"claude-haiku-5-5","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":22.910812181110643,"warning":null},{"id":"5e542e0e15f089d2f9c37083","modelId":"claude-haiku-5-5","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":35.411015132783625,"warning":null},{"id":"efa73122b886c0b4ff588300","modelId":"claude-haiku-5-5","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":28.589614814293164,"warning":null},{"id":"31842dcb7511112fa3607683","modelId":"claude-haiku-5-5","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":35.953889773428195,"warning":null},{"id":"2c356cc8923a93b086247731","modelId":"claude-haiku-5-5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":18.5714285714286,"warning":null},{"id":"c93dd7b9ab2719287c413b24","modelId":"claude-haiku-5-5","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":9.14285714285714,"warning":null},{"id":"6e79b9a027c9f3c3994fd9e9","modelId":"claude-haiku-5-5","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":18.8571428571429,"warning":null},{"id":"f5e2600d3035ec6ff9d2b094","modelId":"claude-haiku-5-5","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":12.8571428571429,"warning":null},{"id":"e20e2ac5f5d97cca8665c048","modelId":"claude-haiku-5-5","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":22.5714285714286,"warning":null},{"id":"a6eb5a937b34b9f9749bf8fd","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":17.2,"warning":null},{"id":"da48da7da16887ff479364e1","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":11.200000000000001,"warning":null},{"id":"fb37eacc1e7bf7618408c670","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":20.8,"warning":null},{"id":"882433e9da8a86564b293325","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":15.2,"warning":null},{"id":"6e49d4b3845b15d79f8d9e7e","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":18.2,"warning":null},{"id":"6547f03f823fd70d8f7cb543","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":1417.55,"warning":null},{"id":"25ad4d77ec1f4579d7332cd1","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":1122.82,"warning":null},{"id":"f0367f1d09be031254666dd5","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":1618.33,"warning":null},{"id":"b60fc3de8a8b9625574914fb","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":1274.5,"warning":null},{"id":"556e07a16bd29ea79d0012fe","modelId":"claude-haiku-5-5","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":1513.05,"warning":null},{"id":"67dcfde23ccdbe31c74ae024","modelId":"claude-haiku-5-5","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":1.111111111111111,"warning":null},{"id":"45f41b968fcc40d7544f102b","modelId":"claude-haiku-5-5","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":1.6666666666666667,"warning":null},{"id":"d29655f1965ac44628d201ae","modelId":"claude-haiku-5-5","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":0.5555555555555555,"warning":null},{"id":"d23124fcf5b40c21b8714c1a","modelId":"claude-haiku-5-5","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":0.5555555555555555,"warning":null},{"id":"2fc3ee36bfe14f597d841cf7","modelId":"claude-haiku-5-5","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":2.7777777777777777,"warning":null},{"id":"e6b08b55be1cad404f7f38ce","modelId":"claude-haiku-5-5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":37.25671918443,"warning":null},{"id":"d129f54bd0882ce474c63b9c","modelId":"claude-haiku-5-5","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":27.0157553290083,"warning":null},{"id":"d659b5095a8ac992f3ef380d","modelId":"claude-haiku-5-5","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":44.3929564411492,"warning":null},{"id":"043fcd45aac117c1034cc8db","modelId":"claude-haiku-5-5","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":33.7812789620019,"warning":null},{"id":"eae5891239f4bd91881b95de","modelId":"claude-haiku-5-5","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":42.678405931418,"warning":null},{"id":"3bd26295cb43d510d7721106","modelId":"claude-haiku-5-5","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":63.3333333333333,"warning":null},{"id":"46e00a4ac847e9e09244d1db","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":5.75,"warning":null},{"id":"0843e6c11c75623edcb16a7c","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":3.25,"warning":null},{"id":"78eec329fdf447cf475f1472","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":10.6666666666667,"warning":null},{"id":"794322d853e75058b2eaaf9d","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":4.46666666666667,"warning":null},{"id":"1c807b583a76289a7f8e81fb","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":6.05,"warning":null},{"id":"ba41841c38e2b596cd43a092","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":34.78333333333333,"warning":null},{"id":"eb00fc84756629c7c66fcd8f","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":33.25,"warning":null},{"id":"ac50dccb8aad18dfce416986","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":36.38333333333333,"warning":null},{"id":"a0af2ce4f38ef9ff7a7e13d9","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":34.03333333333333,"warning":null},{"id":"e166da14c0ff17e226137155","modelId":"claude-haiku-5-5","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":34.949999999999996,"warning":null},{"id":"a149d9aace434b3cd553342b","modelId":"claude-haiku-5-5","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":48.7268518518519,"warning":null},{"id":"594b2c9a79e1cdafab55d368","modelId":"claude-haiku-5-5","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":49.189814814814795,"warning":null},{"id":"92825550c99ad8e6ecf875fd","modelId":"claude-haiku-5-5","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":54.976851851851805,"warning":null},{"id":"43c470bb96d270e8d9f2a82d","modelId":"claude-haiku-5-5","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":48.9583333333333,"warning":null},{"id":"84edbf4d0f88ca3f6ee43e8f","modelId":"claude-haiku-5-5","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":51.73611111111111,"warning":null},{"id":"7a65883bef43d1c98fd6af9d","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":10,"warning":null},{"id":"2e10c68989141ec31666fabd","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminal-bench-science","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":1.9047619047619,"warning":null},{"id":"8843be38ce6e313312e080ea","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":20,"warning":null},{"id":"40c52aa61a7d5de328a0f325","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminal-bench-science","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":7.14285714285714,"warning":null},{"id":"edcfeba2fb38d06a95739463","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminal-bench-science","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":18.095238095238102,"warning":null},{"id":"e153ed6842cf4cf2a6147283","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-high","score":21.7171717171717,"warning":null},{"id":"8c5cee954feccff6c1e3655c","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-low","score":12.6262626262626,"warning":null},{"id":"745618b55e15dcede6ea3cae","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Haiku 5.5 (Max)","configuration":"Claude Haiku 5.5 (Max); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5","score":32.8282828282828,"warning":null},{"id":"43c0aad399d4e0676ae239ea","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-medium","score":15.151515151515198,"warning":null},{"id":"6f41874611955ee4be4baff3","modelId":"claude-haiku-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (Xhigh)","configuration":"Claude Haiku 5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-haiku-5-5-xhigh","score":29.292929292929298,"warning":null},{"id":"47ea478bce608ff06705a26a","modelId":"claude-haiku-5-5","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Haiku 5.5 (High)","configuration":"Claude Haiku 5.5 (High); Source model ID: anthropic-claude-haiku-5-5-high; 120 tasks","score":30.69444444444444,"warning":null},{"id":"31f12232eb5252c6a7f039db","modelId":"claude-haiku-5-5","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Claude Haiku 5.5 (Low)","configuration":"Claude Haiku 5.5 (Low); Source model ID: anthropic-claude-haiku-5-5-low; 120 tasks","score":2.7777777777777777,"warning":null},{"id":"6ba48bd49a77dbfae49f5c9e","modelId":"claude-haiku-5-5","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Claude Haiku 5.5 (Medium)","configuration":"Claude Haiku 5.5 (Medium); Source model ID: anthropic-claude-haiku-5-5-medium; 120 tasks","score":9.444444444444443,"warning":null},{"id":"625f733f63d241339388028a","modelId":"claude-haiku-5-5","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (XHigh)","configuration":"Claude Haiku 5.5 (XHigh); Source model ID: anthropic-claude-haiku-5-5-xhigh; 120 tasks","score":61.25000000000001,"warning":null},{"id":"777d836c2aab8b1a753b4fe5","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":62.66666666666671,"warning":null},{"id":"0f517ae1f51f86c35cfb0be4","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":91.3333333333333,"warning":null},{"id":"364e593f02f62ed75fd18066","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":70.6666666666667,"warning":null},{"id":"85c81b940cce2a4e217f39f1","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":77.3333333333333,"warning":null},{"id":"108fba49cb2e133b5c3292c6","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":0.28571428571428603,"warning":null},{"id":"1958b99c15a27c0d8d0aba84","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":4.57142857142857,"warning":null},{"id":"a1db7ea0ee512320711194d0","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":90.58854166666666,"warning":null},{"id":"a0e35583919bfa925e3253c4","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":91.27604166666666,"warning":null},{"id":"1db3648c4fe84e124fa696fc","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":81.010101010101,"warning":null},{"id":"ea5097c6ae66554091a20a26","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":86.5656565656566,"warning":null},{"id":"be7e6806ef9ffabbe47660e5","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":13.1603336422614,"warning":null},{"id":"c8afa47b2b589899209075fd","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":30.120481927710802,"warning":null},{"id":"555c0f000e8ee371e88e57a2","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":42.9931972789116,"warning":null},{"id":"321deea5aaf37c3322817314","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":57.959183673469404,"warning":null},{"id":"fa884eeb86cb180cbd489bac","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":73.7566137566138,"warning":null},{"id":"3b0e8231a510903a76fb9715","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":87.08994708994709,"warning":null},{"id":"5314fc9ebb9ae8bcdf75b98e","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":88.921210106383,"warning":null},{"id":"6d150283271e045da32ceab8","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":89.453125,"warning":null},{"id":"34e7fcfb97c55289d0bf4b48","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":71.2138728323699,"warning":null},{"id":"46aff7e0e196660d8e321636","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":74.0462427745665,"warning":null},{"id":"299cfb2488b1b12fb8bae121","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":-4.13333333333333,"warning":null},{"id":"06bc173078345ce534aba7f3","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":14,"warning":null},{"id":"313fe229263c5d3be7780b9b","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":40.91666666666667,"warning":null},{"id":"8667c79135a6fbe7cc5fa1e4","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":46.58333333333333,"warning":null},{"id":"993af034623aaedde3a3de9b","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":86.2573099415205,"warning":null},{"id":"08adca9aa4e2043174db9373","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":89.4736842105263,"warning":null},{"id":"42a28e056d264ff2cdaead6f","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Claude Opus 4.5 (Non-reasoning)","configuration":"Claude Opus 4.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5","score":40.9090909090909,"warning":null},{"id":"87149dbfd56c9450a6af41d4","modelId":"claude-opus-4-5-20251101","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Claude Opus 4.5 (Reasoning)","configuration":"Claude Opus 4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/claude-opus-4-5-thinking","score":46.969696969697,"warning":null},{"id":"0ced355c18e4b0a2ef2a9eed","modelId":"claude-opus-4-6","benchmarkId":"aa-apex-agents-aa","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":33.03834808259587,"warning":null},{"id":"cb0ac21433b7ef09d3b34104","modelId":"claude-opus-4-6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":78,"warning":null},{"id":"7082c066347d04c1fe0eff71","modelId":"claude-opus-4-6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":67,"warning":null},{"id":"e7af7bb6a33272e67e662c05","modelId":"claude-opus-4-6","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":12.5714285714286,"warning":null},{"id":"77ac14caa7dab95c25617b11","modelId":"claude-opus-4-6","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":2.84081632653061,"warning":null},{"id":"faf1e40e69e6378501bad7ab","modelId":"claude-opus-4-6","benchmarkId":"aa-global-mmlu-lite","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":92.15104166666667,"warning":null},{"id":"fe0515e03e065f0696032470","modelId":"claude-opus-4-6","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":90.64062500000001,"warning":null},{"id":"9c48abd813ff3ce9a58df725","modelId":"claude-opus-4-6","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":89.5959595959596,"warning":null},{"id":"8be4cd366920585446a61182","modelId":"claude-opus-4-6","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":84.040404040404,"warning":null},{"id":"bdd2a2b5454c337d0d08008e","modelId":"claude-opus-4-6","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":39.9443929564411,"warning":null},{"id":"4069d360565b333c9ffd8301","modelId":"claude-opus-4-6","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":19.0917516218721,"warning":null},{"id":"eaf8f10f1fb7f2401ceb6549","modelId":"claude-opus-4-6","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":53.1292517006803,"warning":null},{"id":"878ba5ccfafc3a3aac69124f","modelId":"claude-opus-4-6","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":44.6258503401361,"warning":null},{"id":"e3fce5f1e1e8d14e10809865","modelId":"claude-opus-4-6","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":75.4335260115607,"warning":null},{"id":"6a3c74c1e096ae95afdbfee4","modelId":"claude-opus-4-6","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":72.5433526011561,"warning":null},{"id":"1a525508a4fe8c81f1a4d4f0","modelId":"claude-opus-4-6","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":13.6666666666667,"warning":null},{"id":"50535977bbb1720c8107ec69","modelId":"claude-opus-4-6","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":2.36666666666667,"warning":null},{"id":"27bba211267d5570417f622c","modelId":"claude-opus-4-6","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":46.983333333333334,"warning":null},{"id":"64ac033ef434490438e5ae15","modelId":"claude-opus-4-6","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":45.78333333333333,"warning":null},{"id":"8962563dcc6f8c32537e7cf9","modelId":"claude-opus-4-6","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":92.1052631578947,"warning":null},{"id":"f5abc0c837684c900a63f681","modelId":"claude-opus-4-6","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":84.7953216374269,"warning":null},{"id":"ed160c208fe0d6fd577443a2","modelId":"claude-opus-4-6","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"Claude Opus 4.6 (Max)","configuration":"Claude Opus 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-6-adaptive","score":46.2121212121212,"warning":null},{"id":"9889f6fc9b3b06d00d1dd02a","modelId":"claude-opus-4-6","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Claude Opus 4.6 (Non-reasoning, High)","configuration":"Claude Opus 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-6","score":48.4848484848485,"warning":null},{"id":"3da84ad297936aa3afed7671","modelId":"claude-opus-4-6","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Opus 4.6 (120K, High)","configuration":"Claude Opus 4.6 (120K, High); Source model ID: claude-opus-4-6-thinking-120K-high; 120 tasks","score":69.17,"warning":null},{"id":"cc808875c949319a9c28aeb4","modelId":"claude-opus-4-6","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Claude Opus 4.6 (120K, Low)","configuration":"Claude Opus 4.6 (120K, Low); Source model ID: claude-opus-4-6-thinking-120K-low; 120 tasks","score":64.58,"warning":null},{"id":"d89c7fdc274f08893e4629d1","modelId":"claude-opus-4-6","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"Claude Opus 4.6 (120K, Max)","configuration":"Claude Opus 4.6 (120K, Max); Source model ID: claude-opus-4-6-thinking-120K-max; 120 tasks","score":68.75,"warning":null},{"id":"06f217cc8cc75c7a1ed581a5","modelId":"claude-opus-4-6","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Claude Opus 4.6 (120K, Medium)","configuration":"Claude Opus 4.6 (120K, Medium); Source model ID: claude-opus-4-6-thinking-120K-medium; 120 tasks","score":66.25,"warning":null},{"id":"f95de089dfd26220d76b06bd","modelId":"claude-opus-4-6","benchmarkId":"swepro-minisweagentstarredrows","level":"thinking","sourceLabel":"claude-opus-4-6 (thinking)*","configuration":"claude-opus-4-6 (thinking)*; 731 tasks; source date 2026-04-08","score":51.9,"warning":null},{"id":"230d3d8b83042b632453157b","modelId":"claude-opus-4-7","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":43.75,"warning":null},{"id":"6142ab1e934d7ff75a871540","modelId":"claude-opus-4-7","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":1254.06,"warning":null},{"id":"804c0eb91ca7984918edf037","modelId":"claude-opus-4-7","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":1067.85,"warning":null},{"id":"45725090e3ebb260f56a646d","modelId":"claude-opus-4-7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":78.6666666666667,"warning":null},{"id":"fe3bce7d99bdaec121ad81a0","modelId":"claude-opus-4-7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":75.6666666666667,"warning":null},{"id":"18ee4e38399f8dc8f62db6af","modelId":"claude-opus-4-7","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":12,"warning":null},{"id":"f56ecb77460063c3aaa9e902","modelId":"claude-opus-4-7","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":5.14285714285714,"warning":null},{"id":"a615ab831a3f6c984efe2a74","modelId":"claude-opus-4-7","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":1352.39,"warning":null},{"id":"d9ae69d15a29c2e348242045","modelId":"claude-opus-4-7","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":91.4141414141414,"warning":null},{"id":"ec4a3a88869c76d0c90cf0da","modelId":"claude-opus-4-7","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":88.4848484848485,"warning":null},{"id":"19efc850d9a3b9e724fc29ec","modelId":"claude-opus-4-7","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":42.3076923076923,"warning":null},{"id":"77e7982cf86fa643461872bc","modelId":"claude-opus-4-7","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":33.3178869323448,"warning":null},{"id":"98e0842c05983f6201644100","modelId":"claude-opus-4-7","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":58.6394557823129,"warning":null},{"id":"56a0744f016cf615838f1aee","modelId":"claude-opus-4-7","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":43.6054421768707,"warning":null},{"id":"36a79ec29a0e790aea0a9691","modelId":"claude-opus-4-7","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":46.6572504708098,"warning":null},{"id":"7c3f2dfeb685623b873b99c4","modelId":"claude-opus-4-7","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":28.888888888888896,"warning":null},{"id":"f66c2dd60241efe31cd7779a","modelId":"claude-opus-4-7","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":78.8439306358382,"warning":null},{"id":"35884c5a3a5d3a0905614cfa","modelId":"claude-opus-4-7","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":76.35838150289021,"warning":null},{"id":"0ac072efffbc910c90cc5e43","modelId":"claude-opus-4-7","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":27.2666666666667,"warning":null},{"id":"dd3f3302d99c6e2a87da9709","modelId":"claude-opus-4-7","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":14.8333333333333,"warning":null},{"id":"9a5d2158fe7907c3a18741a3","modelId":"claude-opus-4-7","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":48.88333333333333,"warning":null},{"id":"2d4d6debc6979aee50be3031","modelId":"claude-opus-4-7","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":44.733333333333334,"warning":null},{"id":"147757ed73e9880fc778d2b4","modelId":"claude-opus-4-7","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":88.59649122807019,"warning":null},{"id":"86cb9a8643a9e957a2818486","modelId":"claude-opus-4-7","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":73.9766081871345,"warning":null},{"id":"3260ef32a5d5523de7b68fdb","modelId":"claude-opus-4-7","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":34.639175257732,"warning":null},{"id":"53e61ad8c67c2d8ca29cdf71","modelId":"claude-opus-4-7","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":51.5151515151515,"warning":null},{"id":"139f2ae5d7aa248b64f48508","modelId":"claude-opus-4-7","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Claude Opus 4.7 (Non-reasoning, High)","configuration":"Claude Opus 4.7 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-opus-4-7-non-reasoning","score":54.5454545454545,"warning":null},{"id":"fc5cf8e34ddef72ce4b067b2","modelId":"claude-opus-4-7","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Claude Opus 4.7 (Max)","configuration":"Claude Opus 4.7 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-7","score":83.1460674157303,"warning":null},{"id":"da713bbecda4ed9bb6e4575a","modelId":"claude-opus-4-7","benchmarkId":"terminal-claudecode","level":"max","sourceLabel":"Opus 4.7","configuration":"Opus 4.7; Reported effort: max; 447 trials; recorded 2026-04-16","score":68.9,"warning":null},{"id":"f1a9e8085f72db6bccf6962b","modelId":"claude-opus-4-7","benchmarkId":"terminal-terminus2","level":"max","sourceLabel":"Opus 4.7","configuration":"Opus 4.7; Reported effort: max; 445 trials; recorded 2026-04-16","score":66.07,"warning":null},{"id":"65ddd6b7454ecf908d3f195a","modelId":"claude-opus-4-8","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":45,"warning":null},{"id":"8886355db52097534e7693f2","modelId":"claude-opus-4-8","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":1321.43,"warning":null},{"id":"fb5d73e50fc866659481b5fa","modelId":"claude-opus-4-8","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":77.6666666666667,"warning":null},{"id":"88573658b1f16a32767f85c5","modelId":"claude-opus-4-8","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":45.59343691527719,"warning":null},{"id":"33629a1e5c67bf1cec936707","modelId":"claude-opus-4-8","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":20.8571428571429,"warning":null},{"id":"11c80c3d3d1f51d0aa8ae988","modelId":"claude-opus-4-8","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":43.957027752909575,"warning":null},{"id":"c4714228dc4d583a0b07323b","modelId":"claude-opus-4-8","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":22.8,"warning":null},{"id":"17cfee3e3a4d9ad1ca6506d6","modelId":"claude-opus-4-8","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":1453.47,"warning":null},{"id":"df2b37e1775ec294a81b7593","modelId":"claude-opus-4-8","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":92.020202020202,"warning":null},{"id":"b7463337503ad39673a476d1","modelId":"claude-opus-4-8","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":48.6561631139944,"warning":null},{"id":"926f42994e94735363f94c2a","modelId":"claude-opus-4-8","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":62.244897959183696,"warning":null},{"id":"72167db58f6c4cf8a86c5024","modelId":"claude-opus-4-8","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":45.5555555555556,"warning":null},{"id":"f512427635d3a1eb3986839a","modelId":"claude-opus-4-8","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":28.75,"warning":null},{"id":"8fb0e686d83be066ccba241f","modelId":"claude-opus-4-8","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":48.833333333333336,"warning":null},{"id":"9b13a6e2c09e8a7ad1375e15","modelId":"claude-opus-4-8","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":54.398148148148195,"warning":null},{"id":"d188aab83722a63c82439dca","modelId":"claude-opus-4-8","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":94.4444444444445,"warning":null},{"id":"b2015a25bd1af98428159ba1","modelId":"claude-opus-4-8","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":34.226804123711304,"warning":null},{"id":"eff7039460ae03fa38a6104a","modelId":"claude-opus-4-8","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":58.3333333333333,"warning":null},{"id":"1044cd275af0efa750177c3e","modelId":"claude-opus-4-8","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":84.6441947565543,"warning":null},{"id":"adb60ff15f3922d889d1d9f2","modelId":"claude-opus-4-8","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Opus 4.8 (Max)","configuration":"Claude Opus 4.8 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-4-8","score":21.7171717171717,"warning":null},{"id":"b5c453e45d5a65111bf8c8fc","modelId":"claude-opus-4-8","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Opus 4.8 (High)","configuration":"Claude Opus 4.8 (High); Source model ID: anthropic-opus-4-8-high; 120 tasks","score":72.08,"warning":null},{"id":"7609122fbb5f21e0336fd79d","modelId":"claude-opus-4-8","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Claude Opus 4.8 (Low)","configuration":"Claude Opus 4.8 (Low); Source model ID: anthropic-opus-4-8-low; 120 tasks","score":62.22,"warning":null},{"id":"099afb744367e0027cb4beb0","modelId":"claude-opus-4-8","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Claude Opus 4.8 (Medium)","configuration":"Claude Opus 4.8 (Medium); Source model ID: anthropic-opus-4-8-medium; 120 tasks","score":71.67,"warning":null},{"id":"3ed225cfbd8da4a68cb773eb","modelId":"claude-opus-4-8","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"claude-opus-4-8","configuration":"claude-opus-4-8; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":51.8,"warning":null},{"id":"6529363c938719d7a1e4c752","modelId":"claude-opus-4-8","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"claude-opus-4-8","configuration":"claude-opus-4-8; Reported effort: low; 113 unique tasks; 451 attempts; 4 runs","score":40.8,"warning":null},{"id":"8158b49aa1444c7311339716","modelId":"claude-opus-4-8","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"claude-opus-4-8","configuration":"claude-opus-4-8; Reported effort: max; 111 unique tasks; 429 attempts; 4 runs","score":59,"warning":"Incomplete task coverage: 111 of 113 tasks. A common task set is required for ranking."},{"id":"00b724564a3c562424e7d7ee","modelId":"claude-opus-4-8","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"claude-opus-4-8","configuration":"claude-opus-4-8; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":48.7,"warning":null},{"id":"c0c9ff5a707bf93dda7c0ef5","modelId":"claude-opus-4-8","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"claude-opus-4-8","configuration":"claude-opus-4-8; Reported effort: xhigh; 113 unique tasks; 447 attempts; 4 runs","score":54.4,"warning":null},{"id":"766dde2b58473794011e3786","modelId":"claude-opus-4-8","benchmarkId":"terminal-claudecode","level":"high","sourceLabel":"Opus 4.8","configuration":"Opus 4.8; Reported effort: high; 445 trials; recorded 2026-05-28","score":78.88,"warning":null},{"id":"753ae551ccda6bd4cc94ecc2","modelId":"claude-opus-5","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":53.75,"warning":null},{"id":"2d7c6616a36f73b9abec34b2","modelId":"claude-opus-5","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":1560.7,"warning":null},{"id":"ffc4652f085361c9357b3c4f","modelId":"claude-opus-5","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":1209.28,"warning":null},{"id":"a3f5eaeeaa9e1a0a014fcd04","modelId":"claude-opus-5","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":1659.48,"warning":null},{"id":"ebf0edd233ce92220723ee86","modelId":"claude-opus-5","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":1434.89,"warning":null},{"id":"57b09c758d8afe45e4a06756","modelId":"claude-opus-5","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":1633.85,"warning":null},{"id":"c18984644d10f401652a36d0","modelId":"claude-opus-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":79,"warning":null},{"id":"9fedb771d38efbc432ec886a","modelId":"claude-opus-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":81.3333333333333,"warning":null},{"id":"311404ebf97a9ba739180a11","modelId":"claude-opus-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":79.3333333333333,"warning":null},{"id":"19dd786f2ad563d3436380ec","modelId":"claude-opus-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":82,"warning":null},{"id":"c0ae562e970e6c00bfe84de3","modelId":"claude-opus-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":80.3333333333333,"warning":null},{"id":"f00c4ba93207aba1401fc923","modelId":"claude-opus-5","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":53.56401051539629,"warning":null},{"id":"749f4979ff22b8103c053757","modelId":"claude-opus-5","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":51.79066298700786,"warning":null},{"id":"0418d18521aa922fc20bdc69","modelId":"claude-opus-5","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":56.57355576493251,"warning":null},{"id":"8bbbc737457c74a5e7dc1909","modelId":"claude-opus-5","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":54.33925431990897,"warning":null},{"id":"63df25e414d3eda4abc48672","modelId":"claude-opus-5","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":53.21919210948297,"warning":null},{"id":"463fcbe8804c69251a3e3311","modelId":"claude-opus-5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":28.2857142857143,"warning":null},{"id":"cdb06553d1a208c2b80caa53","modelId":"claude-opus-5","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":23.1428571428571,"warning":null},{"id":"c999a16f20587c650d92d9ce","modelId":"claude-opus-5","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":29.1428571428571,"warning":null},{"id":"f18538dddf551dd0a9d3e9ba","modelId":"claude-opus-5","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":26.8571428571429,"warning":null},{"id":"30b002253599a4d3ad1be1d5","modelId":"claude-opus-5","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":27.7142857142857,"warning":null},{"id":"d22df38366d64bb621649fa9","modelId":"claude-opus-5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":47.4783646672635,"warning":null},{"id":"4dc641bc65d811cd88c8bb84","modelId":"claude-opus-5","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":19.6,"warning":null},{"id":"b5f9932d4a6d9118062e5262","modelId":"claude-opus-5","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":17.2,"warning":null},{"id":"f5d2fcd4a1a9d03459ee7a47","modelId":"claude-opus-5","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":21.6,"warning":null},{"id":"6adaa58f22135037583de2d5","modelId":"claude-opus-5","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":20,"warning":null},{"id":"749f89dcfba44da317297928","modelId":"claude-opus-5","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":21,"warning":null},{"id":"934904bbe8471f7adfebf686","modelId":"claude-opus-5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":1594.93,"warning":null},{"id":"7c472104532d5a4d4b17d927","modelId":"claude-opus-5","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":1302.46,"warning":null},{"id":"4d290468a58cd88fa8ccf3f9","modelId":"claude-opus-5","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":1721.15,"warning":null},{"id":"39f076a7cd84a756e74bc78c","modelId":"claude-opus-5","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":1492.08,"warning":null},{"id":"c5cb55ab13cfbae90e710aa3","modelId":"claude-opus-5","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":1690.67,"warning":null},{"id":"4f6f3d8f7355f2168e6b797f","modelId":"claude-opus-5","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":93.7373737373737,"warning":null},{"id":"25be030f337ebad41589e420","modelId":"claude-opus-5","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":88.8888888888889,"warning":null},{"id":"77f7445976324c3644b0c3f8","modelId":"claude-opus-5","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":93.2323232323232,"warning":null},{"id":"be7019de2851cbc7372d5bab","modelId":"claude-opus-5","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":91.9191919191919,"warning":null},{"id":"1758f7975c3ff96cd35d5ff8","modelId":"claude-opus-5","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":93.7373737373737,"warning":null},{"id":"908e5a057d8960748d9f6ea6","modelId":"claude-opus-5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":52.8266913809083,"warning":null},{"id":"6be9e5d46bba96241c0886c9","modelId":"claude-opus-5","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":43.4198331788693,"warning":null},{"id":"1d3db39aec7c3192094696d2","modelId":"claude-opus-5","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":54.8656163113994,"warning":null},{"id":"7454dd6f79805b2bec9415af","modelId":"claude-opus-5","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":51.29749768303981,"warning":null},{"id":"4d98fdbfdc90e27f6d278318","modelId":"claude-opus-5","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":54.40222428174239,"warning":null},{"id":"4f1b1a542b24abc8c93945b7","modelId":"claude-opus-5","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":59.4444444444444,"warning":null},{"id":"da9a2215e00354802b05ae8b","modelId":"claude-opus-5","benchmarkId":"aa-mlcr-aa","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":53.8888888888889,"warning":null},{"id":"499d94ebf820a00159a9645b","modelId":"claude-opus-5","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":55.5555555555556,"warning":null},{"id":"c05bd701ad7726124f5f02aa","modelId":"claude-opus-5","benchmarkId":"aa-mlcr-aa","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":56.1111111111111,"warning":null},{"id":"ca2ced40cfb2923b480cd321","modelId":"claude-opus-5","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":58.3333333333333,"warning":null},{"id":"adef80d4a3e3d57512889e03","modelId":"claude-opus-5","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":82.4277456647399,"warning":null},{"id":"b130484a14ad902aeb8042a9","modelId":"claude-opus-5","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":79.8265895953757,"warning":null},{"id":"a0d794a1bd9d73d025241347","modelId":"claude-opus-5","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":84.7398843930636,"warning":null},{"id":"54d725c1e6ae6fc780bb2b69","modelId":"claude-opus-5","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":81.6184971098266,"warning":null},{"id":"ab40bc86f66b2afc5bafbb11","modelId":"claude-opus-5","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":83.9884393063584,"warning":null},{"id":"140db294c5eb909c8ba208ee","modelId":"claude-opus-5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":33.7166666666667,"warning":null},{"id":"88f928df06be55061e149bb6","modelId":"claude-opus-5","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":28.55,"warning":null},{"id":"a164b566a82427182f1a3eab","modelId":"claude-opus-5","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":37.0666666666667,"warning":null},{"id":"da7415490fb983c56c6f4665","modelId":"claude-opus-5","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":31.0166666666667,"warning":null},{"id":"ebdd606353ef09824dff9f22","modelId":"claude-opus-5","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":35.3833333333333,"warning":null},{"id":"a36dd3cab91d107cec849d23","modelId":"claude-opus-5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":58.88333333333333,"warning":null},{"id":"1c875c50a865ffda926b0176","modelId":"claude-opus-5","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":55.95,"warning":null},{"id":"3bdd06bf0b2b43314120a2a0","modelId":"claude-opus-5","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":60.86666666666667,"warning":null},{"id":"ac0d7494f0b9f62a149b11eb","modelId":"claude-opus-5","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":57.06666666666666,"warning":null},{"id":"12d880bfee5e2773707ebc6c","modelId":"claude-opus-5","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":59.5,"warning":null},{"id":"093cf9f9fde9935f885cedee","modelId":"claude-opus-5","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":55.4398148148148,"warning":null},{"id":"c39d270141be864f2639eaff","modelId":"claude-opus-5","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":49.189814814814795,"warning":null},{"id":"9a471570f6a64ed79fca4964","modelId":"claude-opus-5","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":56.3657407407407,"warning":null},{"id":"d95e11216bddf1634c371c2f","modelId":"claude-opus-5","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":51.5046296296296,"warning":null},{"id":"70dfe50b02b73db1cfa4c851","modelId":"claude-opus-5","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":55.671296296296305,"warning":null},{"id":"c53529ff9180aec31ead9038","modelId":"claude-opus-5","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":44.7422680412371,"warning":null},{"id":"cd698440f56c12039bb690d8","modelId":"claude-opus-5","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":30.309278350515502,"warning":null},{"id":"f1564e1613e4cf9d043fb6f7","modelId":"claude-opus-5","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":42.0618556701031,"warning":null},{"id":"ba03f98fa8d60d95820a278d","modelId":"claude-opus-5","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":38.5567010309278,"warning":null},{"id":"5df1da5328450ff388e0f1ca","modelId":"claude-opus-5","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":43.298969072165,"warning":null},{"id":"7a7274da713055e8c4a61ece","modelId":"claude-opus-5","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":28.571428571428598,"warning":null},{"id":"ac04007e87cd03c590d92df3","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":87.6404494382023,"warning":null},{"id":"df3d0e668abd0706d1ac4845","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":76.4044943820225,"warning":null},{"id":"2028eb18368634b3d692b553","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":89.1385767790262,"warning":null},{"id":"2d4925ee56b312a337d44e2c","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":86.1423220973783,"warning":null},{"id":"867bd714a169a36c0d7e7115","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":88.0149812734082,"warning":null},{"id":"04eeb390c041f6adc016f986","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Model source: https://artificialanalysis.ai/models/claude-opus-5-high","score":45.959595959596,"warning":null},{"id":"e5ba8487d8e9fa7485e0b7b8","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Claude Opus 5 (Low)","configuration":"Claude Opus 5 (Low); Model source: https://artificialanalysis.ai/models/claude-opus-5-low","score":26.2626262626263,"warning":null},{"id":"91bc9aa32a6f0e6f8003d581","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Model source: https://artificialanalysis.ai/models/claude-opus-5","score":48.989898989899,"warning":null},{"id":"bb90e9be720ea259988588fa","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Claude Opus 5 (Medium)","configuration":"Claude Opus 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-opus-5-medium","score":34.343434343434296,"warning":null},{"id":"470aec2611ff841aafcbe893","modelId":"claude-opus-5","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Claude Opus 5 (Xhigh)","configuration":"Claude Opus 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-opus-5-xhigh","score":46.4646464646465,"warning":null},{"id":"ee2fdd53366196c1c9cdaf3a","modelId":"claude-opus-5","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Source model ID: anthropic-claude-opus-5-high; 120 tasks","score":88.33,"warning":null},{"id":"48b4a19317b73c4a3f407097","modelId":"claude-opus-5","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Source model ID: anthropic-claude-opus-5-max; 120 tasks","score":90.42,"warning":null},{"id":"e57e43848926eef736a897ba","modelId":"claude-opus-5","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"claude-opus-5","configuration":"claude-opus-5; Reported effort: high; 113 unique tasks; 449 attempts; 4 runs","score":72.8,"warning":null},{"id":"f6a6f991a9a99441b39a26ca","modelId":"claude-opus-5","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"claude-opus-5","configuration":"claude-opus-5; Reported effort: low; 113 unique tasks; 449 attempts; 4 runs","score":58.1,"warning":null},{"id":"a0003347b8421c764d77ca4a","modelId":"claude-opus-5","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"claude-opus-5","configuration":"claude-opus-5; Reported effort: max; 113 unique tasks; 444 attempts; 4 runs","score":73.6,"warning":null},{"id":"0298f83d2256dcf2b5ed6e3f","modelId":"claude-opus-5","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"claude-opus-5","configuration":"claude-opus-5; Reported effort: medium; 113 unique tasks; 447 attempts; 4 runs","score":68.9,"warning":null},{"id":"d138eb10af8dd2cacd7a504a","modelId":"claude-opus-5","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"claude-opus-5","configuration":"claude-opus-5; Reported effort: xhigh; 113 unique tasks; 447 attempts; 4 runs","score":73.2,"warning":null},{"id":"b0d3a7d36775ded3600ba04c","modelId":"claude-opus-5-5","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":56.25,"warning":null},{"id":"82d5e8df767f94baf36abc8e","modelId":"claude-opus-5-5","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":1685.66,"warning":null},{"id":"7db3cf5f5746a4d1f885b6fe","modelId":"claude-opus-5-5","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":1279.79,"warning":null},{"id":"a9b80a0965cef38956583bcf","modelId":"claude-opus-5-5","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":1802.99,"warning":null},{"id":"e6b2c7f3ce3ab706b6c9262d","modelId":"claude-opus-5-5","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":1625.06,"warning":null},{"id":"309b850a473a4eb06ca36502","modelId":"claude-opus-5-5","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":1763.85,"warning":null},{"id":"2541879a89dd37858f47f4dc","modelId":"claude-opus-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":82.6666666666667,"warning":null},{"id":"40b0ded3ac8d8c77b77b1978","modelId":"claude-opus-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":80.6666666666667,"warning":null},{"id":"f60c8f502ff079881572e38a","modelId":"claude-opus-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":84.6666666666667,"warning":null},{"id":"a722f0ba610fcbfb39f04c73","modelId":"claude-opus-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":84.3333333333333,"warning":null},{"id":"e702f9f161d4d0063ffec926","modelId":"claude-opus-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":84.6666666666667,"warning":null},{"id":"4b159a8eb029c9dd3fb7f9d3","modelId":"claude-opus-5-5","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":63.206774831862134,"warning":null},{"id":"94ef189002096d8efa98a00e","modelId":"claude-opus-5-5","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":52.92537042934996,"warning":null},{"id":"1ec5fd417df7526f3ca636f9","modelId":"claude-opus-5-5","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":69.5386812863607,"warning":null},{"id":"868bd12c8cb86e2f5dfa1e0e","modelId":"claude-opus-5-5","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":61.21742919892179,"warning":null},{"id":"22a36344fcb85f0b38d4aad8","modelId":"claude-opus-5-5","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":64.95283516401527,"warning":null},{"id":"5ec4d9e0a8015bcfed6f1c5c","modelId":"claude-opus-5-5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":30.8571428571429,"warning":null},{"id":"2c2140166af8265468bb0264","modelId":"claude-opus-5-5","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":17.714285714285698,"warning":null},{"id":"ac5b653e4da3536a8a886166","modelId":"claude-opus-5-5","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":31.7142857142857,"warning":null},{"id":"2d3fd09d220ab937db58d906","modelId":"claude-opus-5-5","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":27.7142857142857,"warning":null},{"id":"156b01166c71c38ca981b51c","modelId":"claude-opus-5-5","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":31.7142857142857,"warning":null},{"id":"ad278f24421e9a21fd446c62","modelId":"claude-opus-5-5","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":28.799999999999997,"warning":null},{"id":"ba35bb83030538d8aaa57cc0","modelId":"claude-opus-5-5","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":25.6,"warning":null},{"id":"a24fdc81ce4731045936546a","modelId":"claude-opus-5-5","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":26.200000000000003,"warning":null},{"id":"8c689261b773ede1f426b711","modelId":"claude-opus-5-5","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":25.6,"warning":null},{"id":"38b6bc70ded277488f2006c7","modelId":"claude-opus-5-5","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":26.6,"warning":null},{"id":"7c108aaef2a400a56249a732","modelId":"claude-opus-5-5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":1704.37,"warning":null},{"id":"2b6322bd2e8d7c8c95282222","modelId":"claude-opus-5-5","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":1235.23,"warning":null},{"id":"4b220f675b59baa316d2108f","modelId":"claude-opus-5-5","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":1866.27,"warning":null},{"id":"d47868ca08c13ed7101f3906","modelId":"claude-opus-5-5","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":1583.6,"warning":null},{"id":"ed53d6c7079c30eb2b3d818c","modelId":"claude-opus-5-5","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":1831.09,"warning":null},{"id":"22b01e80a44e8d13d21ee3a2","modelId":"claude-opus-5-5","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":4.166666666666666,"warning":null},{"id":"bb1cb717c010fbd2a2e865e2","modelId":"claude-opus-5-5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":55.56070435588511,"warning":null},{"id":"f260aa641732c17a4d1aa9e5","modelId":"claude-opus-5-5","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":48.3317886932345,"warning":null},{"id":"cd1315ba948b558ab71b0fa9","modelId":"claude-opus-5-5","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":61.353104726598694,"warning":null},{"id":"b7f5b845012c3539eaf0234f","modelId":"claude-opus-5-5","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":54.72659870250231,"warning":null},{"id":"c01678aa616d8c15eedcd1d2","modelId":"claude-opus-5-5","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":57.5069508804449,"warning":null},{"id":"21b5c8d9c9be6d1abd7aefcf","modelId":"claude-opus-5-5","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":38.229755178907695,"warning":null},{"id":"25dda3fe4c5e6e9e479310db","modelId":"claude-opus-5-5","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":66.6666666666667,"warning":null},{"id":"6b7f84b4e53309237143020e","modelId":"claude-opus-5-5","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":85.8381502890173,"warning":null},{"id":"8bc32c5733549edd5eaedabd","modelId":"claude-opus-5-5","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":84.6820809248555,"warning":null},{"id":"a10a32e543537b09bf4cd991","modelId":"claude-opus-5-5","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":87.6878612716763,"warning":null},{"id":"1642397d29e0befa66c92b36","modelId":"claude-opus-5-5","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":85.72254335260119,"warning":null},{"id":"c53c05143c6ab5c75a1d84aa","modelId":"claude-opus-5-5","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":86.5895953757225,"warning":null},{"id":"302dc2a5ff2ef491e86105a8","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":40.6333333333333,"warning":null},{"id":"ca17d9492048571313d749b5","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":38.8666666666667,"warning":null},{"id":"24768a7c096deb69be9c64d4","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":46.4166666666667,"warning":null},{"id":"9e1c61275e8bbef77da24d59","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":40.3166666666667,"warning":null},{"id":"be3ac51366605e77ab20a4a1","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":42.65,"warning":null},{"id":"08ec5224a9cebb9ab7549a8e","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":64.58333333333334,"warning":null},{"id":"3bbc40c110afd37161e15188","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":63.53333333333333,"warning":null},{"id":"31265066b6f219fb4770aefe","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":66.21666666666667,"warning":null},{"id":"ab1649abc7ea666ad5ea15fe","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":64.55,"warning":null},{"id":"1e14877f84b551f62ef8af3e","modelId":"claude-opus-5-5","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":65.38333333333334,"warning":null},{"id":"143dbd2e432d661d58233d8c","modelId":"claude-opus-5-5","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":60.4166666666667,"warning":null},{"id":"6943c8595f70352be7cf2957","modelId":"claude-opus-5-5","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":58.5648148148148,"warning":null},{"id":"5a1643112f9311f86075f36b","modelId":"claude-opus-5-5","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":66.8981481481482,"warning":null},{"id":"0ab225a78ac35909f1c0a198","modelId":"claude-opus-5-5","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":59.2592592592593,"warning":null},{"id":"b4f7c0523c077121c46b5f30","modelId":"claude-opus-5-5","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":65.0462962962963,"warning":null},{"id":"d462a80934a3302f9e13b4ed","modelId":"claude-opus-5-5","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":49.047619047619,"warning":null},{"id":"39cae1fce4b9bb1d16501f75","modelId":"claude-opus-5-5","benchmarkId":"aa-terminal-bench-science","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":24.2857142857143,"warning":null},{"id":"deaf3523cf8d3ce67f40f57e","modelId":"claude-opus-5-5","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":59.047619047619,"warning":null},{"id":"600753769dc5e2a230ba37e7","modelId":"claude-opus-5-5","benchmarkId":"aa-terminal-bench-science","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":43.3333333333333,"warning":null},{"id":"a23074f9f8a70b48f67fad6b","modelId":"claude-opus-5-5","benchmarkId":"aa-terminal-bench-science","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":61.9047619047619,"warning":null},{"id":"53b6036a713f4c74f4a70c2b","modelId":"claude-opus-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Claude Opus 5.5 (High, Default Fallback)","configuration":"Claude Opus 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-high","score":56.565656565656596,"warning":null},{"id":"e148a8c2166dcd6a54b4a8e7","modelId":"claude-opus-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Claude Opus 5.5 (Low, Default Fallback)","configuration":"Claude Opus 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-low","score":31.313131313131297,"warning":null},{"id":"e33095ba4c2ed5bddb7f655e","modelId":"claude-opus-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Claude Opus 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5","score":59.5959595959596,"warning":null},{"id":"0a81e1cf246bc1c5eff60038","modelId":"claude-opus-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium, Default Fallback)","configuration":"Claude Opus 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-medium","score":52.5252525252525,"warning":null},{"id":"16016608bc4248d23d282b06","modelId":"claude-opus-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Claude Opus 5.5 (Xhigh, Default Fallback)","configuration":"Claude Opus 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-opus-5-5-xhigh","score":59.5959595959596,"warning":null},{"id":"10908990b8182621c8f4f8d6","modelId":"claude-opus-5-5","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Opus 5.5 (High)","configuration":"Claude Opus 5.5 (High); Source model ID: anthropic-claude-opus-5-5-high; 120 tasks","score":93.33333333333333,"warning":null},{"id":"9b2d75c09b222a335b5b0420","modelId":"claude-opus-5-5","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Claude Opus 5.5 (Low)","configuration":"Claude Opus 5.5 (Low); Source model ID: anthropic-claude-opus-5-5-low; 120 tasks","score":70.1388888888889,"warning":null},{"id":"7406bb0c1b1e4f2cd4e9fed8","modelId":"claude-opus-5-5","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"Claude Opus 5.5 (Max)","configuration":"Claude Opus 5.5 (Max); Source model ID: anthropic-claude-opus-5-5-max; 120 tasks","score":91.66666666666666,"warning":null},{"id":"fcccddaa6664dcd9b4cf9ad3","modelId":"claude-opus-5-5","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Claude Opus 5.5 (Medium)","configuration":"Claude Opus 5.5 (Medium); Source model ID: anthropic-claude-opus-5-5-medium; 120 tasks","score":87.5,"warning":null},{"id":"adc443abcd6bfeb63a5a417c","modelId":"claude-opus-5-5","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"Claude Opus 5.5 (XHigh)","configuration":"Claude Opus 5.5 (XHigh); Source model ID: anthropic-claude-opus-5-5-xhigh; 120 tasks","score":92.5,"warning":null},{"id":"1a1b9aa1227717597b6847b5","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":713.27,"warning":null},{"id":"8a159d4fef16041d56762dd8","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":37,"warning":null},{"id":"de1e01647292e9102ed45aa0","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":88,"warning":null},{"id":"674b1c581fa7dd12ee0b66cc","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":54,"warning":null},{"id":"db806a12c42b98a164ee977d","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":72.3333333333333,"warning":null},{"id":"1774a54d662320565d170e76","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":14.075743049062714,"warning":null},{"id":"34d2f64ecde456514b2e2710","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":0,"warning":null},{"id":"4a36ccc9c0f665a4b53b1d6b","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":1.1428571428571401,"warning":null},{"id":"1016ec1494cf8458139671e2","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":5.2,"warning":null},{"id":"3199e1446609878558cc7e69","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":910.05,"warning":null},{"id":"8f403f5bd47ca2833fc80bf0","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":89.25520833333334,"warning":null},{"id":"2b31de44a316fc16fda8177e","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":72.72727272727269,"warning":null},{"id":"df5144f0bc3dfefc0a136256","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":83.4343434343434,"warning":null},{"id":"f783eb29b70679064120eaa9","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":7.228915662650599,"warning":null},{"id":"c60e757ef68afcde43a56bc8","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":17.7942539388323,"warning":null},{"id":"e9b0ce7d16a5006a78eca66c","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":42.653061224489804,"warning":null},{"id":"752154fc7bb6ab5b9437f3c3","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":57.2789115646258,"warning":null},{"id":"7076314c6b18338bfa363af1","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":59.047619047619,"warning":null},{"id":"6c9293e23e64f29131b3d278","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":71.4285714285714,"warning":null},{"id":"40bbe6df11b8de4e7657ecca","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":85.9873670212766,"warning":null},{"id":"266d8dc9f86ccb95c6645fa9","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":87.4501329787234,"warning":null},{"id":"b3c0845808f574dbc704d157","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":65.2023121387283,"warning":null},{"id":"3e0d99a12bb7de08ad26556c","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":68.728323699422,"warning":null},{"id":"c712e425b0902b25a4bca75e","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":-9.36666666666667,"warning":null},{"id":"3efd3c04564d707cb7af0c5c","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":-0.0833333333333333,"warning":null},{"id":"5d5aed22ee29eec8262d15fa","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":28.383333333333333,"warning":null},{"id":"9e508c8c6b578671ad893039","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":32.86666666666667,"warning":null},{"id":"843d8eff7e70d1387558ec55","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":45.717592592592595,"warning":null},{"id":"d0a93edeaedf53b41d00dbd3","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":70.4678362573099,"warning":null},{"id":"6722be7d2b55c374d9bb637c","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":78.07017543859651,"warning":null},{"id":"f9f243e03759b358a57a4007","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":24.5360824742268,"warning":null},{"id":"6535f19e64380503ae7e87fe","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Claude 4.5 Sonnet (Non-reasoning)","configuration":"Claude 4.5 Sonnet (Non-reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet","score":28.7878787878788,"warning":null},{"id":"4e1797d1fb9ba82a6999f801","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":35.6060606060606,"warning":null},{"id":"bac4143f9a745d98bfeb22e4","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":55.805243445692895,"warning":null},{"id":"1f78303c3188cf807ba58ec0","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Claude 4.5 Sonnet (Reasoning)","configuration":"Claude 4.5 Sonnet (Reasoning); Model source: https://artificialanalysis.ai/models/claude-4-5-sonnet-thinking","score":0,"warning":null},{"id":"3f070c56f3e96293b3502873","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Sonnet 4.5 (Thinking 32K)","configuration":"Claude Sonnet 4.5 (Thinking 32K); Source model ID: claude-sonnet-4-5-20250929-thinking-32k; 120 tasks","score":13.61,"warning":null},{"id":"54de588ab99dc61e79aad8e5","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Sonnet 4.5 (Thinking 16K)","configuration":"Claude Sonnet 4.5 (Thinking 16K); Source model ID: claude-sonnet-4-5-20250929-thinking-16k; 120 tasks","score":6.94,"warning":null},{"id":"e513347234a010f067c79909","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Sonnet 4.5 (Thinking 1K)","configuration":"Claude Sonnet 4.5 (Thinking 1K); Source model ID: claude-sonnet-4-5-20250929-thinking-1k; 120 tasks","score":5.83,"warning":null},{"id":"e78c77d5cf9767ea4f53df60","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Claude Sonnet 4.5 (Thinking 8K)","configuration":"Claude Sonnet 4.5 (Thinking 8K); Source model ID: claude-sonnet-4-5-20250929-thinking-8k; 120 tasks","score":6.94,"warning":null},{"id":"1d61f2abfed894c939457377","modelId":"claude-sonnet-4-6","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":20,"warning":null},{"id":"61d31423c8242d4750fb45e3","modelId":"claude-sonnet-4-6","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":1062.96,"warning":null},{"id":"6e055900e434a08afaa864f2","modelId":"claude-sonnet-4-6","benchmarkId":"aa-apex-agents-aa","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":28.023598820058996,"warning":null},{"id":"9e6344bb7e2b7d1cc6b279f1","modelId":"claude-sonnet-4-6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":80,"warning":null},{"id":"27530616174a8f787e91c784","modelId":"claude-sonnet-4-6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":69.3333333333333,"warning":null},{"id":"8bbb15b7b00975fd75772621","modelId":"claude-sonnet-4-6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":68.3333333333333,"warning":null},{"id":"d40b938958343e508989b1f1","modelId":"claude-sonnet-4-6","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":20.098073664518367,"warning":null},{"id":"f7477c91311ae43b9633f3ba","modelId":"claude-sonnet-4-6","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":3.14285714285714,"warning":null},{"id":"9c93c91cb711addf1c4fa892","modelId":"claude-sonnet-4-6","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":0.857142857142857,"warning":null},{"id":"d4cb98101e2aae6472871fd6","modelId":"claude-sonnet-4-6","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":0.857142857142857,"warning":null},{"id":"d3b0cb69ebe2dd2f074ca9bb","modelId":"claude-sonnet-4-6","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":15.8,"warning":null},{"id":"18c64883cc863ce6b2cfbb5b","modelId":"claude-sonnet-4-6","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":1232,"warning":null},{"id":"ae3dffd71edc88976c1b0647","modelId":"claude-sonnet-4-6","benchmarkId":"aa-global-mmlu-lite","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":90.484375,"warning":null},{"id":"6171bd1b3aeff19b98f03d67","modelId":"claude-sonnet-4-6","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":88.23958333333333,"warning":null},{"id":"c0747a8f7573ade81afb51eb","modelId":"claude-sonnet-4-6","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":89.23958333333334,"warning":null},{"id":"c71d9e6eb0084d3b114d8a80","modelId":"claude-sonnet-4-6","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":87.4747474747475,"warning":null},{"id":"78a71abdf16dd749a78783cb","modelId":"claude-sonnet-4-6","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":79.8989898989899,"warning":null},{"id":"aa1e4adb1a8ec6b26d22469f","modelId":"claude-sonnet-4-6","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":79.6969696969697,"warning":null},{"id":"ff8cfcf684aa81ac724e7516","modelId":"claude-sonnet-4-6","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":33.595922150139,"warning":null},{"id":"8063019679e03cfb746392ce","modelId":"claude-sonnet-4-6","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":13.3456904541242,"warning":null},{"id":"d279070c2d38929fdb77aa85","modelId":"claude-sonnet-4-6","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":11.2140871177016,"warning":null},{"id":"807078434f5494d34d0a79a7","modelId":"claude-sonnet-4-6","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":56.5986394557823,"warning":null},{"id":"030e25b703a598bc1234cf6e","modelId":"claude-sonnet-4-6","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":41.156462585034,"warning":null},{"id":"8afde1d0a2ccf273ca948fc3","modelId":"claude-sonnet-4-6","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":42.3809523809524,"warning":null},{"id":"b9c5a949b02ceffa72995b33","modelId":"claude-sonnet-4-6","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":39.8022598870057,"warning":null},{"id":"0503aab79fb5018379d57412","modelId":"claude-sonnet-4-6","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":24.4444444444444,"warning":null},{"id":"963a3c96c19569be4330ad94","modelId":"claude-sonnet-4-6","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":20,"warning":null},{"id":"d8072c6c9d5fae1f30806393","modelId":"claude-sonnet-4-6","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":20.5555555555556,"warning":null},{"id":"a9c308abff16e8e80548a3de","modelId":"claude-sonnet-4-6","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":73.2947976878613,"warning":null},{"id":"19e63d8aec4216f5120452a9","modelId":"claude-sonnet-4-6","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":70.5780346820809,"warning":null},{"id":"be4e64ba8034537372c9c1d0","modelId":"claude-sonnet-4-6","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":69.19075144508669,"warning":null},{"id":"e7398421eab52c0e986070f4","modelId":"claude-sonnet-4-6","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":12.2166666666667,"warning":null},{"id":"7392150f1198f62f4b2879f6","modelId":"claude-sonnet-4-6","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":-3.55,"warning":null},{"id":"a6fa79a6c14528865e202ae8","modelId":"claude-sonnet-4-6","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":-2.1,"warning":null},{"id":"65431a407554611313177afd","modelId":"claude-sonnet-4-6","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":40.86666666666667,"warning":null},{"id":"8fea1adfc17b5869acd4794a","modelId":"claude-sonnet-4-6","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":36.516666666666666,"warning":null},{"id":"dd37dab3527fc9b73af7a482","modelId":"claude-sonnet-4-6","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":38.550000000000004,"warning":null},{"id":"ebff159fbc5de1bdcff80fec","modelId":"claude-sonnet-4-6","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":50.1157407407407,"warning":null},{"id":"e2d735850e86d178767b312a","modelId":"claude-sonnet-4-6","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":75.7309941520468,"warning":null},{"id":"490a984f281c9931dee460b5","modelId":"claude-sonnet-4-6","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":78.9473684210526,"warning":null},{"id":"b390d26c0a476ac3a49c3ea9","modelId":"claude-sonnet-4-6","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":79.5321637426901,"warning":null},{"id":"1fa859a810a6c1843cc1fe4e","modelId":"claude-sonnet-4-6","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":34.4329896907216,"warning":null},{"id":"27a48a90b965fc5a24cb1fa8","modelId":"claude-sonnet-4-6","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":53.030303030303,"warning":null},{"id":"03dbde466df814133b51e33d","modelId":"claude-sonnet-4-6","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, High)","configuration":"Claude Sonnet 4.6 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6","score":46.2121212121212,"warning":null},{"id":"d0969ff841573b014a09218e","modelId":"claude-sonnet-4-6","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Claude Sonnet 4.6 (Non-reasoning, Low)","configuration":"Claude Sonnet 4.6 (Non-reasoning, Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-non-reasoning-low-effort","score":42.4242424242424,"warning":null},{"id":"93b5cb4ffda39a308b3d11c4","modelId":"claude-sonnet-4-6","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":71.1610486891386,"warning":null},{"id":"7a5aa0d448e5956ea3efc834","modelId":"claude-sonnet-4-6","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-4-6-adaptive","score":3.03030303030303,"warning":null},{"id":"3fc5dd5ad41a965cc918cdf2","modelId":"claude-sonnet-4-6","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Claude Sonnet 4.6 (High)","configuration":"Claude Sonnet 4.6 (High); Source model ID: claude_sonnet_4_6_high; 120 tasks","score":60.419999999999995,"warning":null},{"id":"d1a628a019bde02cfbdacf5e","modelId":"claude-sonnet-4-6","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"Claude Sonnet 4.6 (Max)","configuration":"Claude Sonnet 4.6 (Max); Source model ID: claude_sonnet_4_6_max; 120 tasks","score":58.330000000000005,"warning":null},{"id":"46adf8595dc3ea458bbcbb0e","modelId":"claude-sonnet-4-6","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"claude-sonnet-4-6","configuration":"claude-sonnet-4-6; Reported effort: high; 113 unique tasks; 451 attempts; 4 runs","score":29.9,"warning":null},{"id":"009b3a85e026c832fce72710","modelId":"claude-sonnet-5","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":46.25,"warning":null},{"id":"224850d9d58de545f8a79325","modelId":"claude-sonnet-5","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":1176.53,"warning":null},{"id":"80a39a8657c32d0acac497c3","modelId":"claude-sonnet-5","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":926.63,"warning":null},{"id":"942c2a8a3e7155e91c385135","modelId":"claude-sonnet-5","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":1358.4,"warning":null},{"id":"caeb58993d1166ca60d128e9","modelId":"claude-sonnet-5","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":1057.7,"warning":null},{"id":"ea564a29acde7aa3a293c896","modelId":"claude-sonnet-5","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":1273.1,"warning":null},{"id":"d61804932a7bf796852dc78b","modelId":"claude-sonnet-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":76.6666666666667,"warning":null},{"id":"0a54d910891aae916762fc4d","modelId":"claude-sonnet-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":67.3333333333333,"warning":null},{"id":"56c684b0a01b99025cdd4153","modelId":"claude-sonnet-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":82,"warning":null},{"id":"4671a9f9a0d40e6f9c40643c","modelId":"claude-sonnet-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":73.6666666666667,"warning":null},{"id":"e8b35a81f7ac8b44be687f6b","modelId":"claude-sonnet-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":70,"warning":null},{"id":"12deda62e68f19722c504987","modelId":"claude-sonnet-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":76.6666666666667,"warning":null},{"id":"58059f7e91260ed90996ccbc","modelId":"claude-sonnet-5","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":32.0966388143804,"warning":null},{"id":"3e9b0f413bca69f9ad1d17a6","modelId":"claude-sonnet-5","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":19.90193656897387,"warning":null},{"id":"9248c3853443e53768afc18b","modelId":"claude-sonnet-5","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":36.51267571703639,"warning":null},{"id":"e56fd25b451255bf61748b03","modelId":"claude-sonnet-5","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":27.872150606397167,"warning":null},{"id":"f96eebba51fd83bbf5658d8c","modelId":"claude-sonnet-5","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":34.47626889062095,"warning":null},{"id":"ab3891e46bad11376babc68a","modelId":"claude-sonnet-5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":15.1428571428571,"warning":null},{"id":"98ee3a3176559ed9aaa5f1f2","modelId":"claude-sonnet-5","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":4.57142857142857,"warning":null},{"id":"629e2c1d6874bd954db3e9e1","modelId":"claude-sonnet-5","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":16.8571428571429,"warning":null},{"id":"5c0bc781fda61980b39a2e0a","modelId":"claude-sonnet-5","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":8.57142857142857,"warning":null},{"id":"42c6201fdfcdac3ecbdf3f77","modelId":"claude-sonnet-5","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":1.1428571428571401,"warning":null},{"id":"0fcc65e2732904db0776c612","modelId":"claude-sonnet-5","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":15.4285714285714,"warning":null},{"id":"25696098729be54115fb8457","modelId":"claude-sonnet-5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":44.67323187108326,"warning":null},{"id":"33fdffb77c33d74827f29346","modelId":"claude-sonnet-5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":43.86750223813787,"warning":null},{"id":"23fcce925d6e017cab7e220c","modelId":"claude-sonnet-5","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":9.4,"warning":null},{"id":"4514553ecdbe876883be0bcd","modelId":"claude-sonnet-5","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":9.4,"warning":null},{"id":"9779f73be74f98f7ddd7be9c","modelId":"claude-sonnet-5","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":13.200000000000001,"warning":null},{"id":"f93d10f2a08e4e3dc58366a7","modelId":"claude-sonnet-5","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":11.600000000000001,"warning":null},{"id":"f3c8571f70e5ee0a32b74758","modelId":"claude-sonnet-5","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":12.6,"warning":null},{"id":"fe1f63723d2f9db36d317c2a","modelId":"claude-sonnet-5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":1254.48,"warning":null},{"id":"84a0a733316c5843f82e98eb","modelId":"claude-sonnet-5","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":1067.28,"warning":null},{"id":"289c32294ddd85e591f16ee7","modelId":"claude-sonnet-5","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":1463.81,"warning":null},{"id":"733f407e98726ebb79e39ad3","modelId":"claude-sonnet-5","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":1156.51,"warning":null},{"id":"c4b75cb9bd4d57b0ebb3d8d7","modelId":"claude-sonnet-5","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":1209.69,"warning":null},{"id":"979c87685813c03cf4f6e629","modelId":"claude-sonnet-5","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":1357.91,"warning":null},{"id":"2f3c7528a160f317604bea7d","modelId":"claude-sonnet-5","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":91.1111111111111,"warning":null},{"id":"12f51bb25b612b21a8fbf9e0","modelId":"claude-sonnet-5","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":80,"warning":null},{"id":"6408a04e31b16fa9698fb9c1","modelId":"claude-sonnet-5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":35.7275254865616,"warning":null},{"id":"750cf5738a688721d44d2b35","modelId":"claude-sonnet-5","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":21.9184430027804,"warning":null},{"id":"7c8b4f26ffe553325343a1ac","modelId":"claude-sonnet-5","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":41.2882298424467,"warning":null},{"id":"06ea5cb3c53334b37fcfb923","modelId":"claude-sonnet-5","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":29.9814643188137,"warning":null},{"id":"0039f53ce9442f0c5357ce07","modelId":"claude-sonnet-5","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":19.0454124189064,"warning":null},{"id":"bb7b4fb71fc48b9e1144d5ea","modelId":"claude-sonnet-5","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":39.017608897127,"warning":null},{"id":"1f238be1efecc7dda10af3cb","modelId":"claude-sonnet-5","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":55.00000000000001,"warning":null},{"id":"5a4c80288505f0ad98350099","modelId":"claude-sonnet-5","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":77.28323699421969,"warning":null},{"id":"d42139c140597a61dfc0a1e6","modelId":"claude-sonnet-5","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":71.90751445086701,"warning":null},{"id":"463b38ef7a40f6adff0048b2","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":-3.68333333333333,"warning":null},{"id":"21e96c3f3a26974f3c958514","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":-8.23333333333333,"warning":null},{"id":"f93fe9ba3256a3c652bd27aa","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":16.45,"warning":null},{"id":"8b305b6faddeac3de70ff49b","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":-6.86666666666667,"warning":null},{"id":"a60cdab1630c63ecb318ae4a","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":-0.65,"warning":null},{"id":"9f4089dfbf594cbf7298e930","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":3.18333333333333,"warning":null},{"id":"06a5ee8b96fecf5550834295","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":37.43333333333334,"warning":null},{"id":"d7ae06368198adc01f75b507","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":37.35,"warning":null},{"id":"6f48732eff38308580af22bd","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":40.050000000000004,"warning":null},{"id":"35790006c3ca7096b09e208e","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":37.1,"warning":null},{"id":"e9b70c2e0f05e22df0d3803b","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":33.800000000000004,"warning":null},{"id":"50217728be0b391e8e9415a0","modelId":"claude-sonnet-5","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":38.96666666666667,"warning":null},{"id":"c00af2ab066171f76a56a2ce","modelId":"claude-sonnet-5","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":54.2824074074074,"warning":null},{"id":"a14cc610ff685bf92eb2e9f2","modelId":"claude-sonnet-5","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":50.1157407407407,"warning":null},{"id":"8edb37e82094f84ce2b378b7","modelId":"claude-sonnet-5","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":54.2824074074074,"warning":null},{"id":"5f58711702294d24e90328ad","modelId":"claude-sonnet-5","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":51.6203703703704,"warning":null},{"id":"1b7342699ab8a254f4eccc48","modelId":"claude-sonnet-5","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":54.050925925925895,"warning":null},{"id":"c6ffe36cce75fd7e742dfff3","modelId":"claude-sonnet-5","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":37.319587628866,"warning":null},{"id":"59745d1742e79ff24f2a897e","modelId":"claude-sonnet-5","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":15.6701030927835,"warning":null},{"id":"119c54d4b5b48352b6f0939d","modelId":"claude-sonnet-5","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":80.5243445692884,"warning":null},{"id":"d845260785d3f5133f85ba54","modelId":"claude-sonnet-5","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Claude Sonnet 5 (Non-reasoning, High)","configuration":"Claude Sonnet 5 (Non-reasoning, High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-non-reasoning","score":75.2808988764045,"warning":null},{"id":"7fc5c8ca8c2b4beff316d2d1","modelId":"claude-sonnet-5","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-high","score":5.05050505050505,"warning":null},{"id":"2667f6083d1d783cf746c35e","modelId":"claude-sonnet-5","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Claude Sonnet 5 (Low)","configuration":"Claude Sonnet 5 (Low); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-low","score":2.52525252525253,"warning":null},{"id":"df618e9592963789e14982f5","modelId":"claude-sonnet-5","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Sonnet 5 (Max)","configuration":"Claude Sonnet 5 (Max); Model source: https://artificialanalysis.ai/models/claude-sonnet-5","score":14.1414141414141,"warning":null},{"id":"25a489bc5d0aef6c45763a82","modelId":"claude-sonnet-5","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Claude Sonnet 5 (Medium)","configuration":"Claude Sonnet 5 (Medium); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-medium","score":2.02020202020202,"warning":null},{"id":"f221505cd3c57bec7fd7eafb","modelId":"claude-sonnet-5","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Claude Sonnet 5 (Xhigh)","configuration":"Claude Sonnet 5 (Xhigh); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-xhigh","score":7.07070707070707,"warning":null},{"id":"52163093753564a0e2539873","modelId":"claude-sonnet-5","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"claude-sonnet-5","configuration":"claude-sonnet-5; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":48.2,"warning":null},{"id":"11a57984aea1a8da93db31c3","modelId":"claude-sonnet-5","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"claude-sonnet-5","configuration":"claude-sonnet-5; Reported effort: low; 113 unique tasks; 449 attempts; 4 runs","score":30.5,"warning":null},{"id":"9fd633042f2f5198fccbe085","modelId":"claude-sonnet-5","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"claude-sonnet-5","configuration":"claude-sonnet-5; Reported effort: max; 113 unique tasks; 442 attempts; 4 runs","score":53.8,"warning":null},{"id":"a42f15ae42cd5b945a0d0f4f","modelId":"claude-sonnet-5","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"claude-sonnet-5","configuration":"claude-sonnet-5; Reported effort: medium; 113 unique tasks; 450 attempts; 4 runs","score":39.8,"warning":null},{"id":"856339a53bf647f60852283a","modelId":"claude-sonnet-5","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"claude-sonnet-5","configuration":"claude-sonnet-5; Reported effort: xhigh; 113 unique tasks; 451 attempts; 4 runs","score":49.7,"warning":null},{"id":"aaae9fdfb56cae71deebae41","modelId":"claude-sonnet-5","benchmarkId":"terminal-claudecode","level":"high","sourceLabel":"Sonnet 5","configuration":"Sonnet 5; Reported effort: high; 445 trials; recorded 2026-06-30","score":74.61,"warning":null},{"id":"601a9db0513c94649add0c40","modelId":"claude-sonnet-5-5","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":57.49999999999999,"warning":null},{"id":"f370c0e51bce2e33b7806283","modelId":"claude-sonnet-5-5","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":1635.77,"warning":null},{"id":"fd7593f4bf59d0664e34c13c","modelId":"claude-sonnet-5-5","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":1271.53,"warning":null},{"id":"7e7a93700e440bbd01d760e4","modelId":"claude-sonnet-5-5","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":1817.26,"warning":null},{"id":"c005396a9d501d6bd597f667","modelId":"claude-sonnet-5-5","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":1441.38,"warning":null},{"id":"a5998a94e40b68aa1d267105","modelId":"claude-sonnet-5-5","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":1750.96,"warning":null},{"id":"d8f3bce7ec3db35129f7218f","modelId":"claude-sonnet-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":78,"warning":null},{"id":"a691985bc8b35b0e0b33f6fc","modelId":"claude-sonnet-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":76,"warning":null},{"id":"3898d17f98a92bbea0caf82d","modelId":"claude-sonnet-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":82.6666666666667,"warning":null},{"id":"0941f7ab0e084b6443cc47b5","modelId":"claude-sonnet-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":76.3333333333333,"warning":null},{"id":"321cd2b19ee41eec52277211","modelId":"claude-sonnet-5-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":79.6666666666667,"warning":null},{"id":"aa75f4c6c547b00a4210659e","modelId":"claude-sonnet-5-5","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":59.42923026067375,"warning":null},{"id":"3de18ebfffca818568c0f3b3","modelId":"claude-sonnet-5-5","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":49.42092641344955,"warning":null},{"id":"925b7de70a66a8762f910144","modelId":"claude-sonnet-5-5","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":71.76882128193995,"warning":null},{"id":"5456d663eca0bae68021cf90","modelId":"claude-sonnet-5-5","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":54.90962996121462,"warning":null},{"id":"449cbae6f9d82f3d778f4f20","modelId":"claude-sonnet-5-5","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":65.52026949665557,"warning":null},{"id":"10745788e455ab97c9e194f7","modelId":"claude-sonnet-5-5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":24.571428571428598,"warning":null},{"id":"b81421d2d8a5aee4db6175d2","modelId":"claude-sonnet-5-5","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":11.4285714285714,"warning":null},{"id":"e663da7494adfbd64cee7d9d","modelId":"claude-sonnet-5-5","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":31.4285714285714,"warning":null},{"id":"bfe2896e38c206a3d1431cea","modelId":"claude-sonnet-5-5","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":16.8571428571429,"warning":null},{"id":"737be0249f7fccc7e2c95dfb","modelId":"claude-sonnet-5-5","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":31.1428571428571,"warning":null},{"id":"4abeea8445ce38753fa77a0c","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":25.2,"warning":null},{"id":"e11480fd2c3277993de50a4a","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":16,"warning":null},{"id":"d6af01cc08e9792cfb6dec8f","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":25.8,"warning":null},{"id":"060e54e132551071be47021a","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":20.200000000000003,"warning":null},{"id":"bc120d9bfbf49b482eb3f372","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":24.6,"warning":null},{"id":"3ff6ca9f251f9d482c3b3000","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":1549.39,"warning":null},{"id":"1d61e0b092cd38504a5c871d","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":1175.34,"warning":null},{"id":"dcc1354340394416c73b6ac2","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":1837.82,"warning":null},{"id":"b2cac882750c42fae0e1268e","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":1323.27,"warning":null},{"id":"fe02e809cde21148b6c711f8","modelId":"claude-sonnet-5-5","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":1728.44,"warning":null},{"id":"04e9ed9306db8ac657de38bd","modelId":"claude-sonnet-5-5","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":2.7777777777777777,"warning":null},{"id":"971b31b166049e37f7c09ffd","modelId":"claude-sonnet-5-5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":45.8294717330862,"warning":null},{"id":"47541c8b73ead0d2f21e946b","modelId":"claude-sonnet-5-5","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":36.2372567191844,"warning":null},{"id":"01127ecc8f06fece1dfed761","modelId":"claude-sonnet-5-5","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":54.9582947173309,"warning":null},{"id":"7d0ea4be9dd82b81d3b428e3","modelId":"claude-sonnet-5-5","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":39.805375347544,"warning":null},{"id":"197440608bcfa5d941cc39a2","modelId":"claude-sonnet-5-5","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":50,"warning":null},{"id":"f9b007a887bbab3c15300a10","modelId":"claude-sonnet-5-5","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":75,"warning":null},{"id":"c96e3100b2cb1ebe04115ed3","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":20.95,"warning":null},{"id":"7d297321a0a395af5f0e3b03","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":19.3666666666667,"warning":null},{"id":"70aed5a8f5959e55dea3dfdd","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":32.3,"warning":null},{"id":"a176441933be023f28218629","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":20.1,"warning":null},{"id":"d1dfc5817dd62938a0537a02","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":23.5,"warning":null},{"id":"fbffabaf317dcc99c4aca30e","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":51.983333333333334,"warning":null},{"id":"2cc8c80d15b03c0483dcba6b","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":46.300000000000004,"warning":null},{"id":"1907b656bd509a3647e3f5fc","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":53.949999999999996,"warning":null},{"id":"279e5aa37e94bb0f474a9489","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":47.16666666666667,"warning":null},{"id":"5d8d965962add78d86f0f291","modelId":"claude-sonnet-5-5","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":53.03333333333333,"warning":null},{"id":"78b9446cfe37098965498801","modelId":"claude-sonnet-5-5","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":53.703703703703695,"warning":null},{"id":"68a4f73ee8faf5087dd463fb","modelId":"claude-sonnet-5-5","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":49.0740740740741,"warning":null},{"id":"532918e7edc7acb0bc0e991b","modelId":"claude-sonnet-5-5","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":60.9953703703704,"warning":null},{"id":"d2563bf03bb787a9737d9e91","modelId":"claude-sonnet-5-5","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":52.8935185185185,"warning":null},{"id":"bbf569cf2c38fab3f89b8da3","modelId":"claude-sonnet-5-5","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":57.2916666666667,"warning":null},{"id":"8147bc46a12cdcae42a8b1d7","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":34.285714285714306,"warning":null},{"id":"dd9be8f0efdd52f99175a0f3","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminal-bench-science","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":10.4761904761905,"warning":null},{"id":"c40e2d8e35163bc8252c9297","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":53.3333333333333,"warning":null},{"id":"412d9f66362aa3dac0c23137","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminal-bench-science","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":16.1904761904762,"warning":null},{"id":"3554766c0ee9322f8b9d32ec","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminal-bench-science","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":52.380952380952394,"warning":null},{"id":"acd66def8ec3de8e5109a394","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Claude Sonnet 5.5 (High, Default Fallback)","configuration":"Claude Sonnet 5.5 (High, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-high","score":43.9393939393939,"warning":null},{"id":"d7a20749701815a0f350486c","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Claude Sonnet 5.5 (Low, Default Fallback)","configuration":"Claude Sonnet 5.5 (Low, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-low","score":20.7070707070707,"warning":null},{"id":"0249af7ec555e2c7f7311f58","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max, Default Fallback)","configuration":"Claude Sonnet 5.5 (Max, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5","score":63.636363636363605,"warning":null},{"id":"15ebcefae0ec05757d900b89","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Claude Sonnet 5.5 (Medium, Default Fallback)","configuration":"Claude Sonnet 5.5 (Medium, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-medium","score":29.7979797979798,"warning":null},{"id":"a07ebaa78f18b6f4a6cbdb28","modelId":"claude-sonnet-5-5","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (Xhigh, Default Fallback)","configuration":"Claude Sonnet 5.5 (Xhigh, Default Fallback); Model source: https://artificialanalysis.ai/models/claude-sonnet-5-5-xhigh","score":57.07070707070711,"warning":null},{"id":"78f4b6a0fa1a6e9adedb94e2","modelId":"command-a-plus-05-2026","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":355.85,"warning":null},{"id":"e21fc7e41b06c5f966ba5005","modelId":"command-a-plus-05-2026","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.66666666666669,"warning":null},{"id":"201b76826240a98b3cc2f619","modelId":"command-a-plus-05-2026","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.8104537047977383,"warning":null},{"id":"2ccdbb5004b77a13dc61b015","modelId":"command-a-plus-05-2026","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"c284b9431d0d16bbad751202","modelId":"command-a-plus-05-2026","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":1,"warning":null},{"id":"4dfbfb167ad4313ab89d4560","modelId":"command-a-plus-05-2026","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":514.67,"warning":null},{"id":"10229395b4f5570e9e94c618","modelId":"command-a-plus-05-2026","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.06060606060609,"warning":null},{"id":"0b91e3c66e0cbd10f89c4adf","modelId":"command-a-plus-05-2026","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.9555143651529,"warning":null},{"id":"79f1835a76a8ef96e8c7db02","modelId":"command-a-plus-05-2026","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.9455782312925,"warning":null},{"id":"641499b97a2b15845679955b","modelId":"command-a-plus-05-2026","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.555555555555556,"warning":null},{"id":"ec037f6ae1a0299e9e2a7784","modelId":"command-a-plus-05-2026","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.2369942196532,"warning":null},{"id":"9fe83ac17762f2d5b91590ca","modelId":"command-a-plus-05-2026","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":-4.01666666666667,"warning":null},{"id":"26670e73b2cc07d434922fd0","modelId":"command-a-plus-05-2026","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.883333333333333,"warning":null},{"id":"5ab70dacb88f83e57d5e2076","modelId":"command-a-plus-05-2026","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.5416666666667,"warning":null},{"id":"45dbb1834092da47ce482a2c","modelId":"command-a-plus-05-2026","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.7017543859649,"warning":null},{"id":"12d6380c2e451719a74ba2ee","modelId":"command-a-plus-05-2026","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.97938144329897,"warning":null},{"id":"b428a724421ec67f5d001f52","modelId":"command-a-plus-05-2026","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":25,"warning":null},{"id":"e07117252be6e1880685c4d4","modelId":"command-a-plus-05-2026","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.8464419475655,"warning":null},{"id":"b4b54c7c7c50aac1e09c9338","modelId":"command-a-plus-05-2026","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Command A+","configuration":"Command A+; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/command-a-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.505050505050505,"warning":null},{"id":"0f2099408e3868dd3f5bbd0d","modelId":"command-r-03-2024","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Command-R+ (Apr '24)","configuration":"Command-R+ (Apr '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-plus-04-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.3232323232323,"warning":null},{"id":"e7b2966cb801f40d53a63d0d","modelId":"command-r-03-2024","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Command-R (Mar '24)","configuration":"Command-R (Mar '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-03-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.383838383838402,"warning":null},{"id":"78c2d72883d8472bf84bb942","modelId":"command-r-03-2024","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Command-R (Mar '24)","configuration":"Command-R (Mar '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-03-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.79,"warning":null},{"id":"c51be1971aa7732d0e3c9dec","modelId":"command-r-03-2024","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Command-R+ (Apr '24)","configuration":"Command-R+ (Apr '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-plus-04-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.63,"warning":null},{"id":"8738d7c5dfd75c32c1227708","modelId":"command-r-03-2024","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Command-R (Mar '24)","configuration":"Command-R (Mar '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-03-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.76190476190476,"warning":null},{"id":"f43c9503098639be7baabd71","modelId":"command-r-03-2024","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Command-R+ (Apr '24)","configuration":"Command-R+ (Apr '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-plus-04-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.169312169312201,"warning":null},{"id":"14c3bdfd1b0511c0d8bd8d7b","modelId":"command-r-03-2024","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Command-R (Mar '24)","configuration":"Command-R (Mar '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-03-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.400000000000002,"warning":null},{"id":"9fd73563f403665e135e24bc","modelId":"command-r-03-2024","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Command-R+ (Apr '24)","configuration":"Command-R+ (Apr '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-plus-04-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.8666666666667,"warning":null},{"id":"2deb81947d71e3f165d2316b","modelId":"command-r-03-2024","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Command-R+ (Apr '24)","configuration":"Command-R+ (Apr '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-plus-04-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.1765292553192,"warning":null},{"id":"bfe86af3326b7bb69930545c","modelId":"command-r-03-2024","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Command-R (Mar '24)","configuration":"Command-R (Mar '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/command-r-03-2024; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.843085106383,"warning":null},{"id":"28154d8ea2e87ddacaa50495","modelId":"deepseek-r1-0528","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":76,"warning":null},{"id":"d1f9132429ab5fb1330d2974","modelId":"deepseek-r1-0528","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.6666666666667,"warning":null},{"id":"d457e9fe5dc1d1849bde49ee","modelId":"deepseek-r1-0528","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.42857142857143,"warning":null},{"id":"89c583f754304a1af65e90d8","modelId":"deepseek-r1-0528","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":85.97916666666667,"warning":null},{"id":"89234d302d97c8daf59156c8","modelId":"deepseek-r1-0528","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.31313131313131,"warning":null},{"id":"2894a03cfe0dd1c35eb63b61","modelId":"deepseek-r1-0528","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.8480074142725,"warning":null},{"id":"51289684897e664af54a05d2","modelId":"deepseek-r1-0528","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.5918367346939,"warning":null},{"id":"f3bf8f6694da5448f8d82c11","modelId":"deepseek-r1-0528","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.03703703703701,"warning":null},{"id":"10731a3c1d964999d4ff6598","modelId":"deepseek-r1-0528","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":98.26666666666671,"warning":null},{"id":"ac159579de0c6415663fa834","modelId":"deepseek-r1-0528","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.8570478723404,"warning":null},{"id":"da46f877760d8faa5938e0c0","modelId":"deepseek-r1-0528","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":-27.4333333333333,"warning":null},{"id":"b93f0ed6713171396a82dc72","modelId":"deepseek-r1-0528","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.533333333333335,"warning":null},{"id":"18a2418d73fee9d97b9e6213","modelId":"deepseek-r1-0528","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.5497076023392,"warning":null},{"id":"91a19a0300fec337c23eaa24","modelId":"deepseek-r1-0528","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"DeepSeek R1 0528 (May '25)","configuration":"DeepSeek R1 0528 (May '25); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.909090909090901,"warning":null},{"id":"58ce3731b6917805aad4c062","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.66666666666671,"warning":null},{"id":"08fd0fb9fdd7b4017f5af622","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":15,"warning":null},{"id":"2258149dfb5467587a60f57e","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"30fce54c7ca06f7f9c3150ac","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":61.2121212121212,"warning":null},{"id":"502b0514328b85336c43b79d","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.9314179796107505,"warning":null},{"id":"278e99ccd1bf0c3c5d492f6c","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.863945578231302,"warning":null},{"id":"6a61d3d8a1884f195532baf1","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.3227513227513,"warning":null},{"id":"50f9af8abf72ce37fba0fb3a","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":93.2,"warning":null},{"id":"752650e6b94ff2d88f9139d6","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.8779920212766,"warning":null},{"id":"dfd5bf4e982521e955c382fc","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-63.6166666666667,"warning":null},{"id":"dcaa7d394803f4ee83bee90f","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.2,"warning":null},{"id":"5fbc7e5ecd7f934aed6ef04d","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"29dfea68f2a7733cb8a8cecb","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"DeepSeek R1 0528 Qwen3 8B","configuration":"DeepSeek R1 0528 Qwen3 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-qwen3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.51515151515152,"warning":null},{"id":"243c60a86656d1595c85745b","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.66666666666669,"warning":null},{"id":"8dd78fb19901b64635db44f7","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10,"warning":null},{"id":"2db469a257b0c3f2f9b40bfd","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"e6222d4f378654bfa05436ac","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.2020202020202,"warning":null},{"id":"17ee47efdbcf843333a4b243","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.13,"warning":null},{"id":"ae831c640f2a28b64b82e0c4","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.5510204081633,"warning":null},{"id":"b9505e9c34edbb45f0c1117f","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.560846560846603,"warning":null},{"id":"6fc19818675129310568bb8b","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":93.4666666666667,"warning":null},{"id":"01313163168fd0c915717c34","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.4797207446808,"warning":null},{"id":"c8e893e3e38807fa7c5bc79c","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-46.3833333333333,"warning":null},{"id":"2f9d19dbc5ed104d40cdfb95","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.166666666666668,"warning":null},{"id":"4b37220933b4b948d9b751ff","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.9298245614035,"warning":null},{"id":"63f152f6d7e4690aaad8b67a","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 70B","configuration":"DeepSeek R1 Distill Llama 70B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-70b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.51515151515152,"warning":null},{"id":"717d45d07de6c052364ae732","modelId":"deepseek-r1-distill-llama-8b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 8B","configuration":"DeepSeek R1 Distill Llama 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.3333333333333,"warning":null},{"id":"0411fa720378242cd32f3640","modelId":"deepseek-r1-distill-llama-8b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 8B","configuration":"DeepSeek R1 Distill Llama 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"202acf3ea41812346b29502c","modelId":"deepseek-r1-distill-llama-8b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 8B","configuration":"DeepSeek R1 Distill Llama 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.2020202020202,"warning":null},{"id":"28f69c15c898a22e11d5c5c4","modelId":"deepseek-r1-distill-llama-8b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 8B","configuration":"DeepSeek R1 Distill Llama 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.5510204081633,"warning":null},{"id":"0ea361c2d08a7937fde27b2c","modelId":"deepseek-r1-distill-llama-8b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 8B","configuration":"DeepSeek R1 Distill Llama 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.2804232804233,"warning":null},{"id":"9dead492c7767af300580f03","modelId":"deepseek-r1-distill-llama-8b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 8B","configuration":"DeepSeek R1 Distill Llama 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":85.26666666666671,"warning":null},{"id":"36c671e4534c9ff66d2efdea","modelId":"deepseek-r1-distill-llama-8b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Llama 8B","configuration":"DeepSeek R1 Distill Llama 8B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-llama-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.255319148936195,"warning":null},{"id":"a116bf871d78b619440a0d27","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":22,"warning":null},{"id":"3732ec9546845932aab5bfbe","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.66666666666667,"warning":null},{"id":"9e90674679d47770bee4c60f","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.7979797979798,"warning":null},{"id":"055397757bbab069a14a098c","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.08,"warning":null},{"id":"fa0a9617f5d6b929618a9d65","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.1972789115646,"warning":null},{"id":"967ea0ca7b70675dd27176b0","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.98412698412698,"warning":null},{"id":"b6a1f34951caea6c3455f61b","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.7333333333333,"warning":null},{"id":"78dda2ed784ee488b907a8a8","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 1.5B","configuration":"DeepSeek R1 Distill Qwen 1.5B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-1-5b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.9198803191489,"warning":null},{"id":"df13747a2c4fb0617b290384","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.6666666666667,"warning":null},{"id":"f6729439534a49931b93c6d3","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.3333333333333,"warning":null},{"id":"5a99dc39b2489e5491d366a7","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.3838383838384,"warning":null},{"id":"5db07063f3dc97fd82ec4610","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.130000000000001,"warning":null},{"id":"f0f8badfdf79f42dbd45cc88","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.108843537414998,"warning":null},{"id":"b88a2bb503c88bf49553b4e1","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.5661375661376,"warning":null},{"id":"664755cb2071b85ced978db3","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":94.8666666666667,"warning":null},{"id":"8c5936db8bc598b54ecd95b9","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 14B","configuration":"DeepSeek R1 Distill Qwen 14B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.0026595744681,"warning":null},{"id":"36958495a1c3cc263aa8ca57","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":63,"warning":null},{"id":"48284251b9eed81c130233ab","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.66666666666667,"warning":null},{"id":"ad1db369c6baf73ac00539d7","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":61.5151515151515,"warning":null},{"id":"88242b8c0ec5a510a77d7f1d","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.58,"warning":null},{"id":"a99251c3bb7ceba3f13ebc1d","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.9251700680272,"warning":null},{"id":"833ffe1e8e5b5830b4a596e2","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.984126984127,"warning":null},{"id":"5e67615c87f9dbb2127e0805","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":94.06666666666669,"warning":null},{"id":"9efc3ef712c44c990e27aaa1","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek R1 Distill Qwen 32B","configuration":"DeepSeek R1 Distill Qwen 32B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-r1-distill-qwen-32b; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.936170212766,"warning":null},{"id":"51db25dec7dd9b6ee1af0a92","modelId":"deepseek-v3.2","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":59,"warning":null},{"id":"27b5897f1b8a957838bd8cd2","modelId":"deepseek-v3.2","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":92,"warning":null},{"id":"55fd90923f763bdbaf99506b","modelId":"deepseek-v3.2","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":14.5280235988201,"warning":null},{"id":"7bb7a5261517f9b7393edbbf","modelId":"deepseek-v3.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":45.6666666666667,"warning":null},{"id":"509e652963248d4d5fb3d747","modelId":"deepseek-v3.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":73.3333333333333,"warning":null},{"id":"257d1e613d6552d3fb81fcd4","modelId":"deepseek-v3.2","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":0.857142857142857,"warning":null},{"id":"ca37adba46b2003742a66663","modelId":"deepseek-v3.2","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":2.85714285714286,"warning":null},{"id":"73e6dc3a2c843f32fc95666b","modelId":"deepseek-v3.2","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":691.52,"warning":null},{"id":"406b55d44096c3fc11bd6e0e","modelId":"deepseek-v3.2","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":82.67708333333333,"warning":null},{"id":"c3e8ddd7c4d1151df0f2e9b5","modelId":"deepseek-v3.2","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":86.54166666666667,"warning":null},{"id":"95ce60be501a5563083110d5","modelId":"deepseek-v3.2","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":75.0505050505051,"warning":null},{"id":"078e78833b376471c91df11e","modelId":"deepseek-v3.2","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":84.040404040404,"warning":null},{"id":"aa22308da67df2bbc79a5257","modelId":"deepseek-v3.2","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":11.2140871177016,"warning":null},{"id":"375a2df992ecb3932c79a0a9","modelId":"deepseek-v3.2","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":24.5597775718258,"warning":null},{"id":"91a9cfa474cd837eb548b3df","modelId":"deepseek-v3.2","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":49.0476190476191,"warning":null},{"id":"0258d20c0150addb336ab7da","modelId":"deepseek-v3.2","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":60.6802721088435,"warning":null},{"id":"2e55950f34a54a20d60015ea","modelId":"deepseek-v3.2","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":59.2592592592593,"warning":null},{"id":"abafce31201ae669660cbac9","modelId":"deepseek-v3.2","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":86.2433862433862,"warning":null},{"id":"cd243264a418b155eec7573d","modelId":"deepseek-v3.2","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":83.7101063829787,"warning":null},{"id":"b7211928466c4d961deff26b","modelId":"deepseek-v3.2","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":86.2117686170213,"warning":null},{"id":"ad70fb15b77d81ac74c0ce38","modelId":"deepseek-v3.2","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":-46.8833333333333,"warning":null},{"id":"9bdadba43260d89af264c859","modelId":"deepseek-v3.2","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":-22.4833333333333,"warning":null},{"id":"560f2b6ebb5e7968026d533c","modelId":"deepseek-v3.2","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":24,"warning":null},{"id":"7560286cbd56eaed8556521b","modelId":"deepseek-v3.2","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":32.96666666666667,"warning":null},{"id":"9f6e7bb623a6f650fd1e7a7f","modelId":"deepseek-v3.2","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":78.9473684210526,"warning":null},{"id":"b1cdb72f437d4c38c7e68f79","modelId":"deepseek-v3.2","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":90.6432748538012,"warning":null},{"id":"c6d9bb636099f72e73c8f57f","modelId":"deepseek-v3.2","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"DeepSeek V3.2 (Non-reasoning)","configuration":"DeepSeek V3.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2","score":32.5757575757576,"warning":null},{"id":"54cd965a443d2ae081646656","modelId":"deepseek-v3.2","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":35.6060606060606,"warning":null},{"id":"7fa83144d986891b98ca55e1","modelId":"deepseek-v3.2","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"DeepSeek V3.2 (Reasoning)","configuration":"DeepSeek V3.2 (Reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v3-2-reasoning","score":46.8164794007491,"warning":null},{"id":"5de71161ca30dde6ea78d6e2","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":96.6666666666667,"warning":null},{"id":"edc9ec337f8ec9c00318a38f","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":70,"warning":null},{"id":"da1da6326a1c6365f377d1b3","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.428571428571431,"warning":null},{"id":"19939f3f32bedc31c6f2e423","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.36979166666667,"warning":null},{"id":"40fac96214150f7c9055b2f8","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":87.0707070707071,"warning":null},{"id":"934f82bf796a5f145286aefa","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.7303058387396,"warning":null},{"id":"0d6fbc4c2f6e63c01d55902c","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.8775510204082,"warning":null},{"id":"d7a6befd16d9965c0f855044","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":89.6296296296296,"warning":null},{"id":"841d60685fb658fe47a7ef71","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.3115026595745,"warning":null},{"id":"f6d49d36dd36d19e284b4d40","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":-17.7166666666667,"warning":null},{"id":"9e9c78a7ffb2a500222ed9aa","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.766666666666666,"warning":null},{"id":"866f5d18b5c5f028af163a6c","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"bd04ee261df4be08b3396245","modelId":"deepseek-v3.2-speciale","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"DeepSeek V3.2 Speciale","configuration":"DeepSeek V3.2 Speciale; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/deepseek-v3-2-speciale; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.8484848484849,"warning":null},{"id":"2b9e5024704a0a12064b7224","modelId":"deepseek-v4-flash","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":1258.64,"warning":null},{"id":"1b98edddb4c98044a6d02471","modelId":"deepseek-v4-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":79.6666666666667,"warning":null},{"id":"817e5985ecacbf5c45a69ab7","modelId":"deepseek-v4-flash","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":53.96771855133674,"warning":null},{"id":"c13b1d59ea008020f5ed8fbb","modelId":"deepseek-v4-flash","benchmarkId":"aa-critpt","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":16.5714285714286,"warning":null},{"id":"686a5811535b6316f0589c49","modelId":"deepseek-v4-flash","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":11,"warning":null},{"id":"a6a696aa8920008b2d665efd","modelId":"deepseek-v4-flash","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":1433.74,"warning":null},{"id":"8073965ff2f8e65afc61a17c","modelId":"deepseek-v4-flash","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":90.8080808080808,"warning":null},{"id":"0b125071bb27c0955e5c625e","modelId":"deepseek-v4-flash","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":38.5542168674699,"warning":null},{"id":"97d8368fc0ce32755db66924","modelId":"deepseek-v4-flash","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":13.3333333333333,"warning":null},{"id":"7eddb21b48546623d8b9ceae","modelId":"deepseek-v4-flash","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":-14.2833333333333,"warning":null},{"id":"8cec04b830602f64e1c46fd0","modelId":"deepseek-v4-flash","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":40.38333333333333,"warning":null},{"id":"3788db944176c99c751ddcdd","modelId":"deepseek-v4-flash","benchmarkId":"aa-scicode","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":50.3472222222222,"warning":null},{"id":"4e8004764005d17c2d933c76","modelId":"deepseek-v4-flash","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":39.381443298969096,"warning":null},{"id":"c89ea9debbc596c19e5ad450","modelId":"deepseek-v4-flash","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":78.65168539325839,"warning":null},{"id":"ebeba43f9adf3c0cee448c2d","modelId":"deepseek-v4-flash","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash","score":12.1212121212121,"warning":null},{"id":"39a7bed5705711ac27a52863","modelId":"deepseek-v4-flash","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (High)","configuration":"DeepSeek V4 Flash 0731 (High); Source model ID: deepseek-v4-flash-0731-high; 120 tasks","score":55.972222222222214,"warning":null},{"id":"c846403c911a1743e84cc3c3","modelId":"deepseek-v4-flash","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"DeepSeek V4 Flash 0731 (Low)","configuration":"DeepSeek V4 Flash 0731 (Low); Source model ID: deepseek-v4-flash-0731-low; 120 tasks","score":45.97222222222222,"warning":null},{"id":"bcce4d406862908e10d2d67e","modelId":"deepseek-v4-flash","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (Max)","configuration":"DeepSeek V4 Flash 0731 (Max); Source model ID: deepseek-v4-flash-0731-max; 120 tasks","score":61.38888888888888,"warning":null},{"id":"1a96e7a8059bfa86f6a2589c","modelId":"deepseek-v4-flash","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"DeepSeek V4 Flash 0731 (None)","configuration":"DeepSeek V4 Flash 0731 (None); Source model ID: deepseek-v4-flash-0731-none; 120 tasks","score":2.083333333333333,"warning":null},{"id":"8b3518db91f03758794ffbbc","modelId":"deepseek-v4-flash","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"deepseek-v4-flash","configuration":"deepseek-v4-flash; Reported effort: max; 113 unique tasks; 452 attempts; 4 runs","score":53.3,"warning":null},{"id":"ee3d10cdf6bbb1e427459014","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":25,"warning":null},{"id":"61aca6b7539c84826da44436","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":1043.64,"warning":null},{"id":"9fbb8189a6a5912ee0ce7625","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":835.84,"warning":null},{"id":"bb324ec20532a0e6acc9340c","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":72,"warning":null},{"id":"1006b3effa16e5e3ceaf5df0","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":74.3333333333333,"warning":null},{"id":"8ed7a58f0f02bb1ab6f68d52","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":41.6666666666667,"warning":null},{"id":"a4f28e058e6a8af0d4137df3","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":28.504323745973586,"warning":null},{"id":"09d32e7ee540bddf225e41dd","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":27.985986830616554,"warning":null},{"id":"87b1613ddf79d0ed468aca4b","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-critpt","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":3.42857142857143,"warning":null},{"id":"2c5b0222df63432d50090b0b","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-critpt","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":7.14285714285714,"warning":null},{"id":"a9d1b304b5ead4a80b530f71","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"e0725f2747a50513240537b4","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":39.57027752909579,"warning":null},{"id":"d9522462aacf237d2cac3af2","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":10.4,"warning":null},{"id":"39e793845e5c413eb72e17e2","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":10.8,"warning":null},{"id":"c6fe113d0426bcafa8ca2589","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":1004.54,"warning":null},{"id":"c7cdc43912da0da22dbb5aa0","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":1035.87,"warning":null},{"id":"a0f6250fbf3f72727b7f2cc6","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":86.6666666666667,"warning":null},{"id":"8d0de1ebd011819896803a10","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":89.3939393939394,"warning":null},{"id":"3e61a1984432422f7f33a108","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":71.6161616161616,"warning":null},{"id":"573824f6830aa35a7e3f3152","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":30.259499536607997,"warning":null},{"id":"fc1c32b1c24ae73fe394ce0a","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":34.8470806302132,"warning":null},{"id":"2cfac03b641be502ca31c268","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":7.784986098239111,"warning":null},{"id":"eda87d69cb1f115761d80ea9","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":73.4693877551021,"warning":null},{"id":"75dcc0fba24a84839477187c","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":79.1836734693877,"warning":null},{"id":"92d59e07969e4703740ab1f1","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":47.2108843537415,"warning":null},{"id":"84f13776cc0401cbf5f7f948","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":31.516007532956703,"warning":null},{"id":"a9f01970d3e67495318e3049","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":6.66666666666667,"warning":null},{"id":"3c33ed61e07c7fc8ca399805","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":15,"warning":null},{"id":"2a4f099e8be9e64b14fbaafb","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":-23.0833333333333,"warning":null},{"id":"0c1dafd1d8bac92ab1d83278","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":-23.9,"warning":null},{"id":"d01834dbbcf94f0a87f0400b","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":-43.9666666666667,"warning":null},{"id":"992e9b5c855c997d9d829cbd","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":34.86666666666667,"warning":null},{"id":"b7b9d0ffc9f1ba93d751f70e","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":36.81666666666666,"warning":null},{"id":"34e70355e4ea2d35feb5b31e","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":26,"warning":null},{"id":"e23fd324f0b5dba92baee5f7","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-scicode","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":40.162037037037,"warning":null},{"id":"045e27fe3e938eea61091981","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-scicode","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":45.2546296296296,"warning":null},{"id":"c867a89873d13efc48fe8c0f","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":95.6140350877193,"warning":null},{"id":"33b339eb5d043a4b0c44d54a","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":95.0292397660819,"warning":null},{"id":"aceb2c8a49f02fee8df04f65","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":94.4444444444445,"warning":null},{"id":"6c9f0d1dd8a8d28c325a1bcb","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":26.185567010309303,"warning":null},{"id":"70b988aa6b63a27538b3cb1d","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":30.9278350515464,"warning":null},{"id":"0582ee6c96fa6fa36eb8e651","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":38.636363636363605,"warning":null},{"id":"d92bd46050ce9699b73cfa9d","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":35.6060606060606,"warning":null},{"id":"f7cea72247cf3471dad7aa0e","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"DeepSeek V4 Flash 0420 (Non-reasoning)","configuration":"DeepSeek V4 Flash 0420 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-non-reasoning","score":34.0909090909091,"warning":null},{"id":"cfb0545f6a6da7f52061cf70","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":56.9288389513109,"warning":null},{"id":"1973bb3bed7e15df51b0b385","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":61.797752808988804,"warning":null},{"id":"e291d41d20cd377b87d6af89","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"DeepSeek V4 Flash 0420 (High)","configuration":"DeepSeek V4 Flash 0420 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420-high","score":3.03030303030303,"warning":null},{"id":"466b24114cb0a09078e6225b","modelId":"deepseek-v4-flash-0424","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"DeepSeek V4 Flash 0420 (Max)","configuration":"DeepSeek V4 Flash 0420 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-flash-0420","score":2.52525252525253,"warning":null},{"id":"e404cefc679aed7055063e54","modelId":"deepseek-v4-pro","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":1256.94,"warning":null},{"id":"d2be3baacb2f88392754e02a","modelId":"deepseek-v4-pro","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":928.2,"warning":null},{"id":"318e032120e13f4967057cb8","modelId":"deepseek-v4-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":80.3333333333333,"warning":null},{"id":"08a9a20e6f7fa296fe6eb550","modelId":"deepseek-v4-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":50.66666666666671,"warning":null},{"id":"f85861d2fe3c9512988dc6d4","modelId":"deepseek-v4-pro","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":56.711655463440025,"warning":null},{"id":"bff9e87251f2afc8a55016cb","modelId":"deepseek-v4-pro","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":34.90865332353694,"warning":null},{"id":"739bf9016e97683e8ac3cfeb","modelId":"deepseek-v4-pro","benchmarkId":"aa-critpt","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":18,"warning":null},{"id":"1a2e20b67e9f4c04b3b8ccea","modelId":"deepseek-v4-pro","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":0,"warning":null},{"id":"7985a9eb0c39954e717eccaf","modelId":"deepseek-v4-pro","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":49.59713518352731,"warning":null},{"id":"d9e4ef464fc1c08d66c035b6","modelId":"deepseek-v4-pro","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":11.4,"warning":null},{"id":"ada82bf2eae26a6ac8082e1b","modelId":"deepseek-v4-pro","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":3.5999999999999996,"warning":null},{"id":"e7eca8769a262ba2e4881131","modelId":"deepseek-v4-pro","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":1450.47,"warning":null},{"id":"6238725c91e3d11df7051ebf","modelId":"deepseek-v4-pro","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":1057.89,"warning":null},{"id":"d2fa9b10ad796302be38a96d","modelId":"deepseek-v4-pro","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":92.82828282828281,"warning":null},{"id":"6df3723f01db9a8fe2d2f548","modelId":"deepseek-v4-pro","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":41.0101946246525,"warning":null},{"id":"6c54446ae29130945ad023bb","modelId":"deepseek-v4-pro","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":10.5653382761817,"warning":null},{"id":"f26fb8536c24ab8bc35f248c","modelId":"deepseek-v4-pro","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":17.7777777777778,"warning":null},{"id":"c6f5d7997d3920e36cc9a135","modelId":"deepseek-v4-pro","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":0.833333333333333,"warning":null},{"id":"953fbb84fee7ea9bb2daa7f2","modelId":"deepseek-v4-pro","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":-12.3833333333333,"warning":null},{"id":"3d7d47a50f975b946724a1c9","modelId":"deepseek-v4-pro","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":49.1,"warning":null},{"id":"6fabfe1a747eb2065ac5a1ea","modelId":"deepseek-v4-pro","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":41.449999999999996,"warning":null},{"id":"71fbef21a92b2b9266e84dc4","modelId":"deepseek-v4-pro","benchmarkId":"aa-scicode","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":51.0416666666667,"warning":null},{"id":"8ccb6d811a15f748f06abe30","modelId":"deepseek-v4-pro","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":39.9305555555556,"warning":null},{"id":"a2c058cf47a21d4fb3ed8220","modelId":"deepseek-v4-pro","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":39.587628865979404,"warning":null},{"id":"27d489700e8190d346c80f6d","modelId":"deepseek-v4-pro","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":5.71428571428571,"warning":null},{"id":"cc78985bbe964e95b18f995a","modelId":"deepseek-v4-pro","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":78.65168539325839,"warning":null},{"id":"6545fcf07ca74c168934e7c5","modelId":"deepseek-v4-pro","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro","score":14.1414141414141,"warning":null},{"id":"dbce7ca5622b48cc19ff2f62","modelId":"deepseek-v4-pro","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0813 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0813 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-non-reasoning","score":1.51515151515152,"warning":null},{"id":"c64dd03820fdf345d94145b0","modelId":"deepseek-v4-pro","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"DeepSeek V4 Pro 0813 (High)","configuration":"DeepSeek V4 Pro 0813 (High); Source model ID: deepseek-v4-pro-0813-high; 120 tasks","score":59.72222222222221,"warning":null},{"id":"0f79426f1f5a6b4e49959783","modelId":"deepseek-v4-pro","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"DeepSeek V4 Pro 0813 (Low)","configuration":"DeepSeek V4 Pro 0813 (Low); Source model ID: deepseek-v4-pro-0813-low; 120 tasks","score":56.25,"warning":null},{"id":"25337d15a8b7bed1bf41c830","modelId":"deepseek-v4-pro","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (Max)","configuration":"DeepSeek V4 Pro 0813 (Max); Source model ID: deepseek-v4-pro-0813-max; 120 tasks","score":61.25000000000001,"warning":null},{"id":"d725d4dd6ab63e2174499861","modelId":"deepseek-v4-pro","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"DeepSeek V4 Pro 0813 (None)","configuration":"DeepSeek V4 Pro 0813 (None); Source model ID: deepseek-v4-pro-0813-none; 120 tasks","score":0.8333333333333334,"warning":null},{"id":"ac3167852c54be45702e8943","modelId":"deepseek-v4-pro","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"deepseek-v4-pro","configuration":"deepseek-v4-pro; Reported effort: max; 113 unique tasks; 452 attempts; 4 runs","score":62.8,"warning":"DeepSeek V4 Pro checkpoint is not identified by this source; April and August revisions must not be combined."},{"id":"1acd303b95f08abed126e960","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":18.75,"warning":null},{"id":"5c3e5e1981333cde5f341b12","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":922.01,"warning":null},{"id":"dd986e271fb52c5f6542e9d9","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-apex-agents-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":24.2625368731563,"warning":null},{"id":"0d06ffbfcb8fc52ca595d1bb","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":70.3333333333333,"warning":null},{"id":"f2de6940d94d16486e4e54c7","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":74.6666666666667,"warning":null},{"id":"a799f33ed2c8840436d37022","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":53,"warning":null},{"id":"23bd182f81ca9d273f7b0d26","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":56.289685574342705,"warning":null},{"id":"a3c1d26625581d76b6c31873","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-critpt","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":10,"warning":null},{"id":"1faacdec2388a3389c95daa0","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-critpt","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":12.8571428571429,"warning":null},{"id":"9a7bd9e9fbb331e851d3fa93","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":0.857142857142857,"warning":null},{"id":"1c0ebb8f53f204bde27fa113","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":40.43569083855566,"warning":null},{"id":"5fa0a7b1a4fd9e763ca818f6","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":13.4,"warning":null},{"id":"c3a9570cb5be5e9387e1ea1d","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":1152.15,"warning":null},{"id":"3dd2b41097d4a3d04883887e","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":1156.26,"warning":null},{"id":"9fb983cb66553aed7ceebb46","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":90.5050505050505,"warning":null},{"id":"992e678dc195695cadb3e4a8","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":88.78787878787881,"warning":null},{"id":"6541986de29df292349c312f","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":71.7171717171717,"warning":null},{"id":"92d55a9b5f2e46ce1d82dae9","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":35.2177942539388,"warning":null},{"id":"6d9d95d9ee994dd897c40a5a","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":37.5347544022243,"warning":null},{"id":"2e6c279793e70851ca3830b3","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":8.2483781278962,"warning":null},{"id":"48596082de3f4d8b27c476a4","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":71.2925170068027,"warning":null},{"id":"e2d77be9ab5eb6525e4ddbfd","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":76.4625850340136,"warning":null},{"id":"c5a1684a85f538d6c78dd51d","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":45.782312925170096,"warning":null},{"id":"d7dd2d0823569e12bcd2eaa6","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":38.3239171374765,"warning":null},{"id":"4d64a0fd824b9d504d785a94","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":21.1111111111111,"warning":null},{"id":"e78d8563438a3038e2add529","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":-10.5666666666667,"warning":null},{"id":"1838ec314ba48dfcb0b88949","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":-10.7333333333333,"warning":null},{"id":"3ddbb8446868edde4b9b6f18","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":-29.8666666666667,"warning":null},{"id":"a5eac9e0bb5497c1406b2ecc","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":41.4,"warning":null},{"id":"9a9651670f4e9fd8d28669fe","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":42.95,"warning":null},{"id":"ca20a70d87b992b6c59b809b","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":30.866666666666664,"warning":null},{"id":"26f09833a07c503224ee3196","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-scicode","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":50.8101851851852,"warning":null},{"id":"93d15bba04bad86908516f0c","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":94.15204678362569,"warning":null},{"id":"08f0e9fd5c5a6e0b115f294d","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":96.19883040935669,"warning":null},{"id":"9954df33da7c51a639dd175c","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":91.2280701754386,"warning":null},{"id":"063db2abf6786cf8adafe722","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":26.185567010309303,"warning":null},{"id":"b5ec417d4f5555a3d4af2586","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":30.103092783505197,"warning":null},{"id":"9eb01d7c86816a15077c34cd","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":41.6666666666667,"warning":null},{"id":"1aa95e16a12b0a054a48ea42","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":46.2121212121212,"warning":null},{"id":"27ec74814284b3c7c92ad1d2","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"DeepSeek V4 Pro 0424 (Non-reasoning)","configuration":"DeepSeek V4 Pro 0424 (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-non-reasoning","score":36.363636363636395,"warning":null},{"id":"597079259d0379016df7cab6","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"DeepSeek V4 Pro 0424 (High)","configuration":"DeepSeek V4 Pro 0424 (High); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424-high","score":64.7940074906367,"warning":null},{"id":"ace696d0fea831b2a18a5580","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":64.0449438202247,"warning":null},{"id":"0489348008374e06a125192c","modelId":"deepseek-v4-pro-0424","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"DeepSeek V4 Pro 0424 (Max)","configuration":"DeepSeek V4 Pro 0424 (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-pro-0424","score":14.6464646464646,"warning":null},{"id":"f479a4d953fa96de609f6dd9","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":35,"warning":null},{"id":"b022fee5f1ebaec576b7a62c","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":1420.2,"warning":null},{"id":"e61e7fdc45739d1f249f95b7","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":1117.62,"warning":null},{"id":"760c9354bee10e634d59b796","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":84,"warning":null},{"id":"4b52fbb564515244908b3cd4","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":55.3333333333333,"warning":null},{"id":"58534c171011fa8b72aa9a11","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":68.89097769674058,"warning":null},{"id":"1865ab174f5bf32d1b665aa3","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":47.80780681445035,"warning":null},{"id":"7b9e1ba11d371919e8f1f1d7","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-critpt","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":14.285714285714299,"warning":null},{"id":"3626633e9603f09739456348","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"8c4db4347a73c9f707075aca","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":12.8,"warning":null},{"id":"870209aded1973f7986bd199","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":3.8,"warning":null},{"id":"08a8b5cc057407d47d068a4c","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":1600,"warning":null},{"id":"7b44ccd50dfd8c710f298309","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":1341.23,"warning":null},{"id":"57b594f5af7f2a7da4a69481","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":1.6666666666666667,"warning":null},{"id":"3519bcdb1c5335a014180c5a","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":39.2493049119555,"warning":null},{"id":"fc170b6eec52c69c76c6f6eb","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":10.7970342910102,"warning":null},{"id":"da79582af0e47fffb158e19d","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":46.9161958568738,"warning":null},{"id":"4410183a1bd8c6cab63b7ac9","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":22.7777777777778,"warning":null},{"id":"2b249fe5602b0268a147076a","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":76.9942196531792,"warning":null},{"id":"5329ae993b0076c1567ba62e","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":65.60693641618501,"warning":null},{"id":"73dad6f716b874127cb09975","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":-5.3,"warning":null},{"id":"ae524fe73772814c961a1e1e","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":-10.1833333333333,"warning":null},{"id":"c4096419046c7cfb99747be7","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":46.400000000000006,"warning":null},{"id":"29da5896f795ad4f605fb805","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":28.449999999999996,"warning":null},{"id":"b95593ffc6d7d29e6a085c2d","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-scicode","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":51.851851851851805,"warning":null},{"id":"3a7baacbe5fcf52f98f86cbc","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":35.532407407407405,"warning":null},{"id":"5b358cb5677beba250d4f145","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":9.047619047619051,"warning":null},{"id":"fb64a3a278c7e2f20ebbf7ec","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash","score":26.7676767676768,"warning":null},{"id":"1674a710118a655765c1f376","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"DeepSeek V4.1 Flash (Non-reasoning)","configuration":"DeepSeek V4.1 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/deepseek-v4-1-flash-non-reasoning","score":5.55555555555556,"warning":null},{"id":"4075f035c16894991019b82e","modelId":"deepseek-v4.1-flash","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"DeepSeek V4.1 Flash (High)","configuration":"DeepSeek V4.1 Flash (High); Source model ID: deepseek-v4-1-flash-high; 120 tasks","score":67.5,"warning":null},{"id":"cb297d8a0490e3242252f18d","modelId":"deepseek-v4.1-flash","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"DeepSeek V4.1 Flash (Low)","configuration":"DeepSeek V4.1 Flash (Low); Source model ID: deepseek-v4-1-flash-low; 120 tasks","score":60.55555555555555,"warning":null},{"id":"b2d5a83effbb0b16ee3f6cf9","modelId":"deepseek-v4.1-flash","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"DeepSeek V4.1 Flash (Max); Source model ID: deepseek-v4-1-flash-max; 120 tasks","score":72.91666666666666,"warning":null},{"id":"c1ba08f364d613d9901146d4","modelId":"deepseek-v4.1-flash","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"DeepSeek V4.1 Flash (None)","configuration":"DeepSeek V4.1 Flash (None); Source model ID: deepseek-v4-1-flash-none; 120 tasks","score":1.25,"warning":null},{"id":"ce169f4b75dba56fd007c11a","modelId":"devstral-2512","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":482.33,"warning":null},{"id":"c027bac776af43b4cb938c11","modelId":"devstral-2512","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.6666666666667,"warning":null},{"id":"56b570282eeb2d7f9bc5720b","modelId":"devstral-2512","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.3333333333333,"warning":null},{"id":"f20b6e6193c225e4bde7c4d7","modelId":"devstral-2512","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.106312467124685,"warning":null},{"id":"4818263b0b0cdc6012971d70","modelId":"devstral-2512","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"18a919dc18d96a14e330490f","modelId":"devstral-2512","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.4,"warning":null},{"id":"0c81ceee4b3638bbcc3ad221","modelId":"devstral-2512","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":544.16,"warning":null},{"id":"2b73f3f0e2fb4a983219b412","modelId":"devstral-2512","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.48958333333333,"warning":null},{"id":"6f3c6cbea54937ac773fa184","modelId":"devstral-2512","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":59.3939393939394,"warning":null},{"id":"f67d22c0ec4ea389ca56cbff","modelId":"devstral-2512","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.5681186283595903,"warning":null},{"id":"06996081218bc20d64759001","modelId":"devstral-2512","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.0952380952381,"warning":null},{"id":"0028f41bbd3897a5e62a1019","modelId":"devstral-2512","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.7619047619048,"warning":null},{"id":"b77d44d579e2a5fb4c111b42","modelId":"devstral-2512","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.1635638297872,"warning":null},{"id":"f87d2326544d8fc4fc49d91d","modelId":"devstral-2512","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":-46.7,"warning":null},{"id":"28c7d9bfa5d2fbf77e67a274","modelId":"devstral-2512","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.816666666666666,"warning":null},{"id":"4656ead8373195f1a2848cdd","modelId":"devstral-2512","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.7546296296296,"warning":null},{"id":"8c16d1eb56b6a0b6ad2c7df6","modelId":"devstral-2512","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.8538011695906,"warning":null},{"id":"54bcb36c56b3510fb13024ba","modelId":"devstral-2512","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.5154639175258,"warning":null},{"id":"520f49635049319b43c19aaa","modelId":"devstral-2512","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.939393939393902,"warning":null},{"id":"8338dab33792ea1da2e48ada","modelId":"devstral-2512","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.337078651685403,"warning":null},{"id":"d85460656a5d267173ae81ba","modelId":"devstral-2512","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Devstral 2","configuration":"Devstral 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"4149e6a6efb13371a5919a0a","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.6666666666667,"warning":null},{"id":"16155cc9771f19d724e14d31","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"05f0bdfe993c7013fb2b28da","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":321.41,"warning":null},{"id":"1dd0d4ad446639c8b032f826","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.8686868686869,"warning":null},{"id":"3599b35b722d8252b79f6f35","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.8433734939759,"warning":null},{"id":"d29b6385e38b2fa601e91de5","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":59.4557823129252,"warning":null},{"id":"91892dd7bf1a5f08090f9d9e","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.5317919075144,"warning":null},{"id":"0b3377f33c68c96e20808b3d","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-59.2333333333333,"warning":null},{"id":"0a6b438d49c67c6b0125a1c9","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.583333333333332,"warning":null},{"id":"8e6d3333f88940e6403844c7","modelId":"diffusiongemma-26b-a4b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"DiffusionGemma 26B A4B","configuration":"DiffusionGemma 26B A4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/diffusiongemma-26b-a4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.3595505617978,"warning":null},{"id":"454c740f40aa59103fd640fb","modelId":"doubao-seed-code","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.3333333333333,"warning":null},{"id":"9dda4e6313c33f052ce51f21","modelId":"doubao-seed-code","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.6666666666667,"warning":null},{"id":"93e5d74fa9cd4b8faa10f31c","modelId":"doubao-seed-code","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"d0b2a7f440e8d5397f0978a3","modelId":"doubao-seed-code","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.75520833333334,"warning":null},{"id":"05fc8421995c99a0e7d522af","modelId":"doubao-seed-code","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.3636363636364,"warning":null},{"id":"e5af52daf1449986eaa89dff","modelId":"doubao-seed-code","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.0871177015755,"warning":null},{"id":"b83eea8de1c05302b1b64c2e","modelId":"doubao-seed-code","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.4285714285714,"warning":null},{"id":"e7d99c67e34240e8766aad72","modelId":"doubao-seed-code","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.6137566137566,"warning":null},{"id":"19ea5a48da48057701cfe1cc","modelId":"doubao-seed-code","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":85.3889627659574,"warning":null},{"id":"96e8b68fb86c6c0f5461090a","modelId":"doubao-seed-code","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.092485549133,"warning":null},{"id":"0dbac5be77e69634487925b3","modelId":"doubao-seed-code","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":-35.2,"warning":null},{"id":"111fd2d05544b06f84c1f97c","modelId":"doubao-seed-code","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.883333333333333,"warning":null},{"id":"8f16565827cdaac0168c025b","modelId":"doubao-seed-code","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.187134502923996,"warning":null},{"id":"eab63b4ac56561a7e1509869","modelId":"doubao-seed-code","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Doubao Seed Code","configuration":"Doubao Seed Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/doubao-seed-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.5151515151515,"warning":null},{"id":"ea6dc61a26cd710caaeaa276","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.3333333333333,"warning":null},{"id":"89cb005073731bd86024d136","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"b3ff20db183a764049256077","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.1111111111111,"warning":null},{"id":"03ab887691c2d04305534b99","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.33642261353105,"warning":null},{"id":"a4419233d06025238adf7fc2","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.1156462585034,"warning":null},{"id":"048a4485e7a9a4c8d5650791","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.6666666666667,"warning":null},{"id":"ad7581b9cbb13c55b285c09c","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":93.06666666666669,"warning":null},{"id":"ad7ebbddb49c0e728ba114d7","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.6097074468085,"warning":null},{"id":"d15eb698c8417578a083d5a2","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-35.1333333333333,"warning":null},{"id":"531621bce0f156b7a851eb0b","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.05,"warning":null},{"id":"396c5b6a5799009c00a077a9","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"20f35a16a687d6b454dbaa39","modelId":"ernie-4.5-300b-a47b-pt","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"ERNIE 4.5 300B A47B","configuration":"ERNIE 4.5 300B A47B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ernie-4-5-300b-a47b; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.06060606060606,"warning":null},{"id":"f99755b76c6471b984bef051","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":85,"warning":null},{"id":"c71e752f02177e2307ec0208","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.42857142857143,"warning":null},{"id":"0fc7ba284be33c52df15619c","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.5,"warning":null},{"id":"94de804551d84e436bcfab2c","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.6767676767677,"warning":null},{"id":"e7dfc21f139df43e3647f983","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.253012048192799,"warning":null},{"id":"381abd772bfd4fcc40a9ff1d","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.3605442176871,"warning":null},{"id":"0fa215c1732d7ec7afca35f1","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.16402116402111,"warning":null},{"id":"24b49036b5b6c2c71015d226","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.0285904255319,"warning":null},{"id":"47e6a1dbad4f44e3baab9363","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.5664739884393,"warning":null},{"id":"cec86b4c9567fb892efd3dde","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":-45.6166666666667,"warning":null},{"id":"637a534eabbcc15462e13a67","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.133333333333333,"warning":null},{"id":"a17190accc3a0092e345b5d3","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.9181286549707,"warning":null},{"id":"f2919bcee1f19cc027c0b8e4","modelId":"ernie-5.0-thinking-preview","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"ERNIE 5.0 Thinking Preview","configuration":"ERNIE 5.0 Thinking Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/ernie-5-0-thinking-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":25,"warning":null},{"id":"bab5b3f721c054e194163da2","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":24,"warning":null},{"id":"b597d1e3c86a13c59c590891","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":50.3333333333333,"warning":null},{"id":"d0660023c24169d0375b9331","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":0,"warning":null},{"id":"c83bd4489ef7f16d4df9e0cf","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":0,"warning":null},{"id":"74f33d01f1f6fbcdae474f72","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":0,"warning":null},{"id":"a41dda65db95525981059fa9","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":0,"warning":null},{"id":"b3b843795c22987d82505da7","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":42.4242424242424,"warning":null},{"id":"1ebdbcfcce04b951368b4f81","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":51.5151515151515,"warning":null},{"id":"8caf0e4036cb60fb443a2182","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":5.69972196478221,"warning":null},{"id":"c1b99c8edf1d5b0b83a10907","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":5.977757182576459,"warning":null},{"id":"10e9713791139c6bfc66aea4","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":25.3061224489796,"warning":null},{"id":"ffc52f3cf3a2ef4dd82fc717","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":22.9931972789116,"warning":null},{"id":"4dc04b302fcd46bc8980c038","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":29.3121693121693,"warning":null},{"id":"942dab2d051ced6ab8272528","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":51.6402116402116,"warning":null},{"id":"578f3fc0f7765bd8fde6c59d","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":50.0498670212766,"warning":null},{"id":"a589219818ac81f6b4db247b","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":58.7682845744681,"warning":null},{"id":"de9a29070b73434bde934a9d","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":-82.15,"warning":null},{"id":"91f5aaaa147fa965a95ee6e2","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":-79.8166666666667,"warning":null},{"id":"cd1008d27b1fbfac60e39554","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":5,"warning":null},{"id":"37a86f44ce022d6dafdf777f","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":6.833333333333333,"warning":null},{"id":"81337225c1cfbaecc298a109","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":20.467836257309898,"warning":null},{"id":"e07bdb18a2b42aa76c775099","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":16.374269005848,"warning":null},{"id":"19f33b7a6965d3055111ffb1","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Exaone 4.0 1.2B (Non-reasoning)","configuration":"Exaone 4.0 1.2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b","score":0,"warning":null},{"id":"f639c6955dcc41a97a992d9a","modelId":"exaone-4.0-1.2b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Exaone 4.0 1.2B (Reasoning)","configuration":"Exaone 4.0 1.2B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-1-2b-reasoning","score":0,"warning":null},{"id":"c900cfefe08cce4e82a15a9b","modelId":"exaone-4.0-32b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":39.3333333333333,"warning":null},{"id":"e146162ac9b155d4b57a4230","modelId":"exaone-4.0-32b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":80,"warning":null},{"id":"c48949f7d4d375f4aaf944fa","modelId":"exaone-4.0-32b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":0,"warning":null},{"id":"3dccf3f99ac4f647f345a44e","modelId":"exaone-4.0-32b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":0,"warning":null},{"id":"bbb70ff8ccf50d3761bd741a","modelId":"exaone-4.0-32b","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":73.46354166666667,"warning":null},{"id":"465814ec06b1b537bd6b8db7","modelId":"exaone-4.0-32b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":62.8282828282828,"warning":null},{"id":"182b7997becaf9b5dfe635a2","modelId":"exaone-4.0-32b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":73.93939393939391,"warning":null},{"id":"d3806f14df2e84d68e93bb3b","modelId":"exaone-4.0-32b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":5.004633920296571,"warning":null},{"id":"688816e2f1a961f4d3aff674","modelId":"exaone-4.0-32b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":11.445783132530101,"warning":null},{"id":"96fc68b9a4cdba13bc5714a4","modelId":"exaone-4.0-32b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":33.469387755102,"warning":null},{"id":"c2bb057ef3c15aa7c1a0099c","modelId":"exaone-4.0-32b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":36.3265306122449,"warning":null},{"id":"44e88de85e270c332ba612c8","modelId":"exaone-4.0-32b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":47.1957671957672,"warning":null},{"id":"45918216b4f57506bc18f43f","modelId":"exaone-4.0-32b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":74.70899470899471,"warning":null},{"id":"ab9a17932f0001c18c76fbf4","modelId":"exaone-4.0-32b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":93.93333333333331,"warning":null},{"id":"df985a406bd973b58bb12714","modelId":"exaone-4.0-32b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":97.6666666666667,"warning":null},{"id":"caaf789de95efc45eec9e06f","modelId":"exaone-4.0-32b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":76.8284574468085,"warning":null},{"id":"1cc2edb1a9f3b941f175cda1","modelId":"exaone-4.0-32b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":81.7569813829787,"warning":null},{"id":"28161676227107f4bf1e949f","modelId":"exaone-4.0-32b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":-62.7666666666667,"warning":null},{"id":"d18d1d42d9c3d891e66ccc05","modelId":"exaone-4.0-32b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":-60.0666666666667,"warning":null},{"id":"bd5ae655caef8856694260ac","modelId":"exaone-4.0-32b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":10.616666666666667,"warning":null},{"id":"1eaec2770c32a1d977fe2e9f","modelId":"exaone-4.0-32b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":13.866666666666665,"warning":null},{"id":"f0e22773a86c483105fb9aaf","modelId":"exaone-4.0-32b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":4.09356725146199,"warning":null},{"id":"0c56e3b4c1bd36102a0d428b","modelId":"exaone-4.0-32b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":17.2514619883041,"warning":null},{"id":"27cf0b2a4eed2df72f1b5425","modelId":"exaone-4.0-32b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"EXAONE 4.0 32B (Non-reasoning)","configuration":"EXAONE 4.0 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b","score":1.51515151515152,"warning":null},{"id":"d58e0132dbc6ea2836492ddf","modelId":"exaone-4.0-32b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"EXAONE 4.0 32B (Reasoning)","configuration":"EXAONE 4.0 32B (Reasoning); Model source: https://artificialanalysis.ai/models/exaone-4-0-32b-reasoning","score":3.78787878787879,"warning":null},{"id":"f1324bf4938b9aacc8f22d07","modelId":"exaone-4.5-33b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.6666666666667,"warning":null},{"id":"a3c18d79312713938639cd67","modelId":"exaone-4.5-33b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"7e3f20402edb01db2b395439","modelId":"exaone-4.5-33b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":467.49,"warning":null},{"id":"7b3ea9943861bd5124920a9f","modelId":"exaone-4.5-33b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.39393939393939,"warning":null},{"id":"29e2d8ead248cdf77d8c3ad4","modelId":"exaone-4.5-33b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.8822984244671,"warning":null},{"id":"0aaa45ce8b555887ed48000f","modelId":"exaone-4.5-33b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":57.959183673469404,"warning":null},{"id":"55ecc32abb098053e1990908","modelId":"exaone-4.5-33b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.3410404624277,"warning":null},{"id":"331c0bc96cdcb1aaa53a4c89","modelId":"exaone-4.5-33b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-50.7166666666667,"warning":null},{"id":"efb0e6c109bcc9124fb56389","modelId":"exaone-4.5-33b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.233333333333334,"warning":null},{"id":"98c051d405ab71f78cf3ee7e","modelId":"exaone-4.5-33b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.07017543859651,"warning":null},{"id":"736511a5429ed0e64b138278","modelId":"exaone-4.5-33b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.4545454545455,"warning":null},{"id":"25f82c681fcbed1f3a181de8","modelId":"exaone-4.5-33b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"EXAONE 4.5 33B (Reasoning)","configuration":"EXAONE 4.5 33B (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/exaone-4-5-33b; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.3483146067416,"warning":null},{"id":"778a476fac3a54bc26ad1bdc","modelId":"gemini-2.5-flash","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":60.33333333333331,"warning":null},{"id":"25e8b81f8b05ec46c5e80257","modelId":"gemini-2.5-flash","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":73.3333333333333,"warning":null},{"id":"c843963587d72d6e7fa6ab2b","modelId":"gemini-2.5-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":49.9,"warning":null},{"id":"d1e35c9dfd5a9177d6520490","modelId":"gemini-2.5-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":65.3333333333333,"warning":null},{"id":"8fbe3f48f7d31ed271a50ddf","modelId":"gemini-2.5-flash","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":1.42857142857143,"warning":null},{"id":"03903b802c05359369266f36","modelId":"gemini-2.5-flash","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":1.1428571428571401,"warning":null},{"id":"ae60d7a4d3b3a85a296b38ab","modelId":"gemini-2.5-flash","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":68.2828282828283,"warning":null},{"id":"7a919ff134ab8dd622151ef8","modelId":"gemini-2.5-flash","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":78.989898989899,"warning":null},{"id":"0508741ce4e4c770a963f3a4","modelId":"gemini-2.5-flash","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":4.72659870250232,"warning":null},{"id":"29c12307a2fa27b9adfab4e4","modelId":"gemini-2.5-flash","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":12.1408711770158,"warning":null},{"id":"ad52dd4994d43cfb7388537b","modelId":"gemini-2.5-flash","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":38.9795918367347,"warning":null},{"id":"615f11f6104fbdbc6397c19a","modelId":"gemini-2.5-flash","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":50.2721088435374,"warning":null},{"id":"f2274f10c01fd49af34cc8c7","modelId":"gemini-2.5-flash","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":49.5238095238095,"warning":null},{"id":"a32eda887ff803501ac260c2","modelId":"gemini-2.5-flash","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":69.5238095238095,"warning":null},{"id":"a2a790a25cbcf597773c9220","modelId":"gemini-2.5-flash","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":93.2,"warning":null},{"id":"b3ffd498b7959470e367bd26","modelId":"gemini-2.5-flash","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":98.1333333333333,"warning":null},{"id":"8e548cc9df3852affb6827c7","modelId":"gemini-2.5-flash","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":8.33333333333333,"warning":null},{"id":"e9a8955bfe6f14e6cc47c6be","modelId":"gemini-2.5-flash","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":8.888888888888891,"warning":null},{"id":"5513fb39e65010cbd84258eb","modelId":"gemini-2.5-flash","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":80.9175531914894,"warning":null},{"id":"aa36fae3bef0b68268249bfe","modelId":"gemini-2.5-flash","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":83.1532579787234,"warning":null},{"id":"4871389cdfa8311f33acc44c","modelId":"gemini-2.5-flash","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":65.4913294797688,"warning":null},{"id":"8810b2fe580ca34e86a3b029","modelId":"gemini-2.5-flash","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":69.0751445086705,"warning":null},{"id":"a6ea19c704d0cd6d45804ad9","modelId":"gemini-2.5-flash","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":-42.5666666666667,"warning":null},{"id":"94436d1283ceade19a098e93","modelId":"gemini-2.5-flash","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":-29.8,"warning":null},{"id":"89031d6f4b0b9da5cd80cf55","modelId":"gemini-2.5-flash","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":26.13333333333333,"warning":null},{"id":"be837d02ebf79658b1d356cf","modelId":"gemini-2.5-flash","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":25.95,"warning":null},{"id":"b82146057c975a009a40f506","modelId":"gemini-2.5-flash","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":14.9122807017544,"warning":null},{"id":"b6896924c93716199a8e8b23","modelId":"gemini-2.5-flash","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":31.5789473684211,"warning":null},{"id":"0b1b7e45360ba1982d470496","modelId":"gemini-2.5-flash","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash (Non-reasoning)","configuration":"Gemini 2.5 Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash","score":12.1212121212121,"warning":null},{"id":"a6d86a700f4535aa3b842c46","modelId":"gemini-2.5-flash","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemini 2.5 Flash (Reasoning)","configuration":"Gemini 2.5 Flash (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-reasoning","score":13.6363636363636,"warning":null},{"id":"4b6d985260431c716d5f32f1","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":35.3333333333333,"warning":null},{"id":"a433d84489e1746d7f57ce6b","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":53.3333333333333,"warning":null},{"id":"7c4270b425ce13c10d94c734","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":32,"warning":null},{"id":"d69a72ef36f478ea09b68875","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":55.6666666666667,"warning":null},{"id":"d4da0bef5d2711a9743bf008","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":0,"warning":null},{"id":"434ad978699d9f67751986dc","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":0,"warning":null},{"id":"32549bdfff973940b8fe54d1","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":47.3737373737374,"warning":null},{"id":"25128a1d8bb8e5c053dcad35","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":62.5252525252525,"warning":null},{"id":"838fbf397ffda157ad949a61","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":3.66079703429101,"warning":null},{"id":"60f98d8bd60241b8c86ec5d7","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":6.811862835959221,"warning":null},{"id":"6e11e8fedfcf94c7848f35e4","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":31.496598639455797,"warning":null},{"id":"2d5d302d8a59f5716d532cb4","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":49.8639455782313,"warning":null},{"id":"208bce24d8fb34ba60bd10e9","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":40,"warning":null},{"id":"9f124c6e37fc898fcec9d2aa","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":59.2592592592593,"warning":null},{"id":"b5463a06d82c58c7e8fba350","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":92.60000000000001,"warning":null},{"id":"bc2be91ff8211d91bea25150","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":96.8666666666667,"warning":null},{"id":"3e51c2efec3e1e4facb453c9","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":2.2222222222222197,"warning":null},{"id":"f8b3bc91b10e83be18e8677b","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":72.4401595744681,"warning":null},{"id":"6a0ea081f38de34cf0138ec6","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":75.91422872340429,"warning":null},{"id":"b619e01f7daa0e5ea4cdeda1","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":53.9884393063584,"warning":null},{"id":"4989b8aaa192e67ee9b036ce","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":58.2080924855491,"warning":null},{"id":"bd0e18878398a2cba1e2c630","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":-58.7833333333333,"warning":null},{"id":"54fcfd9dba64059091858233","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":-45.6166666666667,"warning":null},{"id":"67df11bafe9809cfd9e3f3eb","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":15.433333333333332,"warning":null},{"id":"86b22a6d3c908a4896b22e32","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":17.9,"warning":null},{"id":"333ef0ed6319c3f365aae494","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":19.0058479532164,"warning":null},{"id":"402b1b95380a385fcfc42d37","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":18.4210526315789,"warning":null},{"id":"c90f54c21e1329dfe593d000","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Non-reasoning)","configuration":"Gemini 2.5 Flash-Lite (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite","score":2.27272727272727,"warning":null},{"id":"eb222f50e00f52ee760583e4","modelId":"gemini-2.5-flash-lite","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemini 2.5 Flash-Lite (Reasoning)","configuration":"Gemini 2.5 Flash-Lite (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-2-5-flash-lite-reasoning","score":4.545454545454549,"warning":null},{"id":"12b95021d8514d0171d35190","modelId":"gemini-2.5-pro","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":301.88,"warning":null},{"id":"487059beacefde1e01685e8d","modelId":"gemini-2.5-pro","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":87.6666666666667,"warning":null},{"id":"5604a1b3b4ab8c68dc8d01bc","modelId":"gemini-2.5-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":69,"warning":null},{"id":"fe94c818a9ad0ca7d39e39ef","modelId":"gemini-2.5-pro","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.174946419340874,"warning":null},{"id":"9aec0db3dacd7fcf771012a9","modelId":"gemini-2.5-pro","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.57142857142857,"warning":null},{"id":"2a2fff8b12d3d37901295fdf","modelId":"gemini-2.5-pro","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.2,"warning":null},{"id":"f2164d21dbe544336569cf92","modelId":"gemini-2.5-pro","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":455.95,"warning":null},{"id":"defe800b4fa5856397240d2e","modelId":"gemini-2.5-pro","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":90.3125,"warning":null},{"id":"0780118e8e66b234eca4b2a4","modelId":"gemini-2.5-pro","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.4444444444444,"warning":null},{"id":"cb4e80a7918d6b9057e3deed","modelId":"gemini-2.5-pro","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.5208526413346,"warning":null},{"id":"f077b068853bedaf080c341f","modelId":"gemini-2.5-pro","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.707482993197296,"warning":null},{"id":"b5a906c034fd39fe365ed813","modelId":"gemini-2.5-pro","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.1058201058201,"warning":null},{"id":"69605a228871fa6d6a1fe38d","modelId":"gemini-2.5-pro","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":96.7333333333333,"warning":null},{"id":"effe61acb1f2ecd3194c56be","modelId":"gemini-2.5-pro","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.186835106383,"warning":null},{"id":"8f2d5e3be696b0d89edeb2fc","modelId":"gemini-2.5-pro","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.9132947976879,"warning":null},{"id":"fed8f4ab77fcd5d31da67c95","modelId":"gemini-2.5-pro","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":-16.35,"warning":null},{"id":"cf0ed4c7679f5831c0151b93","modelId":"gemini-2.5-pro","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.050000000000004,"warning":null},{"id":"b346e2f4d8c39e5b6a8796ce","modelId":"gemini-2.5-pro","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.296296296296305,"warning":null},{"id":"5336414aba2be78f3187de97","modelId":"gemini-2.5-pro","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.093567251462005,"warning":null},{"id":"80b93729bec1b12b13fec2bd","modelId":"gemini-2.5-pro","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.69072164948454,"warning":null},{"id":"fe7921ca6643b00dc3e8c896","modelId":"gemini-2.5-pro","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.5151515151515,"warning":null},{"id":"0f6648f08f82d8663f02b246","modelId":"gemini-2.5-pro","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.4644194756554,"warning":null},{"id":"9861b3363626336b8e1278dd","modelId":"gemini-2.5-pro","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Gemini 2.5 Pro","configuration":"Gemini 2.5 Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"53fc9ec1dee0accada2433b0","modelId":"gemini-2.5-pro","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Gemini 2.5 Pro (Thinking 8K)","configuration":"Gemini 2.5 Pro (Thinking 8K); Source model ID: gemini-2-5-pro-2025-06-17-thinking-8k; 120 tasks","score":2.92,"warning":null},{"id":"6cc85ba2b198bd02c85a961a","modelId":"gemini-2.5-pro","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Gemini 2.5 Pro (Thinking 16K)","configuration":"Gemini 2.5 Pro (Thinking 16K); Source model ID: gemini-2-5-pro-2025-06-17-thinking-16k; 120 tasks","score":4.03,"warning":null},{"id":"85789420efbd10e4bf8e8a0f","modelId":"gemini-2.5-pro","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Gemini 2.5 Pro (Thinking 1K)","configuration":"Gemini 2.5 Pro (Thinking 1K); Source model ID: gemini-2-5-pro-2025-06-17-thinking-1k; 120 tasks","score":0,"warning":null},{"id":"e1746e4a5f84cad00df32d6b","modelId":"gemini-2.5-pro","benchmarkId":"arc-v2semiprivate","level":"thinking","sourceLabel":"Gemini 2.5 Pro (Thinking 32K)","configuration":"Gemini 2.5 Pro (Thinking 32K); Source model ID: gemini-2-5-pro-2025-06-17-thinking-32k; 120 tasks","score":4.859999999999999,"warning":null},{"id":"076c87b82049cfa2ea7b9df0","modelId":"gemini-3-flash-preview","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":55.6666666666667,"warning":null},{"id":"162b72dff2b62a478a4e3d38","modelId":"gemini-3-flash-preview","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":97,"warning":null},{"id":"b3a845d5ca338864a0c68e84","modelId":"gemini-3-flash-preview","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":27.728613569321496,"warning":null},{"id":"5e4484b725f885d1ff5e913f","modelId":"gemini-3-flash-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":55.3333333333333,"warning":null},{"id":"681dcd67742f3fabcb4e8443","modelId":"gemini-3-flash-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":78,"warning":null},{"id":"b771ca3369e45caa49d4b3fe","modelId":"gemini-3-flash-preview","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":1.42857142857143,"warning":null},{"id":"862b3fc84116cb2abfb29a11","modelId":"gemini-3-flash-preview","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":8.57142857142857,"warning":null},{"id":"766bfc6c0bb47794ea896d7c","modelId":"gemini-3-flash-preview","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":92.74999999999999,"warning":null},{"id":"f0939c33b02de61e8d2569c0","modelId":"gemini-3-flash-preview","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":91.4375,"warning":null},{"id":"d45eab2a552c001f2dbdb704","modelId":"gemini-3-flash-preview","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":81.2121212121212,"warning":null},{"id":"5177df7f65f8989816e07078","modelId":"gemini-3-flash-preview","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":89.79797979797979,"warning":null},{"id":"5a0c8bdb65808c38cc8c6ba8","modelId":"gemini-3-flash-preview","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":14.967562557924,"warning":null},{"id":"f524fa4818eae0de4de4b842","modelId":"gemini-3-flash-preview","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":36.5616311399444,"warning":null},{"id":"8d1a6871f955048317879ba7","modelId":"gemini-3-flash-preview","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":55.1020408163265,"warning":null},{"id":"fe637f0f65778514caf29868","modelId":"gemini-3-flash-preview","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":77.9591836734694,"warning":null},{"id":"3877c7c555480e56e2553353","modelId":"gemini-3-flash-preview","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":79.6825396825397,"warning":null},{"id":"afb2c3230511cde285428634","modelId":"gemini-3-flash-preview","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":90.7936507936508,"warning":null},{"id":"65ec7e6d31d6f6ce02771c4b","modelId":"gemini-3-flash-preview","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":11.6666666666667,"warning":null},{"id":"91700195796ca55967f40f20","modelId":"gemini-3-flash-preview","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":11.1111111111111,"warning":null},{"id":"7161c5ae63b41bca96c70974","modelId":"gemini-3-flash-preview","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":88.2230718085106,"warning":null},{"id":"a699cac3b6da2ec95d940e1a","modelId":"gemini-3-flash-preview","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":89.0292553191489,"warning":null},{"id":"1db4c190ae70acb2a1c1a232","modelId":"gemini-3-flash-preview","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":78.5549132947977,"warning":null},{"id":"57c50354c54c32290ddfc4c1","modelId":"gemini-3-flash-preview","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":79.9421965317919,"warning":null},{"id":"e31d78ed7068e350eae0e017","modelId":"gemini-3-flash-preview","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":-4.31666666666667,"warning":null},{"id":"31aec0c830003796ddb584fb","modelId":"gemini-3-flash-preview","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":10.1333333333333,"warning":null},{"id":"9fae7f3a431a07e9ef03a649","modelId":"gemini-3-flash-preview","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":45.78333333333333,"warning":null},{"id":"6ca294472ce7424a59037593","modelId":"gemini-3-flash-preview","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":53.43333333333333,"warning":null},{"id":"090d329b550ff7c39b218328","modelId":"gemini-3-flash-preview","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":43.2748538011696,"warning":null},{"id":"cb0e8ee7331851f023b16462","modelId":"gemini-3-flash-preview","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":80.4093567251462,"warning":null},{"id":"216792d88d988a2e05a41153","modelId":"gemini-3-flash-preview","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":20.8247422680412,"warning":null},{"id":"66bd4898aef466d04784881c","modelId":"gemini-3-flash-preview","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemini 3 Flash Preview (Non-reasoning)","configuration":"Gemini 3 Flash Preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash","score":31.818181818181802,"warning":null},{"id":"f07bd2d137eff751c94e16e4","modelId":"gemini-3-flash-preview","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemini 3 Flash Preview (Reasoning)","configuration":"Gemini 3 Flash Preview (Reasoning); Model source: https://artificialanalysis.ai/models/gemini-3-flash-reasoning","score":38.636363636363605,"warning":null},{"id":"da304a0f8d4e4dc282b7a95b","modelId":"gemini-3-flash-preview","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Gemini 3 Flash Preview (High)","configuration":"Gemini 3 Flash Preview (High); Source model ID: gemini-3-flash-preview-thinking-high; 120 tasks","score":33.61,"warning":null},{"id":"8f72a698e068a7bc179feb19","modelId":"gemini-3-flash-preview","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Gemini 3 Flash Preview (Low)","configuration":"Gemini 3 Flash Preview (Low); Source model ID: gemini-3-flash-preview-thinking-low; 120 tasks","score":1.25,"warning":null},{"id":"c3906ea3a7c1254fb6ccba8d","modelId":"gemini-3-flash-preview","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Gemini 3 Flash Preview (Medium)","configuration":"Gemini 3 Flash Preview (Medium); Source model ID: gemini-3-flash-preview-thinking-medium; 120 tasks","score":12.78,"warning":null},{"id":"f433631bd967cd374a5a92a9","modelId":"gemini-3-flash-preview","benchmarkId":"arc-v2semiprivate","level":"minimal","sourceLabel":"Gemini 3 Flash Preview (Minimal)","configuration":"Gemini 3 Flash Preview (Minimal); Source model ID: gemini-3-flash-preview-thinking-minimal; 120 tasks","score":3.3300000000000005,"warning":null},{"id":"9ed78211ce574b898ce5c37e","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.75,"warning":null},{"id":"d57e017701a3a92259f15e32","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":207.25,"warning":null},{"id":"95738bc56bfab2218391b063","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.168141592920401,"warning":null},{"id":"e51ce9e31d112599849140e1","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.3333333333333,"warning":null},{"id":"542af94496db9fff0c50c33b","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.796182033119545,"warning":null},{"id":"1f43c952940a4b6d26d536be","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.1428571428571401,"warning":null},{"id":"8416976c2cf8e61964b2a869","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.02148612354521,"warning":null},{"id":"866f484125051166bc2e4453","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.8,"warning":null},{"id":"36f7044bc017e2b899790517","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":430.83,"warning":null},{"id":"f39228d0894775a799d9e673","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.2222222222222,"warning":null},{"id":"a500a1b31b9c03cf2da51aa8","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.191844300278,"warning":null},{"id":"96ee84a5b928f3599cf5d764","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.2108843537415,"warning":null},{"id":"4aae5d6df970c338630860a9","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.11111111111111,"warning":null},{"id":"334d043fe09ccc4d1fdbf484","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.5491329479769,"warning":null},{"id":"897f612ed5343edc51d5b099","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":-16.4166666666667,"warning":null},{"id":"4bc71b7684588e825fa67080","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.266666666666666,"warning":null},{"id":"097b9973b90993f5f5e7d1df","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.4027777777778,"warning":null},{"id":"bf184358925df7a4fbedf4bc","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.2865497076023,"warning":null},{"id":"2684a58ff4342575f23cbc46","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.69072164948454,"warning":null},{"id":"29f5bb04ad1bd05a7465f796","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.2424242424242,"warning":null},{"id":"37c6a530f1ded7b9888ea2c3","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.0861423220974,"warning":null},{"id":"a09a2ae31ee117af6ce3dddb","modelId":"gemini-3.1-flash-lite","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Gemini 3.1 Flash-Lite","configuration":"Gemini 3.1 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-flash-lite-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.505050505050505,"warning":null},{"id":"7c7f774514f7fa2297208e0f","modelId":"gemini-3.1-pro","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.25,"warning":null},{"id":"742bf16f1d27daca444b2e71","modelId":"gemini-3.1-pro","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":456.5,"warning":null},{"id":"37687afd20028c6089e77920","modelId":"gemini-3.1-pro","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.0058997050148,"warning":null},{"id":"5f4e475d1f8d57db24bcb7a6","modelId":"gemini-3.1-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":82,"warning":null},{"id":"db8f74e2e14e95ffdcaeb92d","modelId":"gemini-3.1-pro","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.427718398098364,"warning":null},{"id":"fa3f112c9b6e79acc46b504f","modelId":"gemini-3.1-pro","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.714285714285698,"warning":null},{"id":"93374c1590601b233140cd4a","modelId":"gemini-3.1-pro","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.166517457475386,"warning":null},{"id":"8063177a2a63d1be8648e2f8","modelId":"gemini-3.1-pro","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.8,"warning":null},{"id":"a4d6100a1d6b96708fb4ae5d","modelId":"gemini-3.1-pro","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":791.8,"warning":null},{"id":"39bf8cf4de7d4ae5d9295e26","modelId":"gemini-3.1-pro","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":93.16666666666666,"warning":null},{"id":"a3fd1ba1d2b182c00683d93b","modelId":"gemini-3.1-pro","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":94.1414141414141,"warning":null},{"id":"57fa9da1ef479f1e406f93e3","modelId":"gemini-3.1-pro","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"b3631e2d668ca3ff2ebfe823","modelId":"gemini-3.1-pro","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.0342910101946,"warning":null},{"id":"9aa04e6ff77c61f9b93c9290","modelId":"gemini-3.1-pro","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.1428571428571,"warning":null},{"id":"5c7f8cd7187c61c11375f19b","modelId":"gemini-3.1-pro","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.3309120258273,"warning":null},{"id":"208c2682d2e878ad0bfc59b4","modelId":"gemini-3.1-pro","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.5555555555556,"warning":null},{"id":"45e4f876ba3f9fcf04c8c5fa","modelId":"gemini-3.1-pro","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.4277456647399,"warning":null},{"id":"6a904d86ad7c12c3aac7713b","modelId":"gemini-3.1-pro","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.8833333333333,"warning":null},{"id":"24d00d3350012cb696297151","modelId":"gemini-3.1-pro","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.85,"warning":null},{"id":"72fd11da3d45e05f88f9fba5","modelId":"gemini-3.1-pro","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.6805555555556,"warning":null},{"id":"1d824de33344d5f7c900a32c","modelId":"gemini-3.1-pro","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":95.6140350877193,"warning":null},{"id":"6d20dfb2c84bb4190f7edf29","modelId":"gemini-3.1-pro","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.4432989690722,"warning":null},{"id":"f78b4be38ed10e8ddcd80664","modelId":"gemini-3.1-pro","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.787878787878796,"warning":null},{"id":"154ebcd0575e2428b67b2398","modelId":"gemini-3.1-pro","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.78277153558051,"warning":null},{"id":"fd2e6ea2003857920a1bdfcf","modelId":"gemini-3.1-pro","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-1-pro-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.04040404040404,"warning":null},{"id":"db901e8199046d5c8d479ba0","modelId":"gemini-3.1-pro","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gemini-3-1-pro-preview","configuration":"gemini-3-1-pro-preview; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":11.7,"warning":null},{"id":"8f6dd87d3bec8be094065e59","modelId":"gemini-3.1-pro","benchmarkId":"swepro-minisweagentstarredrows","level":"thinking","sourceLabel":"gemini-3.1-pro (thinking)*","configuration":"gemini-3.1-pro (thinking)*; 731 tasks; source date 2026-04-08","score":46.1,"warning":null},{"id":"09a429372313877fb69d9077","modelId":"gemini-3.1-pro","benchmarkId":"terminal-geminicli","level":"high","sourceLabel":"Gemini 3.1 Pro","configuration":"Gemini 3.1 Pro; Reported effort: high; 445 trials; recorded 2026-02-19","score":65.84,"warning":null},{"id":"a01dc047adae16bf8c7acb63","modelId":"gemini-3.1-pro","benchmarkId":"terminal-terminus2","level":"high","sourceLabel":"Gemini 3.1 Pro","configuration":"Gemini 3.1 Pro; Reported effort: high; 445 trials; recorded 2026-02-19","score":65.62,"warning":null},{"id":"f1a0b7149c8e59cb6d290180","modelId":"gemini-3.5-flash","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":45,"warning":null},{"id":"7aa58de7bae00b57d80a75ce","modelId":"gemini-3.5-flash","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":874.32,"warning":null},{"id":"8e3287c473e31f1aa692b32d","modelId":"gemini-3.5-flash","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":876.82,"warning":null},{"id":"49103875817d61d9002ffe88","modelId":"gemini-3.5-flash","benchmarkId":"aa-apex-agents-aa","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":47.0501474926254,"warning":null},{"id":"f6680cb450982d96a116e675","modelId":"gemini-3.5-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":73.3333333333333,"warning":null},{"id":"6b61c20e2441b52b1eeda6c2","modelId":"gemini-3.5-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":74.3333333333333,"warning":null},{"id":"18382cba8ad41ba0fcd02f09","modelId":"gemini-3.5-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":61.33333333333329,"warning":null},{"id":"0043777f9ac5f56c6f8810d7","modelId":"gemini-3.5-flash","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":42.07973565652745,"warning":null},{"id":"4c3cf9e57d553a8cf6933ff7","modelId":"gemini-3.5-flash","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":13.1428571428571,"warning":null},{"id":"659ec9b0e73e69c4bad930e7","modelId":"gemini-3.5-flash","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":10.8571428571429,"warning":null},{"id":"f677a6cc7f2d3051b0fed1ab","modelId":"gemini-3.5-flash","benchmarkId":"aa-critpt","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":1.42857142857143,"warning":null},{"id":"8926adbd496f3eee8e76881d","modelId":"gemini-3.5-flash","benchmarkId":"aa-enterprise-ops-gym-aa","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":50.07460459564309,"warning":null},{"id":"d4fcaf893082b12bca1a8314","modelId":"gemini-3.5-flash","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":19.8,"warning":null},{"id":"82f400749d1cf6160fb91818","modelId":"gemini-3.5-flash","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":1202.58,"warning":null},{"id":"4ef8ec2c58e784a10fd3dbac","modelId":"gemini-3.5-flash","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":92.2222222222222,"warning":null},{"id":"b8f276e9b26fa04236baa7c9","modelId":"gemini-3.5-flash","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":92.12121212121211,"warning":null},{"id":"c85afda5b7cf54f858d9dc01","modelId":"gemini-3.5-flash","benchmarkId":"aa-gpqa-diamond","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":82.8282828282828,"warning":null},{"id":"3ed2eeaa0561782180d86dc0","modelId":"gemini-3.5-flash","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":42.678405931418,"warning":null},{"id":"f9dd86ac1890669d8fc965b0","modelId":"gemini-3.5-flash","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":41.3345690454124,"warning":null},{"id":"0c26eadda993beb0a60a75c5","modelId":"gemini-3.5-flash","benchmarkId":"aa-humanitys-last-exam","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":24.0963855421687,"warning":null},{"id":"2f271a885905a08cad21f261","modelId":"gemini-3.5-flash","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":76.32653061224491,"warning":null},{"id":"38ff4299df5d2d58795c37de","modelId":"gemini-3.5-flash","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":74.5578231292517,"warning":null},{"id":"e82f2ee0575fd64541be5657","modelId":"gemini-3.5-flash","benchmarkId":"aa-ifbench","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":47.2789115646259,"warning":null},{"id":"26ced063b3c0700c84e75fe6","modelId":"gemini-3.5-flash","benchmarkId":"aa-itbench-aa","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":40.3483992467043,"warning":null},{"id":"ee81d3d329c92527b405ec62","modelId":"gemini-3.5-flash","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":18.3333333333333,"warning":null},{"id":"db67b95544bc8c08234ff699","modelId":"gemini-3.5-flash","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":84.2774566473988,"warning":null},{"id":"3c68e6a4b646856c33b07d36","modelId":"gemini-3.5-flash","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":83.8728323699422,"warning":null},{"id":"dc830fdbd7ea5aecae7ec143","modelId":"gemini-3.5-flash","benchmarkId":"aa-mmmu-pro","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":80.1156069364162,"warning":null},{"id":"d5a0fc830fec173a656f2b27","modelId":"gemini-3.5-flash","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":21.1833333333333,"warning":null},{"id":"f2e6667b45d444dc018f8257","modelId":"gemini-3.5-flash","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":20.8166666666667,"warning":null},{"id":"cacd42cf1c650df08447f0cc","modelId":"gemini-3.5-flash","benchmarkId":"aa-omniscience","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":0.666666666666667,"warning":null},{"id":"45186d23f0dbf63b0ed2098d","modelId":"gemini-3.5-flash","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":51.4,"warning":null},{"id":"93e5a5b0f32eb190c9e72fc0","modelId":"gemini-3.5-flash","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":51.05,"warning":null},{"id":"0706386e9f89892ee07c08e5","modelId":"gemini-3.5-flash","benchmarkId":"aa-omniscience-accuracy","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":43.016666666666666,"warning":null},{"id":"8512428c087d4ab5e53a51cf","modelId":"gemini-3.5-flash","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":53.9351851851852,"warning":null},{"id":"4013a7b16f5271363e7943cb","modelId":"gemini-3.5-flash","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":95.3216374269006,"warning":null},{"id":"3d016cb2392999fd837c5a19","modelId":"gemini-3.5-flash","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":95.6140350877193,"warning":null},{"id":"4fd4845c687fa5057dfe1805","modelId":"gemini-3.5-flash","benchmarkId":"aa-tau2-bench","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":58.7719298245614,"warning":null},{"id":"52ebca0bd16f48fe28c87869","modelId":"gemini-3.5-flash","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":32.1649484536082,"warning":null},{"id":"2974f703c94be05603bc5edf","modelId":"gemini-3.5-flash","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":40.9090909090909,"warning":null},{"id":"e3819166dd2bc4c0989fc91e","modelId":"gemini-3.5-flash","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"Gemini 3.5 Flash (Medium)","configuration":"Gemini 3.5 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-medium","score":39.3939393939394,"warning":null},{"id":"809ea178ba06ea5f19a38095","modelId":"gemini-3.5-flash","benchmarkId":"aa-terminalbench-hard","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-minimal","score":46.2121212121212,"warning":null},{"id":"c34df879d252e8d7047bc1e0","modelId":"gemini-3.5-flash","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":78.65168539325839,"warning":null},{"id":"ec5bc8a918512d8932edcc1b","modelId":"gemini-3.5-flash","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-5-flash","score":6.56565656565657,"warning":null},{"id":"fb38b0f3dad2c23ea05267b5","modelId":"gemini-3.5-flash","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Gemini 3.5 Flash (High)","configuration":"Gemini 3.5 Flash (High); Source model ID: gemini-3-5-flash-high; 120 tasks","score":72.08,"warning":null},{"id":"0e2bb982e97ed49d227de10d","modelId":"gemini-3.5-flash","benchmarkId":"arc-v2semiprivate","level":"minimal","sourceLabel":"Gemini 3.5 Flash (Minimal)","configuration":"Gemini 3.5 Flash (Minimal); Source model ID: gemini-3-5-flash-minimal; 120 tasks","score":8.89,"warning":null},{"id":"d995aaf1dca8340a4a693d8a","modelId":"gemini-3.5-flash","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gemini-3-5-flash","configuration":"gemini-3-5-flash; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":36.1,"warning":null},{"id":"056acd83e1017f8328f99918","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":650.26,"warning":null},{"id":"aa3e7986ac74b1d41c049659","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":76,"warning":null},{"id":"14a902d4f36d2bebe6ff22a0","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.011270511840955,"warning":null},{"id":"2e8ef8685e32a69af902f783","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"93f57b1a7efa63a3d40f7270","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.345568487018795,"warning":null},{"id":"ae13f406495e677556b5a0c3","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.600000000000001,"warning":null},{"id":"6bafee414b326ac2fd534566","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":983.28,"warning":null},{"id":"bf28d2d6dffc1a351c70a760","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.8383838383838,"warning":null},{"id":"50ac9bba9acf7056e2d17474","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"a1cfe34cd9aecb7162043dcc","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.813716404077798,"warning":null},{"id":"fc74bcb7728d73322b4c984c","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.2222222222222205,"warning":null},{"id":"fcc617a3338c6253fdf4353c","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.0173410404624,"warning":null},{"id":"2996b27e98c9a488eb6dd898","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.23333333333333,"warning":null},{"id":"199cbf5b9764d44dc0a19484","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.46666666666667,"warning":null},{"id":"26f351bc48bbc25a81380a4b","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.3194444444444,"warning":null},{"id":"ae5702c7b4374ba438f5ea3d","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.5257731958763,"warning":null},{"id":"4133479f872bb7865947541b","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-terminal-bench-science","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f3e410a3346994f7eea97dc6","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.558052434456904,"warning":null},{"id":"e253ea51b6a540f66de5395f","modelId":"gemini-3.5-flash-lite","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Gemini 3.5 Flash-Lite","configuration":"Gemini 3.5 Flash-Lite; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/gemini-3-5-flash-lite; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.01010101010101,"warning":null},{"id":"8b0081bbd7294e66bbccf771","modelId":"gemini-3.5-flash-lite","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Gemini 3.5 Flash-Lite (High)","configuration":"Gemini 3.5 Flash-Lite (High); Source model ID: gemini-3-5-flash-lite-high; 120 tasks","score":10.277777777777779,"warning":null},{"id":"673c903462f2e3ace1f8a3a5","modelId":"gemini-3.5-flash-lite","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Gemini 3.5 Flash-Lite (Low)","configuration":"Gemini 3.5 Flash-Lite (Low); Source model ID: gemini-3-5-flash-lite-low; 120 tasks","score":1.5277777777777777,"warning":null},{"id":"38942fa126421e15bf1592ee","modelId":"gemini-3.5-flash-lite","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Gemini 3.5 Flash-Lite (Medium)","configuration":"Gemini 3.5 Flash-Lite (Medium); Source model ID: gemini-3-5-flash-lite-medium; 120 tasks","score":5.277777777777778,"warning":null},{"id":"f591fd70b9b02e903ae88a25","modelId":"gemini-3.5-flash-lite","benchmarkId":"arc-v2semiprivate","level":"minimal","sourceLabel":"Gemini 3.5 Flash-Lite (Minimal)","configuration":"Gemini 3.5 Flash-Lite (Minimal); Source model ID: gemini-3-5-flash-lite-minimal; 120 tasks","score":0.8333333333333334,"warning":null},{"id":"92ca07ec8fbd91ad5b8d40fc","modelId":"gemini-3.6-flash","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":953.92,"warning":null},{"id":"f6ffd9abea94086795b1b021","modelId":"gemini-3.6-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":80,"warning":null},{"id":"f7feb1b4c5dcc54ceeb081e5","modelId":"gemini-3.6-flash","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":53.01669491509355,"warning":null},{"id":"e5b3726bd56bde48a05ea653","modelId":"gemini-3.6-flash","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":10.5714285714286,"warning":null},{"id":"1b780fdbb3ba1783df7f48fd","modelId":"gemini-3.6-flash","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":17.4,"warning":null},{"id":"3122c70b465bb5750ffb649d","modelId":"gemini-3.6-flash","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":1284.34,"warning":null},{"id":"853d1d98eb6de55793d0cc29","modelId":"gemini-3.6-flash","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":92.82828282828281,"warning":null},{"id":"c8e7c88617e57d2c35a87d27","modelId":"gemini-3.6-flash","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":40.824837812789596,"warning":null},{"id":"b17a8a203a5700bf6b24b1dd","modelId":"gemini-3.6-flash","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":14.444444444444398,"warning":null},{"id":"f04dc462968ee573e79360f3","modelId":"gemini-3.6-flash","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":83.2369942196532,"warning":null},{"id":"3e2c1bfc05ec197865af08da","modelId":"gemini-3.6-flash","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":22.1333333333333,"warning":null},{"id":"67ab689539a46af639c2b744","modelId":"gemini-3.6-flash","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":49.96666666666666,"warning":null},{"id":"78d802fcdda621aaeea60836","modelId":"gemini-3.6-flash","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":53.3564814814815,"warning":null},{"id":"78d774eebc1ccfbeb52630b6","modelId":"gemini-3.6-flash","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":29.8969072164948,"warning":null},{"id":"f045d061de3521153304de3e","modelId":"gemini-3.6-flash","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":77.52808988764049,"warning":null},{"id":"23637f59cee916ee9d76d1c3","modelId":"gemini-3.6-flash","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-6-flash","score":7.07070707070707,"warning":null},{"id":"a40850daad4b8e157f489b7d","modelId":"gemini-3.6-flash","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Source model ID: gemini-3-6-flash-high; 120 tasks","score":60.416666666666664,"warning":null},{"id":"05c31f600e16e2079feb1579","modelId":"gemini-3.6-flash","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Gemini 3.6 Flash (Low)","configuration":"Gemini 3.6 Flash (Low); Source model ID: gemini-3-6-flash-low; 120 tasks","score":30.416666666666664,"warning":null},{"id":"3965013d4d14d4352a1e30c3","modelId":"gemini-3.6-flash","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Gemini 3.6 Flash (Medium)","configuration":"Gemini 3.6 Flash (Medium); Source model ID: gemini-3-6-flash-medium; 120 tasks","score":50.416666666666664,"warning":null},{"id":"d701fa1ba2e4bdda269e42e3","modelId":"gemini-3.6-flash","benchmarkId":"arc-v2semiprivate","level":"minimal","sourceLabel":"Gemini 3.6 Flash (Minimal)","configuration":"Gemini 3.6 Flash (Minimal); Source model ID: gemini-3-6-flash-minimal; 120 tasks","score":2.638888888888889,"warning":null},{"id":"c81497e333fcbd26ddd6cc87","modelId":"gemini-3.6-flash","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gemini-3-6-flash","configuration":"gemini-3-6-flash; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":46.7,"warning":null},{"id":"7d626458f757a704cb40993e","modelId":"gemini-3.7-flash","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":60,"warning":null},{"id":"02452f43f0906bf211af8052","modelId":"gemini-3.7-flash","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":1113.53,"warning":null},{"id":"ef410fd2e280ce86327cab12","modelId":"gemini-3.7-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":81.6666666666667,"warning":null},{"id":"a50ff552278552c7b0864b1c","modelId":"gemini-3.7-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":78.6666666666667,"warning":null},{"id":"61c9e7a5e9a8ca4064af9438","modelId":"gemini-3.7-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":83,"warning":null},{"id":"ae356c029756cf900a13d275","modelId":"gemini-3.7-flash","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":62.02749292053469,"warning":null},{"id":"12d553e565c15bb0a436a501","modelId":"gemini-3.7-flash","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":14.285714285714299,"warning":null},{"id":"2fb89ccac3f947e075f1ceb5","modelId":"gemini-3.7-flash","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":5.71428571428571,"warning":null},{"id":"33ea17dc0e746d0c0278d8f2","modelId":"gemini-3.7-flash","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":9.42857142857143,"warning":null},{"id":"2632ae4dfa63f7ee807f83de","modelId":"gemini-3.7-flash","benchmarkId":"aa-enterprise-ops-gym-aa","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":50.40286481647269,"warning":null},{"id":"eb122259c5dcd10719add943","modelId":"gemini-3.7-flash","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":23.599999999999998,"warning":null},{"id":"cfb85e050b5f7f5bfc44cde5","modelId":"gemini-3.7-flash","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":1391.99,"warning":null},{"id":"f50168725f95f70d08472acd","modelId":"gemini-3.7-flash","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":1323.42,"warning":null},{"id":"b8bf30f93c2254241465dd36","modelId":"gemini-3.7-flash","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":1360.38,"warning":null},{"id":"cc958ab0fa3f87a043cab6ca","modelId":"gemini-3.7-flash","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":94.5454545454545,"warning":null},{"id":"f80db42b566efd24283ebe94","modelId":"gemini-3.7-flash","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":90.1010101010101,"warning":null},{"id":"be18150ddc1f24099133fe59","modelId":"gemini-3.7-flash","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":92.12121212121211,"warning":null},{"id":"e1118315b9730430003d9eea","modelId":"gemini-3.7-flash","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":47.8683966635774,"warning":null},{"id":"fa9c4a6960f3fb31ffeac709","modelId":"gemini-3.7-flash","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":35.1251158480074,"warning":null},{"id":"96a98b14aa65488b1c1f2149","modelId":"gemini-3.7-flash","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":38.9712696941613,"warning":null},{"id":"0db210b57bf80d7a345bf5f3","modelId":"gemini-3.7-flash","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":15,"warning":null},{"id":"e9b19a3ed698a239681a9983","modelId":"gemini-3.7-flash","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":85.4913294797688,"warning":null},{"id":"5fa7fcc9b565b0ae44398e29","modelId":"gemini-3.7-flash","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":84.8554913294798,"warning":null},{"id":"7c33befa4e3fefff9354880f","modelId":"gemini-3.7-flash","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":84.7398843930636,"warning":null},{"id":"21324d1cf4d89f2207e1873c","modelId":"gemini-3.7-flash","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":26.4833333333333,"warning":null},{"id":"53f605c1b4240fda286c18ec","modelId":"gemini-3.7-flash","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":22.1333333333333,"warning":null},{"id":"1aa81ba21023b341862cbd02","modelId":"gemini-3.7-flash","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":23.7,"warning":null},{"id":"a4efd213390629ce7f6f4eae","modelId":"gemini-3.7-flash","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":55.31666666666667,"warning":null},{"id":"9df4e59a667c0282be38c7bb","modelId":"gemini-3.7-flash","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":53.56666666666666,"warning":null},{"id":"43f8d7b751a3dc63fd082125","modelId":"gemini-3.7-flash","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":54,"warning":null},{"id":"734f4b8552268ce3703442f7","modelId":"gemini-3.7-flash","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":57.175925925925895,"warning":null},{"id":"deb39ca13b6306b2fac84c4d","modelId":"gemini-3.7-flash","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":55.671296296296305,"warning":null},{"id":"22cf15a6eebc465978ec8936","modelId":"gemini-3.7-flash","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":59.837962962963,"warning":null},{"id":"e8edec766f81f7e749b1262d","modelId":"gemini-3.7-flash","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":32.7835051546392,"warning":null},{"id":"f8f269f7b68e85665e3ad933","modelId":"gemini-3.7-flash","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":29.4845360824742,"warning":null},{"id":"9441b7c178799b1719461f80","modelId":"gemini-3.7-flash","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":35.4639175257732,"warning":null},{"id":"cd5d287ac8ca758d4312bd29","modelId":"gemini-3.7-flash","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":85.76779026217231,"warning":null},{"id":"a47deb78a3ba7964e3889171","modelId":"gemini-3.7-flash","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-low","score":79.7752808988764,"warning":null},{"id":"5a34f66134d452fb3426faf7","modelId":"gemini-3.7-flash","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash-medium","score":78.2771535580524,"warning":null},{"id":"7e8996c52abfcd6cbed2439d","modelId":"gemini-3.7-flash","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-7-flash","score":13.6363636363636,"warning":null},{"id":"1e7c651364e9ee25998857ec","modelId":"gemini-3.7-flash","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Source model ID: google-gemini-3-7-flash-high; 120 tasks","score":84.58333333333333,"warning":null},{"id":"c66a9ac890388e11ea72db7a","modelId":"gemini-3.7-flash","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Gemini 3.7 Flash (Low)","configuration":"Gemini 3.7 Flash (Low); Source model ID: google-gemini-3-7-flash-low; 120 tasks","score":52.916666666666664,"warning":null},{"id":"198d114dc938db6f199de747","modelId":"gemini-3.7-flash","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Gemini 3.7 Flash (Medium)","configuration":"Gemini 3.7 Flash (Medium); Source model ID: google-gemini-3-7-flash-medium; 120 tasks","score":63.74999999999999,"warning":null},{"id":"b75fe3ec33c013c7d01ac1f8","modelId":"gemini-3.7-flash","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gemini-3-7-flash","configuration":"gemini-3-7-flash; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":65.3,"warning":null},{"id":"0a8c13fbff0cf24590cc04f5","modelId":"gemini-3.7-flash","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"gemini-3-7-flash","configuration":"gemini-3-7-flash; Reported effort: low; 113 unique tasks; 452 attempts; 4 runs","score":53.8,"warning":null},{"id":"3ee91dfd4e3a98ca959ff95c","modelId":"gemini-3.7-flash","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"gemini-3-7-flash","configuration":"gemini-3-7-flash; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":65.5,"warning":null},{"id":"2a2a039f4926e6fd4a8da723","modelId":"gemini-3.8-flash","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":46.25,"warning":null},{"id":"7be8043dde5cf363a5fc9ecf","modelId":"gemini-3.8-flash","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":1202.59,"warning":null},{"id":"b4f2b43c0dad9f2ba018836a","modelId":"gemini-3.8-flash","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":1062.43,"warning":null},{"id":"b10ecd3613b0fcc89caaafc6","modelId":"gemini-3.8-flash","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":1181.5,"warning":null},{"id":"9710e86093818542d0c8bcf2","modelId":"gemini-3.8-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":81.3333333333333,"warning":null},{"id":"52f9cd9d945c0b6f7aad58c1","modelId":"gemini-3.8-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":80.6666666666667,"warning":null},{"id":"9b0b430c5d4e0a6491874fdf","modelId":"gemini-3.8-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":84,"warning":null},{"id":"ca1bdfcefa97946259ce1904","modelId":"gemini-3.8-flash","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":59.93009432118242,"warning":null},{"id":"5d6a55e75f6dd3fb47c9b024","modelId":"gemini-3.8-flash","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":36.509404614862525,"warning":null},{"id":"338a154cee063e3244bacc88","modelId":"gemini-3.8-flash","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":60.8899739211415,"warning":null},{"id":"25fded040edcdd58fe6281c1","modelId":"gemini-3.8-flash","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":18.2857142857143,"warning":null},{"id":"026468c856fa6dda31a694f9","modelId":"gemini-3.8-flash","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":4,"warning":null},{"id":"1a3b20ba58bc7134ac849d73","modelId":"gemini-3.8-flash","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":12.285714285714299,"warning":null},{"id":"7e04cafb849e55a40b17473a","modelId":"gemini-3.8-flash","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":21,"warning":null},{"id":"d3346567f9f0c8ecc20a2568","modelId":"gemini-3.8-flash","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":18.6,"warning":null},{"id":"501240564b9382311038b8bb","modelId":"gemini-3.8-flash","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":22.8,"warning":null},{"id":"36fcd9ce4641201fecc5acba","modelId":"gemini-3.8-flash","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":1436.3,"warning":null},{"id":"9f0a026875a5564f30d798de","modelId":"gemini-3.8-flash","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":1307.18,"warning":null},{"id":"9507f172a530c76fd2c6cd02","modelId":"gemini-3.8-flash","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":1421.18,"warning":null},{"id":"5c96cda83f2bff02c6b2347c","modelId":"gemini-3.8-flash","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":95.2525252525253,"warning":null},{"id":"8b4a4bcaebabd3e99b35f5e7","modelId":"gemini-3.8-flash","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":92.020202020202,"warning":null},{"id":"ce50cafe7ebde5801ff708c5","modelId":"gemini-3.8-flash","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":93.53535353535351,"warning":null},{"id":"8bb756f5fcc15b1da5bce321","modelId":"gemini-3.8-flash","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":0,"warning":null},{"id":"f93fd8bca15766daeb66b106","modelId":"gemini-3.8-flash","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":47.8220574606117,"warning":null},{"id":"7af02628a67f7b382599658b","modelId":"gemini-3.8-flash","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":37.0713623725672,"warning":null},{"id":"1ffda298068f0b850f837a12","modelId":"gemini-3.8-flash","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":42.122335495829496,"warning":null},{"id":"2f8a3c5699dfb805cb90155c","modelId":"gemini-3.8-flash","benchmarkId":"aa-itbench-aa","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":52.5423728813559,"warning":null},{"id":"4edf03a7323d23a294152d0b","modelId":"gemini-3.8-flash","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":21.6666666666667,"warning":null},{"id":"1daa1042c5774b84eec55f32","modelId":"gemini-3.8-flash","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":85.606936416185,"warning":null},{"id":"3cfdb2c1b20c6a6e8c8a9f96","modelId":"gemini-3.8-flash","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":84.5086705202312,"warning":null},{"id":"beb49c14f528ca7119bbee39","modelId":"gemini-3.8-flash","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":84.2196531791908,"warning":null},{"id":"51ac259d6b101f8a33c37a10","modelId":"gemini-3.8-flash","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":29.55,"warning":null},{"id":"78ab7ea85240b0570a09366a","modelId":"gemini-3.8-flash","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":21.35,"warning":null},{"id":"0f099fd18d469ae207c64533","modelId":"gemini-3.8-flash","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":28.5833333333333,"warning":null},{"id":"10d8f946aa1fb0538e073234","modelId":"gemini-3.8-flash","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":54.6,"warning":null},{"id":"7f08233113582c14779dd206","modelId":"gemini-3.8-flash","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":52.21666666666667,"warning":null},{"id":"32bf02014540ea947b5399f8","modelId":"gemini-3.8-flash","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":52.983333333333334,"warning":null},{"id":"002be94559a69a5b5d604dd9","modelId":"gemini-3.8-flash","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":56.5972222222222,"warning":null},{"id":"2368f4b2a96dd8c54c52aa88","modelId":"gemini-3.8-flash","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":54.976851851851805,"warning":null},{"id":"45ab3f2daab0eb2d65b7194f","modelId":"gemini-3.8-flash","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":55.0925925925926,"warning":null},{"id":"b9fe49f5e7fee07bc997d99d","modelId":"gemini-3.8-flash","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":44.948453608247405,"warning":null},{"id":"0e2e3b61a8161a00d8ee30df","modelId":"gemini-3.8-flash","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":33.1958762886598,"warning":null},{"id":"05d93146879a224c432c044d","modelId":"gemini-3.8-flash","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":45.7731958762887,"warning":null},{"id":"42127560cf7d7fa2db6e2b68","modelId":"gemini-3.8-flash","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":10,"warning":null},{"id":"87c00bb7dff60e21a1a7137d","modelId":"gemini-3.8-flash","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":87.6404494382023,"warning":null},{"id":"34df6b362fc5a672a3dcc0ef","modelId":"gemini-3.8-flash","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":83.1460674157303,"warning":null},{"id":"50818b6cb40adbc83dff4e15","modelId":"gemini-3.8-flash","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":83.8951310861423,"warning":null},{"id":"6068591c1dee2ad5a007b455","modelId":"gemini-3.8-flash","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash","score":19.6969696969697,"warning":null},{"id":"c373d8b69d9c0c0dfb86c1b9","modelId":"gemini-3.8-flash","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-low","score":10.1010101010101,"warning":null},{"id":"dc2bfe8291f383a94144c57c","modelId":"gemini-3.8-flash","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Model source: https://artificialanalysis.ai/models/gemini-3-8-flash-medium","score":19.6969696969697,"warning":null},{"id":"691b8c228bc7e3ba32ee2a9e","modelId":"gemini-3.8-flash","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Source model ID: google-gemini-3-8-flash-high; 120 tasks","score":89.16666666666667,"warning":null},{"id":"9c94dc48198fd29a28514fe5","modelId":"gemini-3.8-flash","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Gemini 3.8 Flash (Low)","configuration":"Gemini 3.8 Flash (Low); Source model ID: google-gemini-3-8-flash-low; 120 tasks","score":77.5,"warning":null},{"id":"5f9086d402496731bb3ffe0b","modelId":"gemini-3.8-flash","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Gemini 3.8 Flash (Medium)","configuration":"Gemini 3.8 Flash (Medium); Source model ID: google-gemini-3-8-flash-medium; 120 tasks","score":82.91666666666667,"warning":null},{"id":"405dcd6b0cb92b2c4c19515e","modelId":"gemini-3.8-flash","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gemini-3-8-flash","configuration":"gemini-3-8-flash; Reported effort: high; 113 unique tasks; 447 attempts; 4 runs","score":73.8,"warning":null},{"id":"02f746944e782131d58a2c53","modelId":"gemini-3.8-flash","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"gemini-3-8-flash","configuration":"gemini-3-8-flash; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":71,"warning":null},{"id":"443e76799dd16afee71b7bd6","modelId":"gemini-4-argon","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":48.75,"warning":null},{"id":"e75cc6921daf800d61ef3306","modelId":"gemini-4-argon","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":1486.44,"warning":null},{"id":"5215139d2dec8f72e73588e8","modelId":"gemini-4-argon","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":79.6666666666667,"warning":null},{"id":"b197a99a18f6279260b7488f","modelId":"gemini-4-argon","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":77.51346945899195,"warning":null},{"id":"47f00fc795429de6e5f6b690","modelId":"gemini-4-argon","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":27.142857142857103,"warning":null},{"id":"6b6d73339472f9f816104b66","modelId":"gemini-4-argon","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":21.8,"warning":null},{"id":"efc098fcec90cb0785a51935","modelId":"gemini-4-argon","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":1623.99,"warning":null},{"id":"19ed06217cb5b3ef907563b5","modelId":"gemini-4-argon","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":57.089898053753494,"warning":null},{"id":"b593330a2d29d36cd3398658","modelId":"gemini-4-argon","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":42.35,"warning":null},{"id":"24683e454c5272d5eedaccd5","modelId":"gemini-4-argon","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":49.9,"warning":null},{"id":"0aa4cb5bd3413e5f599b47fd","modelId":"gemini-4-argon","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":61.8055555555556,"warning":null},{"id":"998b5265b532b075b0b1aab8","modelId":"gemini-4-argon","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":52.857142857142904,"warning":null},{"id":"a6bcb32f825f13a672910a9c","modelId":"gemini-4-argon","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Model source: https://artificialanalysis.ai/models/gemini-4-argon","score":57.07070707070711,"warning":null},{"id":"2952a1bddfebef759c0a3287","modelId":"gemini-4-argon","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Source model ID: google-gemini-4-argon-high; 120 tasks","score":91.25,"warning":null},{"id":"a840a820040dddd91f8cee08","modelId":"gemini-4-argon","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Gemini 4 Argon (Low)","configuration":"Gemini 4 Argon (Low); Source model ID: google-gemini-4-argon-low; 120 tasks","score":42.083333333333336,"warning":null},{"id":"a62f2d97a7ae5872052b57da","modelId":"gemini-4-argon","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Gemini 4 Argon (Medium)","configuration":"Gemini 4 Argon (Medium); Source model ID: google-gemini-4-argon-medium; 120 tasks","score":87.08333333333333,"warning":null},{"id":"bae0ffc1da12b473f74a2b1f","modelId":"gemma-3-12b-it","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.3333333333333,"warning":null},{"id":"b1ac9024e3d4ec61f9a27b73","modelId":"gemma-3-12b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.33333333333333,"warning":null},{"id":"9f5c45f943e79042aec9e467","modelId":"gemma-3-12b-it","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.19111706206154658,"warning":null},{"id":"e92bfd8b7eb0fc97830fe231","modelId":"gemma-3-12b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"a6589219b0e0884945d77b28","modelId":"gemma-3-12b-it","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1,"warning":null},{"id":"1b9c1f0350e49f7f646eb850","modelId":"gemma-3-12b-it","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-408.21,"warning":null},{"id":"475a94e16f23c1a63ea69240","modelId":"gemma-3-12b-it","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.93229166666666,"warning":null},{"id":"adccbe9b822ee7f815ae4a80","modelId":"gemma-3-12b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.949494949494905,"warning":null},{"id":"641540b21e3f392cf8f0bd9a","modelId":"gemma-3-12b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.21,"warning":null},{"id":"71da35fd94ce73f427f2e7ce","modelId":"gemma-3-12b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.734693877551,"warning":null},{"id":"22e8ad1e201236cae3a05725","modelId":"gemma-3-12b-it","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.650793650793599,"warning":null},{"id":"080268458062c51d69e21edf","modelId":"gemma-3-12b-it","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":85.3333333333333,"warning":null},{"id":"c0ba5642ee15db5fba8b4514","modelId":"gemma-3-12b-it","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":59.4913563829787,"warning":null},{"id":"e09f5f3a1d31437d4e2c6d10","modelId":"gemma-3-12b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.514450867052,"warning":null},{"id":"5ac2366a97ebc55c10126e78","modelId":"gemma-3-12b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-77.0666666666667,"warning":null},{"id":"65e36893b56d652cbfd11b4a","modelId":"gemma-3-12b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.299999999999999,"warning":null},{"id":"4dfa25465b0259154b3eed24","modelId":"gemma-3-12b-it","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.4351851851852,"warning":null},{"id":"bb329105825c1bb19bf3af65","modelId":"gemma-3-12b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.8187134502924,"warning":null},{"id":"a374e979dbabf157744673dc","modelId":"gemma-3-12b-it","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.824742268041237,"warning":null},{"id":"2b0bbcfc9a419ea70d04d3c8","modelId":"gemma-3-12b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.757575757575758,"warning":null},{"id":"e99a62a6a4b3677568cbb008","modelId":"gemma-3-12b-it","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"fe7aac0ecbb0c2849f090886","modelId":"gemma-3-12b-it","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Gemma 3 12B Instruct","configuration":"Gemma 3 12B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-12b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"870a7107c43ca88307ffdbaf","modelId":"gemma-3-1b-it","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.33333333333333,"warning":null},{"id":"ea441b9a6112bdfae0f748f8","modelId":"gemma-3-1b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"93c701d15fcbd95282d8974f","modelId":"gemma-3-1b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"ea7330bf72144219f9f1610c","modelId":"gemma-3-1b-it","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.020833333333336,"warning":null},{"id":"a85c5930a31d561cbd2a37be","modelId":"gemma-3-1b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.737373737373698,"warning":null},{"id":"5d6447d51ae1eeef0c7328cd","modelId":"gemma-3-1b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.32900834105653,"warning":null},{"id":"a8f8c4241e62ccc94f5c438c","modelId":"gemma-3-1b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.9319727891156,"warning":null},{"id":"967c42becbfc1b7f69a023c4","modelId":"gemma-3-1b-it","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.6931216931216901,"warning":null},{"id":"b405af3f368ddd8defb911f5","modelId":"gemma-3-1b-it","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.4,"warning":null},{"id":"19ee2523e0c844fa8d5056a4","modelId":"gemma-3-1b-it","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.4724069148936,"warning":null},{"id":"d572ac181de7f4a7bd81c4e9","modelId":"gemma-3-1b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-75.45,"warning":null},{"id":"986669b662cfc130d20c3054","modelId":"gemma-3-1b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.75,"warning":null},{"id":"e7781a6c80a8cf308f349051","modelId":"gemma-3-1b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.526315789473701,"warning":null},{"id":"fd8260748e66a80c27ada5ea","modelId":"gemma-3-1b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 3 1B Instruct","configuration":"Gemma 3 1B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-1b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"52d80905949128ccdd8df2d8","modelId":"gemma-3-270m-it","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.33333333333333,"warning":null},{"id":"76fadff0e219977139924094","modelId":"gemma-3-270m-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"9fbe186ff0d120f39021406b","modelId":"gemma-3-270m-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"3309f5884078782f5109c1aa","modelId":"gemma-3-270m-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.4242424242424,"warning":null},{"id":"296e9bf178830a18dc28aea5","modelId":"gemma-3-270m-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.6144578313252995,"warning":null},{"id":"b36408c26ff599f500a313f0","modelId":"gemma-3-270m-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.108843537415,"warning":null},{"id":"5d43728608ad6d1780bac10c","modelId":"gemma-3-270m-it","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.317460317460317,"warning":null},{"id":"ae1485e7ba3805532af5d722","modelId":"gemma-3-270m-it","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.52692819148936,"warning":null},{"id":"20bea5cafe0c9e45a478f6d6","modelId":"gemma-3-270m-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":-29.2666666666667,"warning":null},{"id":"da6a60d5479d339a034ebb01","modelId":"gemma-3-270m-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.8833333333333334,"warning":null},{"id":"5d60afe4189ee37f28c901da","modelId":"gemma-3-270m-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.06432748538012,"warning":null},{"id":"5482e2f6a8f5c425ecbe1b94","modelId":"gemma-3-270m-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 3 270M","configuration":"Gemma 3 270M; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-270m; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"6171d7d4fbe5f0e0c06bf689","modelId":"gemma-3-27b-it","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.6666666666667,"warning":null},{"id":"879b05bb7efdb989b22b0dc3","modelId":"gemma-3-27b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.33333333333333,"warning":null},{"id":"06a85450b2a17ff417c7e6eb","modelId":"gemma-3-27b-it","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.19111706206154658,"warning":null},{"id":"2a72e48f0c589d27795a37bd","modelId":"gemma-3-27b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"1ba477dce1a0ef87c127a612","modelId":"gemma-3-27b-it","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.2,"warning":null},{"id":"b4e8816ad4212bc32c6c8349","modelId":"gemma-3-27b-it","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-407.98,"warning":null},{"id":"5421892f4adacf93e8314be4","modelId":"gemma-3-27b-it","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.36979166666666,"warning":null},{"id":"8ff7f32b33421c1c9c6ce965","modelId":"gemma-3-27b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.8282828282828,"warning":null},{"id":"b72a49afa1eabcb932b719b4","modelId":"gemma-3-27b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.3999999999999995,"warning":null},{"id":"c9a7a6546e6e6d3d50d90e69","modelId":"gemma-3-27b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.8367346938776,"warning":null},{"id":"c238d9191ee2d48b6a84b9f0","modelId":"gemma-3-27b-it","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.650793650793599,"warning":null},{"id":"77df7e4831b72de2ed9a1961","modelId":"gemma-3-27b-it","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.26666666666671,"warning":null},{"id":"7c0d40f3fd2faa8072484fd1","modelId":"gemma-3-27b-it","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.86336436170211,"warning":null},{"id":"1d78d7410ecf03f58df197db","modelId":"gemma-3-27b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.0346820809249,"warning":null},{"id":"09f9d18827eea3ab094fd5d4","modelId":"gemma-3-27b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-67.2166666666667,"warning":null},{"id":"2e45792d6883b2e89cc3c62c","modelId":"gemma-3-27b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.950000000000001,"warning":null},{"id":"fe3bf9bdec38cedda2c4603a","modelId":"gemma-3-27b-it","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.2638888888889,"warning":null},{"id":"ea5e72c4e76b0f441b7b788e","modelId":"gemma-3-27b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.526315789473701,"warning":null},{"id":"0382a1aa06c99f35d0e5edc0","modelId":"gemma-3-27b-it","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.824742268041237,"warning":null},{"id":"6c6288b8deb73d4839d789d6","modelId":"gemma-3-27b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.78787878787879,"warning":null},{"id":"7c690cd70e1220383d0f7914","modelId":"gemma-3-27b-it","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.49438202247191,"warning":null},{"id":"bbca9a0d6ee1668274b86945","modelId":"gemma-3-27b-it","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Gemma 3 27B Instruct","configuration":"Gemma 3 27B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-27b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"d22ebd4720d57b1d0764161d","modelId":"gemma-3-4b-it","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.666666666666702,"warning":null},{"id":"613fa8b94c50285eabaf9c8f","modelId":"gemma-3-4b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.66666666666667,"warning":null},{"id":"a442736b72445b13048e7d53","modelId":"gemma-3-4b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"cf36cb19af8c5e42ca17f4dd","modelId":"gemma-3-4b-it","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.447916666666664,"warning":null},{"id":"0cb0a7036c6d87b2926d3b6e","modelId":"gemma-3-4b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.0909090909091,"warning":null},{"id":"7d8ce4772f34fadf12075913","modelId":"gemma-3-4b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.29,"warning":null},{"id":"0b13cb76860c726d995622e9","modelId":"gemma-3-4b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.2993197278912,"warning":null},{"id":"62201b3cf05086b0b77a78bb","modelId":"gemma-3-4b-it","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.2169312169312,"warning":null},{"id":"555c7d3fe8cc0863494ec509","modelId":"gemma-3-4b-it","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.6,"warning":null},{"id":"616fdd55adeeae6be1242f52","modelId":"gemma-3-4b-it","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.688829787234,"warning":null},{"id":"99f7e488d25cf68d9af18b26","modelId":"gemma-3-4b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.942196531791897,"warning":null},{"id":"84dea6e1e3ef3afc74262d95","modelId":"gemma-3-4b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-82.8833333333333,"warning":null},{"id":"024ea9c47f612bdb82a2af67","modelId":"gemma-3-4b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.733333333333333,"warning":null},{"id":"b46e070e0d48161a9317ca32","modelId":"gemma-3-4b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.97076023391813,"warning":null},{"id":"94f7565276574925a5e8b94f","modelId":"gemma-3-4b-it","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.412371134020619,"warning":null},{"id":"7c2754460e15897101c6669f","modelId":"gemma-3-4b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.757575757575758,"warning":null},{"id":"3a5f475968c70faa3e41558a","modelId":"gemma-3-4b-it","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Gemma 3 4B Instruct","configuration":"Gemma 3 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.374531835205993,"warning":null},{"id":"254339b1bd7ba893610b28a5","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.3333333333333,"warning":null},{"id":"5cc3c6f529c5fcaac01b08b9","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"d24e94beebb7749ea1d8ffea","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"3e9213eb8cb6f7bda1d0f1e5","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.9292929292929,"warning":null},{"id":"6e6e05f44c22c523f43e0263","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.17052826691381,"warning":null},{"id":"fc57f0fecccf2986223f97b0","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.0408163265306,"warning":null},{"id":"265009dd01320779e45d8889","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.52380952380952,"warning":null},{"id":"6c6b8a1e0fb87413ec47be0e","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.1333333333333,"warning":null},{"id":"e2a432a9133e962dcc3af709","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.815824468085104,"warning":null},{"id":"b27d10aabf75f3a064bc461e","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-80.1666666666667,"warning":null},{"id":"085cdd0603d97d7410437c05","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.833333333333333,"warning":null},{"id":"6beb07747331dd05fccc4d18","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f0214606125afc28f6556fde","modelId":"gemma-3n-e2b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 3n E2B Instruct","configuration":"Gemma 3n E2B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.757575757575758,"warning":null},{"id":"ea42f23527add8b9c39d253d","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.3333333333333,"warning":null},{"id":"99d58e99ccd89cd35c2b912b","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"696ce85764f9c9d4b1ba6e1c","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"c0d8d76e50c09ce3e119f8ba","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.5959595959596,"warning":null},{"id":"fc586044e84b48830b7341ca","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.49490268767377,"warning":null},{"id":"01760f7dbd3659373c5cb7d0","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.891156462585,"warning":null},{"id":"0a88eeceeac43fa8c6eca502","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.603174603174601,"warning":null},{"id":"8cd759735b2f8417924de7a2","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.0666666666667,"warning":null},{"id":"90efdea1edf0d70b9c91f2e0","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.828125,"warning":null},{"id":"f604330d5bdfb9b2f5bfd3f5","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.242774566473997,"warning":null},{"id":"9082176c134078aa550097ff","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-81.3333333333333,"warning":null},{"id":"2ae2a55f56fb56a01959e687","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.983333333333334,"warning":null},{"id":"5a340915e75cfe29b8a881e4","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.97076023391813,"warning":null},{"id":"a0fdfcd89c7b44fa4f91571d","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.206185567010309,"warning":null},{"id":"b88afa6c66f560b3e323a93e","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.27272727272727,"warning":null},{"id":"42b31d167ae12b063c4787ce","modelId":"gemma-3n-e4b-it","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Gemma 3n E4B Instruct","configuration":"Gemma 3n E4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gemma-3n-e4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.7490636704119851,"warning":null},{"id":"2fa190ad552336df40fc5759","modelId":"gemma-4-12b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":35,"warning":null},{"id":"7f25db26af07cec60d504c7c","modelId":"gemma-4-12b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":63.66666666666671,"warning":null},{"id":"716efdabc122b732d667420c","modelId":"gemma-4-12b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":0,"warning":null},{"id":"bea6c787579f9a1237d89ed4","modelId":"gemma-4-12b-it","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":0,"warning":null},{"id":"9ce2a98a13e9e059a62ee45e","modelId":"gemma-4-12b-it","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":427.31,"warning":null},{"id":"bf1dc487a514004349865eb5","modelId":"gemma-4-12b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":66.06060606060609,"warning":null},{"id":"a333a9160f2666010facc8ed","modelId":"gemma-4-12b-it","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":75.2525252525252,"warning":null},{"id":"9b0cac4689133115cef3a717","modelId":"gemma-4-12b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":6.34847080630213,"warning":null},{"id":"e9d2340a77f08483cc4db4b8","modelId":"gemma-4-12b-it","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":15.6626506024096,"warning":null},{"id":"01faa81311cd7b4f50088ac4","modelId":"gemma-4-12b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":45.1700680272109,"warning":null},{"id":"a15f95933b64d5141645bcf6","modelId":"gemma-4-12b-it","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":73.5374149659864,"warning":null},{"id":"d44fa4443a7ce8124d0a70d3","modelId":"gemma-4-12b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":61.965317919075105,"warning":null},{"id":"38462004b6ccf5545dc45a40","modelId":"gemma-4-12b-it","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":69.6531791907514,"warning":null},{"id":"2b90c9332daa18e534fd0fcf","modelId":"gemma-4-12b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":-52.8,"warning":null},{"id":"ce50829c14122f3091899493","modelId":"gemma-4-12b-it","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":-52.7166666666667,"warning":null},{"id":"2751c624590967f5d5249682","modelId":"gemma-4-12b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":11.883333333333333,"warning":null},{"id":"f5850b32526048230419690d","modelId":"gemma-4-12b-it","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":15.616666666666667,"warning":null},{"id":"5e5cb56ee5eb6803aa8bd6a2","modelId":"gemma-4-12b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":31.8713450292398,"warning":null},{"id":"d9aa2621884ddc873af35623","modelId":"gemma-4-12b-it","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":36.2573099415205,"warning":null},{"id":"350822620ac30f66a2da1d9b","modelId":"gemma-4-12b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 4 12B (Non-reasoning)","configuration":"Gemma 4 12B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b-non-reasoning","score":11.363636363636399,"warning":null},{"id":"5ca27f0cf0791ac115c1d87e","modelId":"gemma-4-12b-it","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":18.181818181818198,"warning":null},{"id":"afa273af8ca9474f2626494e","modelId":"gemma-4-12b-it","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemma 4 12B (Reasoning)","configuration":"Gemma 4 12B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-12b","score":27.340823970037498,"warning":null},{"id":"b740a64246adba5c08c32d75","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":315.08,"warning":null},{"id":"949edb06f6af88efae6f287b","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":42.3333333333333,"warning":null},{"id":"2ecf2ada691fdd1c0a03b480","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":65.6666666666667,"warning":null},{"id":"b7dddcd63170c619f5d1707c","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":1.9430215909597492,"warning":null},{"id":"6fc2dd3e0d651bd9249e2633","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":0,"warning":null},{"id":"f6dcd64e671508e237b35918","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":0,"warning":null},{"id":"72ff96edc2e4b3bfe7936b1c","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":565.75,"warning":null},{"id":"5fcfd34173707143f2a99925","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":71.4141414141414,"warning":null},{"id":"297b1d6d72e973e016718eff","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":79.1919191919192,"warning":null},{"id":"0311d8249343cd490439681b","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":11.4921223354958,"warning":null},{"id":"8b3cd8bd20e702eedb423658","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":19.3234476367006,"warning":null},{"id":"4b414517eb3739571023617d","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":45.442176870748305,"warning":null},{"id":"b0b8625f23efeef97463192d","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":72.4489795918367,"warning":null},{"id":"123c6d4efe200397de794a3b","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":23.6346516007533,"warning":null},{"id":"f245bbe38d52f3b84a64e4dc","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":66.7052023121387,"warning":null},{"id":"7177df54c2337ce7eaf62e08","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":69.2485549132948,"warning":null},{"id":"67499adc1b46f586a1c3d2b0","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":-62.3166666666667,"warning":null},{"id":"49acca20990729f11bd62386","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":-50.8333333333333,"warning":null},{"id":"f2a7a0301ac7cdb230df1663","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":15.483333333333333,"warning":null},{"id":"319e9e5b1b3e864344a503b3","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":19.1,"warning":null},{"id":"7f48b80e0edac6838dffb50d","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":40.046296296296305,"warning":null},{"id":"2bb7da5d78c57f44d827d9f9","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":40.3508771929825,"warning":null},{"id":"d8a5c15274c4f18730b4d9f5","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":43.5672514619883,"warning":null},{"id":"23c4101446e544ed2ac5b82f","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":11.9587628865979,"warning":null},{"id":"6cd407ecade291b598835150","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 4 26B A4B (Non-reasoning)","configuration":"Gemma 4 26B A4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b-non-reasoning","score":25,"warning":null},{"id":"81715bf25a80b1a5d222b5cf","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":13.639999999999999,"warning":null},{"id":"60c606d7f6815f665071f6e1","modelId":"gemma-4-26b-a4b-it","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemma 4 26B A4B (Reasoning)","configuration":"Gemma 4 26B A4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-26b-a4b","score":38.9513108614232,"warning":null},{"id":"00ca1349b4ad03059b696d93","modelId":"gemma-4-31b-it","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":367.87,"warning":null},{"id":"0a11579402aa7ddc9dd6348f","modelId":"gemma-4-31b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":46.6666666666667,"warning":null},{"id":"706f0676dad235cacaacf815","modelId":"gemma-4-31b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":69.6666666666667,"warning":null},{"id":"6d730c431d429efc51487ae1","modelId":"gemma-4-31b-it","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":4.933063448681232,"warning":null},{"id":"89a616e7c917c56392cafec6","modelId":"gemma-4-31b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":0,"warning":null},{"id":"601a3ec51179b6fffcfaead1","modelId":"gemma-4-31b-it","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":1.42857142857143,"warning":null},{"id":"5e1832dadb83cec7d9868f04","modelId":"gemma-4-31b-it","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":28.349746344374815,"warning":null},{"id":"b21e911e97485a34e9502144","modelId":"gemma-4-31b-it","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":6,"warning":null},{"id":"9d403a9e0e26ba60727627fc","modelId":"gemma-4-31b-it","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":547.44,"warning":null},{"id":"7d63b6a3c77cb4a775dd9bcb","modelId":"gemma-4-31b-it","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":619.79,"warning":null},{"id":"7fa903a68ff27edd309baae9","modelId":"gemma-4-31b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":76.2626262626263,"warning":null},{"id":"a0a564da6c85edee9be55904","modelId":"gemma-4-31b-it","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":85.6565656565657,"warning":null},{"id":"7701f8f5dda40cd06fa4c776","modelId":"gemma-4-31b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":11.8164967562558,"warning":null},{"id":"8e3a35ca0c938e259f5e70b4","modelId":"gemma-4-31b-it","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":23.6329935125116,"warning":null},{"id":"559274bb4d2db9e3dee3156b","modelId":"gemma-4-31b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":53.469387755102005,"warning":null},{"id":"26dcfe7fc4f0f3ca73ea75cd","modelId":"gemma-4-31b-it","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":75.57823129251699,"warning":null},{"id":"25e5a78d91ee39f3070e8ee1","modelId":"gemma-4-31b-it","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":37.288135593220304,"warning":null},{"id":"fca70e45bb76253f1f919976","modelId":"gemma-4-31b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":70.3468208092486,"warning":null},{"id":"6cbd352ef10ac02c691f3078","modelId":"gemma-4-31b-it","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":73.4104046242775,"warning":null},{"id":"ba8e82e53d70d6f8aafa303e","modelId":"gemma-4-31b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":-51.6833333333333,"warning":null},{"id":"098d80e8ba47d8a6ef0c241f","modelId":"gemma-4-31b-it","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":-47.9333333333333,"warning":null},{"id":"c7458f1927e4b1a2a3dabfd7","modelId":"gemma-4-31b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":16.633333333333333,"warning":null},{"id":"31dd36e84421125afbaf73c8","modelId":"gemma-4-31b-it","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":20.033333333333335,"warning":null},{"id":"e131f49fc0290c2c251ad59d","modelId":"gemma-4-31b-it","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":45.4861111111111,"warning":null},{"id":"5fb1a0a25e052c05799b4a8d","modelId":"gemma-4-31b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":65.4970760233918,"warning":null},{"id":"46fcde8fb24e810ecbd0e3fd","modelId":"gemma-4-31b-it","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":59.9415204678363,"warning":null},{"id":"38550b8a2a207eb0da07c6e6","modelId":"gemma-4-31b-it","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":8.8659793814433,"warning":null},{"id":"2d0393ab86c8e9008391b8f9","modelId":"gemma-4-31b-it","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":14.8453608247423,"warning":null},{"id":"4c693dcad0bad4f75e983ec7","modelId":"gemma-4-31b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":30.303030303030297,"warning":null},{"id":"6a59f286887b9777ef477767","modelId":"gemma-4-31b-it","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":36.363636363636395,"warning":null},{"id":"7082033eb30f427626a4ffe5","modelId":"gemma-4-31b-it","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Gemma 4 31B (Non-reasoning)","configuration":"Gemma 4 31B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b-non-reasoning","score":29.213483146067396,"warning":null},{"id":"e0daed7f980e0bd9dd3d6801","modelId":"gemma-4-31b-it","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":43.4456928838951,"warning":null},{"id":"632c819ff919f14202c38e37","modelId":"gemma-4-31b-it","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Gemma 4 31B (Reasoning)","configuration":"Gemma 4 31B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-31b","score":0,"warning":null},{"id":"5110c2885f54e024cd329be3","modelId":"gemma-4-e2b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":16.3333333333333,"warning":null},{"id":"6983a9fbb42127d71547b6b4","modelId":"gemma-4-e2b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":16.3333333333333,"warning":null},{"id":"64f0265c1003812ec781556f","modelId":"gemma-4-e2b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":0,"warning":null},{"id":"673c72fe749a675bb48c99c2","modelId":"gemma-4-e2b-it","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":0,"warning":null},{"id":"8752044e2ef5ae1845876a0d","modelId":"gemma-4-e2b-it","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":-187.7,"warning":null},{"id":"9c01f4e5c42675a3c6f3d9aa","modelId":"gemma-4-e2b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":40.5050505050505,"warning":null},{"id":"117449d508219e8cd6e43180","modelId":"gemma-4-e2b-it","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":43.3333333333333,"warning":null},{"id":"e0d5158b58c9798a658ce75a","modelId":"gemma-4-e2b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":4.72659870250232,"warning":null},{"id":"e3dabdefbac77e863019f219","modelId":"gemma-4-e2b-it","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":4.77293790546803,"warning":null},{"id":"2135126d4e4981631a2395d2","modelId":"gemma-4-e2b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":33.605442176870696,"warning":null},{"id":"a3fed7a0571b3df4e44deacb","modelId":"gemma-4-e2b-it","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":38.0272108843537,"warning":null},{"id":"935ddb78227e9b20558bbd5e","modelId":"gemma-4-e2b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":41.7919075144509,"warning":null},{"id":"9dbacd79912a83e02071a8f7","modelId":"gemma-4-e2b-it","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":44.5664739884393,"warning":null},{"id":"cc429fcc1335cb09f66d90b4","modelId":"gemma-4-e2b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":-60.1833333333333,"warning":null},{"id":"e73a2705c80206386d342583","modelId":"gemma-4-e2b-it","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":-23.5833333333333,"warning":null},{"id":"c21205c79690fad58d74523b","modelId":"gemma-4-e2b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":8.15,"warning":null},{"id":"0b86207217f529cfa79a01fb","modelId":"gemma-4-e2b-it","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":6.633333333333333,"warning":null},{"id":"4fa492fe78a6b6374bdaad9c","modelId":"gemma-4-e2b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":22.2222222222222,"warning":null},{"id":"381279b32093010e53f60864","modelId":"gemma-4-e2b-it","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":20.7602339181287,"warning":null},{"id":"ed4c579b74058292c68ba773","modelId":"gemma-4-e2b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 4 E2B (Non-reasoning)","configuration":"Gemma 4 E2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b-non-reasoning","score":2.27272727272727,"warning":null},{"id":"e62ae3c573bc0ed516e86b34","modelId":"gemma-4-e2b-it","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":3.03030303030303,"warning":null},{"id":"94f2c4df5cb23012e2ed42af","modelId":"gemma-4-e2b-it","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemma 4 E2B (Reasoning)","configuration":"Gemma 4 E2B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e2b","score":0.374531835205993,"warning":null},{"id":"b5dac129747405f7170bea58","modelId":"gemma-4-e4b-it","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":0,"warning":null},{"id":"7d5dc30c068300c623a91a48","modelId":"gemma-4-e4b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":24,"warning":null},{"id":"b93409ba27f2f5c6f9a53d14","modelId":"gemma-4-e4b-it","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":32,"warning":null},{"id":"16eab8318f6e4610ce33fe77","modelId":"gemma-4-e4b-it","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":0.46154542449164326,"warning":null},{"id":"7d1148ef014c1313b8bb33c9","modelId":"gemma-4-e4b-it","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"892bc5fe99c93c72090843f4","modelId":"gemma-4-e4b-it","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":0.571428571428571,"warning":null},{"id":"818529aecfebb3dd22cdf6f4","modelId":"gemma-4-e4b-it","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":1,"warning":null},{"id":"f357344d051689050257e609","modelId":"gemma-4-e4b-it","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":-36.44,"warning":null},{"id":"09f4fa9699ed45e540bccdf5","modelId":"gemma-4-e4b-it","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":54.949494949494905,"warning":null},{"id":"dc6981039f93bb34311c5aaa","modelId":"gemma-4-e4b-it","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":57.5757575757576,"warning":null},{"id":"94598d71e00dbc497012ed1f","modelId":"gemma-4-e4b-it","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":4.81927710843374,"warning":null},{"id":"8014fef86e630db2b5b16922","modelId":"gemma-4-e4b-it","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":3.75347544022243,"warning":null},{"id":"7c949f5675b407fd8b972461","modelId":"gemma-4-e4b-it","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":40.5442176870748,"warning":null},{"id":"8b37f6fe8c07bb8bf8c55d62","modelId":"gemma-4-e4b-it","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":44.2176870748299,"warning":null},{"id":"fb8d6f02b4833a2bfacb1689","modelId":"gemma-4-e4b-it","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":51.2138728323699,"warning":null},{"id":"8edf8e0f16e79a110f25a4db","modelId":"gemma-4-e4b-it","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":51.38728323699419,"warning":null},{"id":"6ecea92ea2a8f843cc72ea55","modelId":"gemma-4-e4b-it","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":-40.9833333333333,"warning":null},{"id":"bf37eb7ed991ae1c14f31e61","modelId":"gemma-4-e4b-it","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":-19.7,"warning":null},{"id":"411b36bb3f028339dc23a40a","modelId":"gemma-4-e4b-it","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":8.316666666666666,"warning":null},{"id":"bb15c5f471328aba08aaf5b4","modelId":"gemma-4-e4b-it","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":8.583333333333334,"warning":null},{"id":"bc1bd2b2d7480b7b9373c1c2","modelId":"gemma-4-e4b-it","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":24.421296296296298,"warning":null},{"id":"97f873fbbc402712668863f5","modelId":"gemma-4-e4b-it","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":26.023391812865498,"warning":null},{"id":"7bf99cf2364cac7350bf17b5","modelId":"gemma-4-e4b-it","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":20.7602339181287,"warning":null},{"id":"d6faa5c06f974d9bd6bd6075","modelId":"gemma-4-e4b-it","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Gemma 4 E4B (Non-reasoning)","configuration":"Gemma 4 E4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b-non-reasoning","score":7.57575757575758,"warning":null},{"id":"7e25eac2feb6eedc0f86c175","modelId":"gemma-4-e4b-it","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":8.33333333333333,"warning":null},{"id":"17d5928b1cc07aea7cc8ca5c","modelId":"gemma-4-e4b-it","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":1.8726591760299598,"warning":null},{"id":"7a38fe13fa75c4f325079a9e","modelId":"gemma-4-e4b-it","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Gemma 4 E4B (Reasoning)","configuration":"Gemma 4 E4B (Reasoning); Model source: https://artificialanalysis.ai/models/gemma-4-e4b","score":0,"warning":null},{"id":"8298ccc202056776af853910","modelId":"glm-4.5","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":73.6666666666667,"warning":null},{"id":"a005dca05213b4a73bd1f631","modelId":"glm-4.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":52.66666666666669,"warning":null},{"id":"9d40825300ffb78278062990","modelId":"glm-4.5","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":0,"warning":null},{"id":"fa89cbb630e45386ecb9a21b","modelId":"glm-4.5","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":78.1818181818182,"warning":null},{"id":"0a25e325728168d4811f7328","modelId":"glm-4.5","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":12.9749768303985,"warning":null},{"id":"53c7ad0bcc622084c5c97bac","modelId":"glm-4.5","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":44.0816326530612,"warning":null},{"id":"5252efc29a3f74f35b2cc18b","modelId":"glm-4.5","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":73.7566137566138,"warning":null},{"id":"fed2921be1ef626eaf2a2bbe","modelId":"glm-4.5","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":97.8666666666667,"warning":null},{"id":"dbbfc474edba92c7dab9b479","modelId":"glm-4.5","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":83.51063829787229,"warning":null},{"id":"f36782bceb8f45f46260015d","modelId":"glm-4.5","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":-27.3833333333333,"warning":null},{"id":"9532998fc69d20d9ec16b054","modelId":"glm-4.5","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":25.133333333333336,"warning":null},{"id":"08f28aaf925a9d0fe859921a","modelId":"glm-4.5","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":42.9824561403509,"warning":null},{"id":"370832ef81cc12e7336b2145","modelId":"glm-4.5","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-4.5 (Reasoning)","configuration":"GLM-4.5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4.5","score":21.969696969697,"warning":null},{"id":"30778ad113a0f6285db2b6ff","modelId":"glm-4.5-air","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.6666666666667,"warning":null},{"id":"450ec3814ad0983a05e7a649","modelId":"glm-4.5-air","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.6666666666667,"warning":null},{"id":"cf0dba8bcf1455f91c347f09","modelId":"glm-4.5-air","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"3d56ef2f2e608d5f1b411ba5","modelId":"glm-4.5-air","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.3333333333333,"warning":null},{"id":"1d3e677c3bf1025f2a641c60","modelId":"glm-4.5-air","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.043558850787771,"warning":null},{"id":"4b433f8594cf7a6275f6bdda","modelId":"glm-4.5-air","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.551020408163296,"warning":null},{"id":"e8cfae35f6575df2ef5f407a","modelId":"glm-4.5-air","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.3597883597884,"warning":null},{"id":"527bf2fab21343432d365235","modelId":"glm-4.5-air","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":96.5333333333333,"warning":null},{"id":"97020ae33cadfba84dafac57","modelId":"glm-4.5-air","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.4660904255319,"warning":null},{"id":"01d4601ce35baf7582b2b1d6","modelId":"glm-4.5-air","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":-61.5333333333333,"warning":null},{"id":"e7949f905f5354d088ef79b6","modelId":"glm-4.5-air","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.25,"warning":null},{"id":"eb25975b22bc9375e5cd191c","modelId":"glm-4.5-air","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.4912280701754,"warning":null},{"id":"dc0e4f99afd02c7db9b646f0","modelId":"glm-4.5-air","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-4.5-Air","configuration":"GLM-4.5-Air; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-4-5-air; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.4545454545455,"warning":null},{"id":"4871200d6165cb682b2f4db0","modelId":"glm-4.5v","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":15.333333333333298,"warning":null},{"id":"e4545c8e6a9f48f87e9183ab","modelId":"glm-4.5v","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":73,"warning":null},{"id":"0729665e03aab12509527098","modelId":"glm-4.5v","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":0,"warning":null},{"id":"922c35fbdd3776fc47267594","modelId":"glm-4.5v","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":0,"warning":null},{"id":"eca393b7c441fead8a16594f","modelId":"glm-4.5v","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":0,"warning":null},{"id":"f40e1888dda90b48a7c89c2d","modelId":"glm-4.5v","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":0,"warning":null},{"id":"1d03512bcc278ddfc0116961","modelId":"glm-4.5v","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":57.272727272727295,"warning":null},{"id":"e8adc6d5ef578a1953c26485","modelId":"glm-4.5v","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":68.38383838383841,"warning":null},{"id":"4dcf51bad5f69f32997356ae","modelId":"glm-4.5v","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":3.47544022242817,"warning":null},{"id":"03c90c119703b5139fdd86d2","modelId":"glm-4.5v","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":6.302131603336419,"warning":null},{"id":"713f47b76c588cdeed905e1a","modelId":"glm-4.5v","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":28.6394557823129,"warning":null},{"id":"7c2c5f83154687619618e3e1","modelId":"glm-4.5v","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":34.2176870748299,"warning":null},{"id":"c761f12b0a18361cd10bacb2","modelId":"glm-4.5v","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":35.2380952380952,"warning":null},{"id":"b524713fda2561eae7d1aa5f","modelId":"glm-4.5v","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":60.4232804232804,"warning":null},{"id":"f5091aafad1bbf696b1ec861","modelId":"glm-4.5v","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":75.0748005319149,"warning":null},{"id":"24ecfed1451beb78d27f9aad","modelId":"glm-4.5v","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":78.82313829787229,"warning":null},{"id":"35ffcd0393390b9865e927f4","modelId":"glm-4.5v","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":42.7745664739884,"warning":null},{"id":"09112bc5833f03e221eb9dcd","modelId":"glm-4.5v","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":50.4624277456647,"warning":null},{"id":"3448b6b298eb30f990513f92","modelId":"glm-4.5v","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":-55.5833333333333,"warning":null},{"id":"d6ed32645d4db9e47a014182","modelId":"glm-4.5v","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":-46.2833333333333,"warning":null},{"id":"b391e172392ad37ac3d246b5","modelId":"glm-4.5v","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":18.21666666666667,"warning":null},{"id":"bcd773de3a0b8d695f5c99e9","modelId":"glm-4.5v","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":20.833333333333336,"warning":null},{"id":"089e4a139e5493e3585dcc44","modelId":"glm-4.5v","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":19.5906432748538,"warning":null},{"id":"e34271c056b08e75014f0dbd","modelId":"glm-4.5v","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":22.5146198830409,"warning":null},{"id":"4503f7cd70608a6263c8a831","modelId":"glm-4.5v","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GLM-4.5V (Non-reasoning)","configuration":"GLM-4.5V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v","score":6.81818181818182,"warning":null},{"id":"ba4b24c2832b05c124a17fb1","modelId":"glm-4.5v","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-4.5V (Reasoning)","configuration":"GLM-4.5V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-5v-reasoning","score":5.3030303030303,"warning":null},{"id":"94b2207ac99b20596113e239","modelId":"glm-4.6","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":44.3333333333333,"warning":null},{"id":"4c0bdeccb0c354ff92adf063","modelId":"glm-4.6","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":86,"warning":null},{"id":"6f35e6b7857e8f7192a4a601","modelId":"glm-4.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":26.333333333333297,"warning":null},{"id":"0344a85a899d64b5d461c992","modelId":"glm-4.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":54,"warning":null},{"id":"315001284817defdcd52844a","modelId":"glm-4.6","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":0,"warning":null},{"id":"fed9d2112a78902f9f4f0e51","modelId":"glm-4.6","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":1.1428571428571401,"warning":null},{"id":"c67f2a6f971536ea02be034c","modelId":"glm-4.6","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":757.12,"warning":null},{"id":"1bae8321f9546badb45d0fb9","modelId":"glm-4.6","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":85.56250000000001,"warning":null},{"id":"f82dc3cb8c29059c143578fe","modelId":"glm-4.6","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":63.232323232323196,"warning":null},{"id":"ccadcb53ecb9ac27bab7aeb6","modelId":"glm-4.6","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":77.97979797979801,"warning":null},{"id":"6d3bc66cda454029e3c6020a","modelId":"glm-4.6","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":5.4680259499536605,"warning":null},{"id":"3b40312601b47b4babec1ab4","modelId":"glm-4.6","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":14.457831325301198,"warning":null},{"id":"ebea8e8659f63d3cc4eee8fd","modelId":"glm-4.6","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":36.734693877551,"warning":null},{"id":"3fb0125cf298b473640c0aec","modelId":"glm-4.6","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":43.401360544217695,"warning":null},{"id":"40fe745d4d40c37a7d37481e","modelId":"glm-4.6","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":56.0846560846561,"warning":null},{"id":"c88494c785e24d0e1cdaf51f","modelId":"glm-4.6","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":69.5238095238095,"warning":null},{"id":"a3c0ba45f30db91ea4c183c1","modelId":"glm-4.6","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":78.407579787234,"warning":null},{"id":"1db53b8e924e88e0152b0bca","modelId":"glm-4.6","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":82.920545212766,"warning":null},{"id":"34147f34008b905a9aaccb96","modelId":"glm-4.6","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":-31.6833333333333,"warning":null},{"id":"ed586d5fb6e4774692dbe3d6","modelId":"glm-4.6","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":-41.8833333333333,"warning":null},{"id":"9c2e61e1aa9164545b79fcd2","modelId":"glm-4.6","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":21.416666666666668,"warning":null},{"id":"c53213ee6f0915802c451959","modelId":"glm-4.6","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":26.883333333333333,"warning":null},{"id":"d316796d0ffdebfc5767c923","modelId":"glm-4.6","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":76.9005847953216,"warning":null},{"id":"9199e4208a43755ee68b1b8e","modelId":"glm-4.6","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":70.4678362573099,"warning":null},{"id":"5bda5b832f8326e62befe8f7","modelId":"glm-4.6","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":13.4020618556701,"warning":null},{"id":"d02b87f4d69acd7535e38333","modelId":"glm-4.6","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GLM-4.6 (Non-reasoning)","configuration":"GLM-4.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6","score":28.7878787878788,"warning":null},{"id":"2b53925946f0f877aafc59ae","modelId":"glm-4.6","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":25,"warning":null},{"id":"2f27314674aef5a4b49f2640","modelId":"glm-4.6","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"GLM-4.6 (Reasoning)","configuration":"GLM-4.6 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6-reasoning","score":49.438202247191,"warning":null},{"id":"4928b0276fd6c9a8c444b71e","modelId":"glm-4.6v","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":26.333333333333297,"warning":null},{"id":"837343c2ac3011d1ba300da8","modelId":"glm-4.6v","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":85.3333333333333,"warning":null},{"id":"e5e6ec6c228c7287c39d279f","modelId":"glm-4.6v","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":17,"warning":null},{"id":"d141580c3462963d595336e2","modelId":"glm-4.6v","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":48.6666666666667,"warning":null},{"id":"3905dcd5f1aeeae6a2266f50","modelId":"glm-4.6v","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":0,"warning":null},{"id":"afd55634a947e6c3807e9226","modelId":"glm-4.6v","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":0,"warning":null},{"id":"eba8c8cdf1f756b264cfd259","modelId":"glm-4.6v","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":74.34375,"warning":null},{"id":"7122a72135860cf4a9ad81f7","modelId":"glm-4.6v","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":80.62500000000001,"warning":null},{"id":"e7fd67d8815e3bbfa5923edc","modelId":"glm-4.6v","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":56.565656565656596,"warning":null},{"id":"ba3ca764bdc9824f995c4087","modelId":"glm-4.6v","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":71.9191919191919,"warning":null},{"id":"c68005b6921e3ff990365ab4","modelId":"glm-4.6v","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":3.66079703429101,"warning":null},{"id":"ecd415b5cced2b02caaf3681","modelId":"glm-4.6v","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":9.63855421686747,"warning":null},{"id":"c716d0178a13ce9a95ae3cd0","modelId":"glm-4.6v","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":27.891156462585,"warning":null},{"id":"0421e7c6e87d6634872e2e40","modelId":"glm-4.6v","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":30.068027210884402,"warning":null},{"id":"7d7305f2366a230af2f05859","modelId":"glm-4.6v","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":41.0582010582011,"warning":null},{"id":"7c7927de9636469ac8d473a7","modelId":"glm-4.6v","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":15.978835978836,"warning":null},{"id":"384accf5ced6dc13a246274e","modelId":"glm-4.6v","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":75.1579122340426,"warning":null},{"id":"fdb7e30eb45da4db40e78049","modelId":"glm-4.6v","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":79.8869680851064,"warning":null},{"id":"c89a56b6d6bbbcefc7ae2df1","modelId":"glm-4.6v","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":42.196531791907496,"warning":null},{"id":"2827544ddf30cfce3f005022","modelId":"glm-4.6v","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":48.5549132947977,"warning":null},{"id":"39a8e6640c8191a779ffb48c","modelId":"glm-4.6v","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":-37.6333333333333,"warning":null},{"id":"6fddb525b5d0b94467ba8661","modelId":"glm-4.6v","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":-26.95,"warning":null},{"id":"e1cded7b38ba045706848908","modelId":"glm-4.6v","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":17.4,"warning":null},{"id":"b6ce561dc26c02cc5b8f2b40","modelId":"glm-4.6v","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":16.183333333333334,"warning":null},{"id":"7cfeb8b11e9274ce0f4325c1","modelId":"glm-4.6v","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":30.701754385964904,"warning":null},{"id":"52cdb947037479b540381f2c","modelId":"glm-4.6v","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":31.5789473684211,"warning":null},{"id":"56681bacf5ef1774b2acc379","modelId":"glm-4.6v","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GLM-4.6V (Non-reasoning)","configuration":"GLM-4.6V (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v","score":3.03030303030303,"warning":null},{"id":"35b1c4bf5d6fa8c5baaf36b7","modelId":"glm-4.6v","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-4.6V (Reasoning)","configuration":"GLM-4.6V (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-6v-reasoning","score":14.3939393939394,"warning":null},{"id":"0771a4f46146e10e4235d642","modelId":"glm-4.7","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":48,"warning":null},{"id":"9d958d31de170d1f70fe36e3","modelId":"glm-4.7","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":95,"warning":null},{"id":"2e0855d0f0f6762f6c2ffdef","modelId":"glm-4.7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":40.6666666666667,"warning":null},{"id":"600cd8ff296dbf4c45f8dfb1","modelId":"glm-4.7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":71,"warning":null},{"id":"be2db485b78d0a1b46ed5356","modelId":"glm-4.7","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":0,"warning":null},{"id":"7c1a7f1baae339fd5be9c307","modelId":"glm-4.7","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":1.7102040816326498,"warning":null},{"id":"6e26a11771f840576d856325","modelId":"glm-4.7","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":1014.22,"warning":null},{"id":"5bc781f14b12760c914bf4ff","modelId":"glm-4.7","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":79.94270833333333,"warning":null},{"id":"d361191da2645058d234a667","modelId":"glm-4.7","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":66.3636363636364,"warning":null},{"id":"e6b2c07688573b5ef356beb2","modelId":"glm-4.7","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":85.8585858585858,"warning":null},{"id":"f2ab30db0772c1973993e6ed","modelId":"glm-4.7","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":6.39481000926784,"warning":null},{"id":"b20d65229b0b347c90b78a5c","modelId":"glm-4.7","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":27.386468952734,"warning":null},{"id":"580c7737090958a16068b38c","modelId":"glm-4.7","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":54.6258503401361,"warning":null},{"id":"bf52da1f53cc20f4500101b1","modelId":"glm-4.7","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":67.891156462585,"warning":null},{"id":"6382cf43b3a8232f5674c7b0","modelId":"glm-4.7","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":56.190476190476204,"warning":null},{"id":"1e4ba8a5e58f7505da22c681","modelId":"glm-4.7","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":89.41798941798939,"warning":null},{"id":"6011d3f6af7d6023a6300e86","modelId":"glm-4.7","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":79.42985372340429,"warning":null},{"id":"6ab0834a7bf56a40feab81d4","modelId":"glm-4.7","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":85.61336436170211,"warning":null},{"id":"99a720d42d9ab77eb33f8999","modelId":"glm-4.7","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":-47.3166666666667,"warning":null},{"id":"4e7b558ad298402697e6731f","modelId":"glm-4.7","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":-36.4333333333333,"warning":null},{"id":"b5f9786af6208eb41d64f34a","modelId":"glm-4.7","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":23.633333333333333,"warning":null},{"id":"c2a7e666a331885b7d3dc6b5","modelId":"glm-4.7","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":29.31666666666667,"warning":null},{"id":"1df34276ac9f506932188cd2","modelId":"glm-4.7","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":94.15204678362569,"warning":null},{"id":"66a84bbe9d4d9beec66aa404","modelId":"glm-4.7","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":95.906432748538,"warning":null},{"id":"2c709f287f79e8409cab1324","modelId":"glm-4.7","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":12.1649484536082,"warning":null},{"id":"ed118dd9b03ad11a94dbc60a","modelId":"glm-4.7","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GLM-4.7 (Non-reasoning)","configuration":"GLM-4.7 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-non-reasoning","score":30.303030303030297,"warning":null},{"id":"fcf8a79a09b5a2fbc1bf7633","modelId":"glm-4.7","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":31.818181818181802,"warning":null},{"id":"c0a612c35973a4492ab95c32","modelId":"glm-4.7","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"GLM-4.7 (Reasoning)","configuration":"GLM-4.7 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7","score":45.3183520599251,"warning":null},{"id":"9cd24111c9965cfb6c1082ac","modelId":"glm-4.7-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":20.3333333333333,"warning":null},{"id":"cd1295cb2630e66e6f215155","modelId":"glm-4.7-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":41.6666666666667,"warning":null},{"id":"ed9425416741e32642acdd40","modelId":"glm-4.7-flash","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":0,"warning":null},{"id":"0db7ae079fcd8b12ea4ae666","modelId":"glm-4.7-flash","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":0.28571428571428603,"warning":null},{"id":"7d0bb65c43e4e65421db7974","modelId":"glm-4.7-flash","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":45.1515151515151,"warning":null},{"id":"1c6b02da06598a9564b5f0a1","modelId":"glm-4.7-flash","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":58.0808080808081,"warning":null},{"id":"64779412a942295b2f1fcc41","modelId":"glm-4.7-flash","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":4.95829471733086,"warning":null},{"id":"f063c4ae2b226ece4cb36abb","modelId":"glm-4.7-flash","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":7.5996292863762696,"warning":null},{"id":"9dd6590fc2f380b2c4012ea3","modelId":"glm-4.7-flash","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":46.2585034013605,"warning":null},{"id":"4c84ae887b9f12258846745d","modelId":"glm-4.7-flash","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":60.816326530612194,"warning":null},{"id":"a401455d8af56dc225e840c7","modelId":"glm-4.7-flash","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":-68.9333333333333,"warning":null},{"id":"7ce9b77baf9eb443d943cfaf","modelId":"glm-4.7-flash","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":-62.6166666666667,"warning":null},{"id":"b9f06013867513c4304ef9e3","modelId":"glm-4.7-flash","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":13.05,"warning":null},{"id":"4f604a4640d84fc42eceed08","modelId":"glm-4.7-flash","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":16.150000000000002,"warning":null},{"id":"6e4ef0923f16cd949083876d","modelId":"glm-4.7-flash","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":91.812865497076,"warning":null},{"id":"70b12115dcee433abbc02722","modelId":"glm-4.7-flash","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":98.8304093567251,"warning":null},{"id":"12babe8e33433fd23f2ac171","modelId":"glm-4.7-flash","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GLM-4.7-Flash (Non-reasoning)","configuration":"GLM-4.7-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash-non-reasoning","score":3.78787878787879,"warning":null},{"id":"29f7ca5fdc620111b0d18ca8","modelId":"glm-4.7-flash","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-4.7-Flash (Reasoning)","configuration":"GLM-4.7-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/glm-4-7-flash","score":21.969696969697,"warning":null},{"id":"371e32e413b5c1e2b63024dc","modelId":"glm-5","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":14.454277286135694,"warning":null},{"id":"3428cabd424a2de0154d82fe","modelId":"glm-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":43.6666666666667,"warning":null},{"id":"fe7a9aae7813ba9b022e0a72","modelId":"glm-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":75.6666666666667,"warning":null},{"id":"08067a5c23ad9c568ec00812","modelId":"glm-5","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":0,"warning":null},{"id":"ae84ac005a4fdeda6ace0f05","modelId":"glm-5","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":2,"warning":null},{"id":"498eceb3c7be3ce505460dd0","modelId":"glm-5","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":81.91666666666666,"warning":null},{"id":"d7755d2a416f32e67386047d","modelId":"glm-5","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":66.5656565656566,"warning":null},{"id":"b9a1cabde7a1e7348cc33f64","modelId":"glm-5","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":82.020202020202,"warning":null},{"id":"16de6f0d8d08fe75b8727d3a","modelId":"glm-5","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":7.645968489341979,"warning":null},{"id":"3fa3e46f37c7db7f5e084225","modelId":"glm-5","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":29.286376274328102,"warning":null},{"id":"13e503950411783d5ec9b173","modelId":"glm-5","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":55.170068027210895,"warning":null},{"id":"b2025179161cc4f662dcc530","modelId":"glm-5","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":72.312925170068,"warning":null},{"id":"7ee2916f2a458729d0bf1b6b","modelId":"glm-5","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":-11.8,"warning":null},{"id":"9d54baf80501bc9f77e431c4","modelId":"glm-5","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":0.266666666666667,"warning":null},{"id":"5a3b5525815de32a23e559dd","modelId":"glm-5","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":23.05,"warning":null},{"id":"8c3a2030b1af48fb9ae3dde5","modelId":"glm-5","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":26.3,"warning":null},{"id":"9229265b86c1968b93d86775","modelId":"glm-5","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":97.3684210526316,"warning":null},{"id":"7cbdf43c125960c2074118eb","modelId":"glm-5","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":98.2456140350877,"warning":null},{"id":"ad5cb316707e497431201b78","modelId":"glm-5","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GLM-5 (Non-reasoning)","configuration":"GLM-5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-non-reasoning","score":39.3939393939394,"warning":null},{"id":"2e99d17495fc238ef6244f56","modelId":"glm-5","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-5 (Reasoning)","configuration":"GLM-5 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5","score":43.1818181818182,"warning":null},{"id":"1f452159e71bb7cf6c9df55c","modelId":"glm-5-turbo","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.6666666666667,"warning":null},{"id":"38fd5b20ee074591cd6ee178","modelId":"glm-5-turbo","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"82f973f0bcc93805188086f9","modelId":"glm-5-turbo","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.74747474747471,"warning":null},{"id":"4768564e465e1b47587ec2d7","modelId":"glm-5-turbo","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.757182576459698,"warning":null},{"id":"6da62a2c7f9a3a460475f9f5","modelId":"glm-5-turbo","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.1972789115646,"warning":null},{"id":"0bdaf0c08a9f5d6437f640a7","modelId":"glm-5-turbo","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":-16.3666666666667,"warning":null},{"id":"0a4a8c34851842bb451eabc3","modelId":"glm-5-turbo","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.449999999999996,"warning":null},{"id":"776868e5b235788549bdbee8","modelId":"glm-5-turbo","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":98.5380116959064,"warning":null},{"id":"556b2cb34128790508b0a194","modelId":"glm-5-turbo","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-5-Turbo","configuration":"GLM-5-Turbo; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.3333333333333,"warning":null},{"id":"565b39e1026432a05c278d39","modelId":"glm-5.1","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":964.54,"warning":null},{"id":"c39102d86367d9fb4babc11b","modelId":"glm-5.1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":53.3333333333333,"warning":null},{"id":"595338c539b2c4564f1d642c","modelId":"glm-5.1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":73.6666666666667,"warning":null},{"id":"fc73ee323b1e28e5bc9197ae","modelId":"glm-5.1","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":20.308354687857076,"warning":null},{"id":"30a13e7c9545cc95c3213936","modelId":"glm-5.1","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":0,"warning":null},{"id":"fc8df65d5ad38aaa1f26ee2c","modelId":"glm-5.1","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":4.57142857142857,"warning":null},{"id":"f5c0ac6e2cf24335a4aafeb8","modelId":"glm-5.1","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":8.4,"warning":null},{"id":"f52996f64fec4af28a0ff2d1","modelId":"glm-5.1","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":1116.24,"warning":null},{"id":"3af2e846ff223b71f85e5654","modelId":"glm-5.1","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":83.93939393939391,"warning":null},{"id":"54f0aef98b6d8f147c847061","modelId":"glm-5.1","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":86.7676767676768,"warning":null},{"id":"bdf6dc6e16a18d863fbad514","modelId":"glm-5.1","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":27.9425393883225,"warning":null},{"id":"ae183160f013826b3737ac2c","modelId":"glm-5.1","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":30.0741427247451,"warning":null},{"id":"9c83698c69eebf7abc6a73de","modelId":"glm-5.1","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":51.9727891156463,"warning":null},{"id":"485e3cc25dc48e1a9389fc9a","modelId":"glm-5.1","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":76.2585034013605,"warning":null},{"id":"4fb343bd4b97cb79826e7420","modelId":"glm-5.1","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":40.254237288135606,"warning":null},{"id":"d5b509e455ea18ca2e688dc8","modelId":"glm-5.1","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":-22.4333333333333,"warning":null},{"id":"7fdcae53e99c38cc6d93e736","modelId":"glm-5.1","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":0.85,"warning":null},{"id":"e4cb9ffa9390c3c6084fbc6a","modelId":"glm-5.1","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":25.183333333333337,"warning":null},{"id":"29edc738b17c43dc468be90f","modelId":"glm-5.1","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":23.7,"warning":null},{"id":"751a38b92383dd44f4f4b6f8","modelId":"glm-5.1","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":44.7916666666667,"warning":null},{"id":"49be3f22ea03e85c39c33f83","modelId":"glm-5.1","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":97.07602339181291,"warning":null},{"id":"2b4a47e946d0bad8e61425fe","modelId":"glm-5.1","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":97.6608187134503,"warning":null},{"id":"824624d738ab3059a6e8fbd8","modelId":"glm-5.1","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":13.6082474226804,"warning":null},{"id":"f1da9a555bd7b230511bbc67","modelId":"glm-5.1","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GLM-5.1 (Non-reasoning)","configuration":"GLM-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1-non-reasoning","score":35.6060606060606,"warning":null},{"id":"a6d79b63b7974c406cbce5a0","modelId":"glm-5.1","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":43.1818181818182,"warning":null},{"id":"407c1f4b591ee717a0225582","modelId":"glm-5.1","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":61.797752808988804,"warning":null},{"id":"4942502e2f28e870df1cfb21","modelId":"glm-5.1","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"GLM-5.1 (Reasoning)","configuration":"GLM-5.1 (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5-1","score":2.02020202020202,"warning":null},{"id":"b5077a6fd8bbb23f0cf30638","modelId":"glm-5.1","benchmarkId":"terminal-claudecode","level":"max","sourceLabel":"GLM-5.1","configuration":"GLM-5.1; Reported effort: max; 445 trials; recorded 2026-03-27","score":58.65,"warning":null},{"id":"fa12ffdb02b55783f75edfee","modelId":"glm-5.2","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":1228.86,"warning":null},{"id":"33d9fb395613087a301f2a61","modelId":"glm-5.2","benchmarkId":"aa-apex-agents-aa","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":33.7020648967552,"warning":null},{"id":"429d0a99aeabac40020e989b","modelId":"glm-5.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":78.3333333333333,"warning":null},{"id":"e94927ce7fbb41be1446de36","modelId":"glm-5.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":42.3333333333333,"warning":null},{"id":"04f82f716e1b5d30b1617a89","modelId":"glm-5.2","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":28.398977210054348,"warning":null},{"id":"336e74c6150f0f5ace5c55bc","modelId":"glm-5.2","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":20.8571428571429,"warning":null},{"id":"509692cc7e198e1eba713675","modelId":"glm-5.2","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":3.14285714285714,"warning":null},{"id":"0755c822a36d26ca1261696e","modelId":"glm-5.2","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":42.733512384362875,"warning":null},{"id":"d3d3f788d3369ce5160384f4","modelId":"glm-5.2","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":10.4,"warning":null},{"id":"d3f3517ac6a7b0942ef72054","modelId":"glm-5.2","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":1372.01,"warning":null},{"id":"a4d13d20bb5c97fe673d5900","modelId":"glm-5.2","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":1247.03,"warning":null},{"id":"77fea71a1ce54ced8e0ca8eb","modelId":"glm-5.2","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":89.49494949494951,"warning":null},{"id":"3a6e78ca66808b560219a6ef","modelId":"glm-5.2","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":68.58585858585859,"warning":null},{"id":"a9d17567e1c3bc69ec048824","modelId":"glm-5.2","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":41.1492122335496,"warning":null},{"id":"5734fedef2340bfa5ea45ecc","modelId":"glm-5.2","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":9.7775718257646,"warning":null},{"id":"bf3dedc20e083f8dbc3e72e3","modelId":"glm-5.2","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":73.3333333333333,"warning":null},{"id":"21b1e9df4380160bfd474b8c","modelId":"glm-5.2","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":42.6553672316384,"warning":null},{"id":"b0d037f207d2790141ce7041","modelId":"glm-5.2","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":7.2222222222222205,"warning":null},{"id":"8b1cd09ce3680fd3ede877c5","modelId":"glm-5.2","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":4.43333333333333,"warning":null},{"id":"d905bff8aaed379cbae34a55","modelId":"glm-5.2","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":-6.56666666666667,"warning":null},{"id":"50877e213c6cc4dddaf95446","modelId":"glm-5.2","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":24.333333333333336,"warning":null},{"id":"18c771702d3ba14c5fc02b8a","modelId":"glm-5.2","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":20.283333333333335,"warning":null},{"id":"c5c56d14c19f67f323d70f3d","modelId":"glm-5.2","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":51.1574074074074,"warning":null},{"id":"e157bbf9c740b15105665bd6","modelId":"glm-5.2","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":99.1228070175439,"warning":null},{"id":"9124066006e0baf36525ed0a","modelId":"glm-5.2","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":34.639175257732,"warning":null},{"id":"d298d6a3d4fc3a4d17703a24","modelId":"glm-5.2","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":16.7010309278351,"warning":null},{"id":"97f371340132c102563885cc","modelId":"glm-5.2","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":50.7575757575758,"warning":null},{"id":"12292e8fc20951ebf06da478","modelId":"glm-5.2","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":77.9026217228464,"warning":null},{"id":"1714719aac77d33cf0b3aa16","modelId":"glm-5.2","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GLM-5.2 (Non-reasoning)","configuration":"GLM-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/glm-5-2-non-reasoning","score":51.685393258427005,"warning":null},{"id":"a9db7e91ffc6cd912a8ca26d","modelId":"glm-5.2","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GLM-5.2 (Max)","configuration":"GLM-5.2 (Max); Model source: https://artificialanalysis.ai/models/glm-5-2","score":1.01010101010101,"warning":null},{"id":"41fb67eca9e0c6af98ac79d1","modelId":"glm-5.2","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"glm-5-2","configuration":"glm-5-2; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":36.3,"warning":null},{"id":"036eacdadd55f02cf88f8c65","modelId":"glm-5.2","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"glm-5-2","configuration":"glm-5-2; Reported effort: max; 113 unique tasks; 450 attempts; 4 runs","score":43.8,"warning":null},{"id":"893b27f62d0e7cffc882bf5b","modelId":"glm-5.3","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":38.75,"warning":null},{"id":"21717bd0de21523f12772afc","modelId":"glm-5.3","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":1186.83,"warning":null},{"id":"884f0f7e8e72e6e4bfc87f38","modelId":"glm-5.3","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":1506.86,"warning":null},{"id":"1018283acf15b3df50dce77d","modelId":"glm-5.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":72.6666666666667,"warning":null},{"id":"6acdedcfd5d8a5df2ba34d43","modelId":"glm-5.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":79.6666666666667,"warning":null},{"id":"0522a5197833d22074f01b11","modelId":"glm-5.3","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":51.979098375134605,"warning":null},{"id":"893ee084f931c82904e2cac9","modelId":"glm-5.3","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":62.202864996264196,"warning":null},{"id":"171d2e9e09a1a7920241f75e","modelId":"glm-5.3","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":14.5714285714286,"warning":null},{"id":"2817b1c263e6d7fb9ff6832a","modelId":"glm-5.3","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":19.1428571428571,"warning":null},{"id":"c59712c22da5ee10b4c63c5d","modelId":"glm-5.3","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":36.436884512085946,"warning":null},{"id":"3100db71ce65024e3f451662","modelId":"glm-5.3","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":8.799999999999999,"warning":null},{"id":"2ddb00dc9161fd272729ea2e","modelId":"glm-5.3","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":11.200000000000001,"warning":null},{"id":"8f8cadacb6fad186ed2685d0","modelId":"glm-5.3","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":1322.45,"warning":null},{"id":"dd7ea45d6e5db710738799d0","modelId":"glm-5.3","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":1650.31,"warning":null},{"id":"76d349f13fedfc8108cf21a8","modelId":"glm-5.3","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":91.7171717171717,"warning":null},{"id":"aa53bb1266c5e3247086fd0b","modelId":"glm-5.3","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":0.27777777777777773,"warning":null},{"id":"5fb9fe13bad3f5f8f4850661","modelId":"glm-5.3","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":36.5616311399444,"warning":null},{"id":"1740408b871303a0dc607cd7","modelId":"glm-5.3","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":42.2613531047266,"warning":null},{"id":"1003cf4c06eed62ea3c37db0","modelId":"glm-5.3","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":46.0828625235405,"warning":null},{"id":"c73e3b5f0a9da381f3eecb18","modelId":"glm-5.3","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":48.3333333333333,"warning":null},{"id":"4cf17b3752920a22bddf0931","modelId":"glm-5.3","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":-8.4,"warning":null},{"id":"c3d5b20dd24762f5952922e6","modelId":"glm-5.3","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":14.3,"warning":null},{"id":"ae37d8ee54d5d519c4982b78","modelId":"glm-5.3","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":33.916666666666664,"warning":null},{"id":"917e799f2f5dd97ec426c7a9","modelId":"glm-5.3","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":33.85,"warning":null},{"id":"c6d35b78e0277163670c59a7","modelId":"glm-5.3","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":42.0138888888889,"warning":null},{"id":"ce9d876bde1429ca9f36479c","modelId":"glm-5.3","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":59.0277777777778,"warning":null},{"id":"0b780bfa6334abad3028cd21","modelId":"glm-5.3","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":50.3092783505155,"warning":null},{"id":"305e5adf1836cc1346a3d6e0","modelId":"glm-5.3","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":9.52380952380952,"warning":null},{"id":"c98307775902e958dc99fba0","modelId":"glm-5.3","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":83.8951310861423,"warning":null},{"id":"e490da5b429a10309a8e228b","modelId":"glm-5.3","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GLM-5.3 (Low)","configuration":"GLM-5.3 (Low); Model source: https://artificialanalysis.ai/models/glm-5-3-low","score":34.8484848484849,"warning":null},{"id":"396b09dda11d0652428aa763","modelId":"glm-5.3","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"GLM-5.3 (Max); Model source: https://artificialanalysis.ai/models/glm-5-3","score":41.9191919191919,"warning":null},{"id":"9e2a553a0ef9312c689f04ed","modelId":"glm-5.3","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"glm-5-3","configuration":"glm-5-3; Reported effort: max; 113 unique tasks; 451 attempts; 4 runs","score":69,"warning":null},{"id":"f88a425fff09825196a737c6","modelId":"glm-5.3-flash","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":25,"warning":null},{"id":"d8e1ad32ca42028aeb086c2b","modelId":"glm-5.3-flash","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":1453,"warning":null},{"id":"2edf5c6a680fd2d04a804ddf","modelId":"glm-5.3-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":80,"warning":null},{"id":"451a12a824fab4ecfd7a24e5","modelId":"glm-5.3-flash","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":60.36862782167782,"warning":null},{"id":"2e29f3fa7a4442d420295d70","modelId":"glm-5.3-flash","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.4285714285714,"warning":null},{"id":"7d01a0db6aef0175ea0ebcc2","modelId":"glm-5.3-flash","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.21396598030439,"warning":null},{"id":"c84d1f50e5616dea725fb651","modelId":"glm-5.3-flash","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.4,"warning":null},{"id":"f61eb8bca2b2e3b26276d83d","modelId":"glm-5.3-flash","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":1642.18,"warning":null},{"id":"bcadd9e7664a3739734c4683","modelId":"glm-5.3-flash","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":91.2121212121212,"warning":null},{"id":"245897d23e4c162d31275f46","modelId":"glm-5.3-flash","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.8333333333333334,"warning":null},{"id":"14bbcc046d8a9fe05aa2991a","modelId":"glm-5.3-flash","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.8517145505097,"warning":null},{"id":"25b0988a21d522d1125c6094","modelId":"glm-5.3-flash","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.24293785310739,"warning":null},{"id":"18dfc980033586824c069087","modelId":"glm-5.3-flash","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.11111111111109,"warning":null},{"id":"ed40864bcdafb5eb329dcd23","modelId":"glm-5.3-flash","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.46666666666667,"warning":null},{"id":"9bdb6cca3acc7ef5de99c107","modelId":"glm-5.3-flash","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.500000000000004,"warning":null},{"id":"e1954fd811496a176698b207","modelId":"glm-5.3-flash","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.6203703703704,"warning":null},{"id":"8e4e0f16955a0fdb1d162c5f","modelId":"glm-5.3-flash","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.2164948453608,"warning":null},{"id":"b22eb85ee574455ca3191b6c","modelId":"glm-5.3-flash","benchmarkId":"aa-terminal-bench-science","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.76190476190476,"warning":null},{"id":"3f2e3da28b530a9beae28ba4","modelId":"glm-5.3-flash","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.2696629213483,"warning":null},{"id":"7d247681ca8d8ade3d4bf48a","modelId":"glm-5.3-flash","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"GLM-5.3-Flash","configuration":"GLM-5.3-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/glm-5-3-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.8282828282828,"warning":null},{"id":"f0dcb72f5917162e78faffb2","modelId":"glm-5.3-flash","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GLM-5.3-Flash (High)","configuration":"GLM-5.3-Flash (High); Source model ID: zai-glm-5-3-flash-high; 120 tasks","score":50.138888888888886,"warning":null},{"id":"ad8f5a7ae13ab8c586359be2","modelId":"glm-5.3-flash","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GLM-5.3-Flash (Low)","configuration":"GLM-5.3-Flash (Low); Source model ID: zai-glm-5-3-flash-low; 120 tasks","score":27.916666666666668,"warning":null},{"id":"d2a82e0d42cfe5fbe120beb5","modelId":"glm-5.3-flash","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GLM-5.3-Flash (Max)","configuration":"GLM-5.3-Flash (Max); Source model ID: zai-glm-5-3-flash-max; 120 tasks","score":65.83333333333333,"warning":null},{"id":"ea4b0ccbfe7ece48ee812e8b","modelId":"glm-5.3-flash","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"glm-5-3-flash","configuration":"glm-5-3-flash; Reported effort: max; 113 unique tasks; 448 attempts; 4 runs","score":63.4,"warning":null},{"id":"f6783606396b8bce8ac51b24","modelId":"glm-5v-turbo","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":70.3333333333333,"warning":null},{"id":"12b6566ca4ca6c13138f9e96","modelId":"glm-5v-turbo","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":0.571428571428571,"warning":null},{"id":"44ad128942762572d234bdd2","modelId":"glm-5v-turbo","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":80.90909090909089,"warning":null},{"id":"16870d51f189414eb5dd849d","modelId":"glm-5v-turbo","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":17.1455050973123,"warning":null},{"id":"37982711b6c81550d0bf0ced","modelId":"glm-5v-turbo","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":61.0884353741497,"warning":null},{"id":"0bac296b6bcdc867575332aa","modelId":"glm-5v-turbo","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":72.7745664739884,"warning":null},{"id":"6f7ad24f8763d179c05b02ec","modelId":"glm-5v-turbo","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":-19.2833333333333,"warning":null},{"id":"4d967b270cc1b306d32a8d0a","modelId":"glm-5v-turbo","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":29.333333333333332,"warning":null},{"id":"8fbfbcae27f2a663fec6ea03","modelId":"glm-5v-turbo","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":98.5380116959064,"warning":null},{"id":"919e117fe8b99db2a9b0924f","modelId":"glm-5v-turbo","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"GLM 5V Turbo (Reasoning)","configuration":"GLM 5V Turbo (Reasoning); Model source: https://artificialanalysis.ai/models/glm-5v-turbo","score":32.5757575757576,"warning":null},{"id":"e86dccdbe57cd031461b162a","modelId":"gpt-3.5-turbo","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-3.5 Turbo","configuration":"GPT-3.5 Turbo; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-35-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.6969696969697,"warning":null},{"id":"7b7b5a85ad55f9f5265d61d4","modelId":"gpt-3.5-turbo","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-3.5 Turbo","configuration":"GPT-3.5 Turbo; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-35-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.0666666666667,"warning":null},{"id":"b4572ee6f0b5d5ed597a0caf","modelId":"gpt-3.5-turbo","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-3.5 Turbo","configuration":"GPT-3.5 Turbo; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-35-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.1851728723404,"warning":null},{"id":"31ad92730d85a268e23aea8c","modelId":"gpt-4","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-4","configuration":"GPT-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":-19.88,"warning":null},{"id":"33dcba80de235a7178254ae4","modelId":"gpt-4","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4","configuration":"GPT-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.949494949494905,"warning":null},{"id":"038ce8230ca4b6eb3d669d9a","modelId":"gpt-4","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-4","configuration":"GPT-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.197278911564595,"warning":null},{"id":"5d475d04cc97cb96522544e6","modelId":"gpt-4","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4","configuration":"GPT-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":56.8,"warning":null},{"id":"6a974fcb30a93f5f4469e5f6","modelId":"gpt-4","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4","configuration":"GPT-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":56.208444148936195,"warning":null},{"id":"1f1a98c6041a0347cedc067b","modelId":"gpt-4","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-4","configuration":"GPT-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":-33.8333333333333,"warning":null},{"id":"9fe1f5fef2e00423fce3cdba","modelId":"gpt-4","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-4","configuration":"GPT-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":21,"warning":null},{"id":"f633b74d416f6bd43ad13675","modelId":"gpt-4-turbo","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4 Turbo","configuration":"GPT-4 Turbo; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.1300000000000003,"warning":null},{"id":"9560dd4fd1468931e9052113","modelId":"gpt-4-turbo","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4 Turbo","configuration":"GPT-4 Turbo; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.100529100529098,"warning":null},{"id":"6ffe256a0e1313ed70520f43","modelId":"gpt-4-turbo","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4 Turbo","configuration":"GPT-4 Turbo; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.6666666666667,"warning":null},{"id":"02bbce731291738c87436522","modelId":"gpt-4-turbo","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4 Turbo","configuration":"GPT-4 Turbo; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-turbo; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.3733377659574,"warning":null},{"id":"71e63379395e6cbc0d8af312","modelId":"gpt-4.1","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.6666666666667,"warning":null},{"id":"6258690b3586212c0ba891ed","modelId":"gpt-4.1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.3333333333333,"warning":null},{"id":"62391a002947467a08c83e49","modelId":"gpt-4.1","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"6341242383674a88e3732cb2","modelId":"gpt-4.1","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.5656565656566,"warning":null},{"id":"08a0c963b233621f21dd0ccb","modelId":"gpt-4.1","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.18,"warning":null},{"id":"4d67dd78b382a9e804734322","modelId":"gpt-4.1","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.9931972789116,"warning":null},{"id":"615f90985d1bd979b83a9644","modelId":"gpt-4.1","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":45.7142857142857,"warning":null},{"id":"fbd303abbd8215370445d77e","modelId":"gpt-4.1","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":91.2666666666667,"warning":null},{"id":"cda7d9aee07435b67e05e255","modelId":"gpt-4.1","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.61835106382979,"warning":null},{"id":"babd604f90f1908b5bafa26a","modelId":"gpt-4.1","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":61.2138728323699,"warning":null},{"id":"516b003dce56d5a7659fc6e1","modelId":"gpt-4.1","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":-39.6333333333333,"warning":null},{"id":"45f0a1fb80d261d7173f15b7","modelId":"gpt-4.1","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.766666666666666,"warning":null},{"id":"2228ebd23f321bb595a53cb8","modelId":"gpt-4.1","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.076023391812896,"warning":null},{"id":"7c2b1459ddc5b975cc6374e5","modelId":"gpt-4.1","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-4.1","configuration":"GPT-4.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.6363636363636,"warning":null},{"id":"eebaae45907cf005abada7c4","modelId":"gpt-4.1-mini","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.3333333333333,"warning":null},{"id":"78e5e2b4996ff8eb159b948d","modelId":"gpt-4.1-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":44,"warning":null},{"id":"fef727774450464c08bdd0f7","modelId":"gpt-4.1-mini","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"953e81fe2ed1bb5edb755613","modelId":"gpt-4.1-mini","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":270.29,"warning":null},{"id":"2aa8871a8db562e9d18d39f5","modelId":"gpt-4.1-mini","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.3636363636364,"warning":null},{"id":"98e459e9baa066381c77aed4","modelId":"gpt-4.1-mini","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.0200000000000005,"warning":null},{"id":"b899d6847a198d9782271fcf","modelId":"gpt-4.1-mini","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.2993197278912,"warning":null},{"id":"f3555e10497e1abe114c7141","modelId":"gpt-4.1-mini","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.253968253968296,"warning":null},{"id":"22a4815b9fac3ade7ef81d55","modelId":"gpt-4.1-mini","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":92.5333333333333,"warning":null},{"id":"6a981d16086d951f236f90c7","modelId":"gpt-4.1-mini","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.0834441489362,"warning":null},{"id":"e223341eecccef740f030378","modelId":"gpt-4.1-mini","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.728323699422006,"warning":null},{"id":"d5aa5d02a1b67889fcf36b3d","modelId":"gpt-4.1-mini","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":-53.5666666666667,"warning":null},{"id":"b85f2dfb7db4041669025541","modelId":"gpt-4.1-mini","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.316666666666666,"warning":null},{"id":"e0c1df7e5bd17d33c2f2bd4e","modelId":"gpt-4.1-mini","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.9239766081871,"warning":null},{"id":"a234bcb3480d187f78e691a1","modelId":"gpt-4.1-mini","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.36082474226804,"warning":null},{"id":"c9e7e69728999f70de8da6ac","modelId":"gpt-4.1-mini","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.57575757575758,"warning":null},{"id":"4bdb55f172fca20477d7ecfe","modelId":"gpt-4.1-mini","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GPT-4.1 mini","configuration":"GPT-4.1 mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.1123595505618,"warning":null},{"id":"28ea576bcc8fbc7f7d7c10fa","modelId":"gpt-4.1-nano","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":24,"warning":null},{"id":"489b206164aa46bc047e9442","modelId":"gpt-4.1-nano","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.3333333333333,"warning":null},{"id":"622f6b3b250d7dd489c93fdd","modelId":"gpt-4.1-nano","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"cf513ca8c95b4a92239280d5","modelId":"gpt-4.1-nano","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":-213.03,"warning":null},{"id":"0776aad3ae4e24e33e23a782","modelId":"gpt-4.1-nano","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.2121212121212,"warning":null},{"id":"a3e067b47919de9fb8db721d","modelId":"gpt-4.1-nano","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.75347544022243,"warning":null},{"id":"8a4df2fd7696ccf624dc7ec5","modelId":"gpt-4.1-nano","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.0408163265306,"warning":null},{"id":"b423f62f5dc4c141f8ed488b","modelId":"gpt-4.1-nano","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.5925925925926,"warning":null},{"id":"fd5c7d1cec7975be45682128","modelId":"gpt-4.1-nano","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.8,"warning":null},{"id":"77df975eff4ea000f47f5a15","modelId":"gpt-4.1-nano","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.6831781914894,"warning":null},{"id":"588685ec1855237622d161dc","modelId":"gpt-4.1-nano","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.1156069364162,"warning":null},{"id":"00077c09ee5961a857ea9812","modelId":"gpt-4.1-nano","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":-57.5833333333333,"warning":null},{"id":"9ac947bf063f982fe5973290","modelId":"gpt-4.1-nano","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.683333333333334,"warning":null},{"id":"123959e1480ca28caac46f0a","modelId":"gpt-4.1-nano","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.2514619883041,"warning":null},{"id":"f223b4e74f1c051e27d17567","modelId":"gpt-4.1-nano","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.50515463917526,"warning":null},{"id":"caebc613dd7096aa01c68408","modelId":"gpt-4.1-nano","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.78787878787879,"warning":null},{"id":"645eb8d2b06c1ca684efea03","modelId":"gpt-4.1-nano","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GPT-4.1 nano","configuration":"GPT-4.1 nano; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4-1-nano; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.7453183520599302,"warning":null},{"id":"60cd4a91f69f107025375b3f","modelId":"gpt-4o","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-4o (March 2025, chatgpt-4o-latest)","configuration":"GPT-4o (March 2025, chatgpt-4o-latest); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt-03-25; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.6666666666667,"warning":null},{"id":"bc7edbd7f528efe446e8db18","modelId":"gpt-4o","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":6,"warning":null},{"id":"870a8a46a2b531f6c5af62e8","modelId":"gpt-4o","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":41,"warning":null},{"id":"ca8731b9c382bd00c0d8665b","modelId":"gpt-4o","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-4o (ChatGPT)","configuration":"GPT-4o (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":56.00000000000001,"warning":null},{"id":"fddc09f5e0e8bc06d04874e4","modelId":"gpt-4o","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.3333333333333,"warning":null},{"id":"2ae788a49fc1d3e582e9f839","modelId":"gpt-4o","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"c2e74edde37a3d7442f9667e","modelId":"gpt-4o","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"8504120408952353f6c14d15","modelId":"gpt-4o","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":-16.24,"warning":null},{"id":"24a10eeb3a46c7763c5a63b2","modelId":"gpt-4o","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4o (May '24)","configuration":"GPT-4o (May '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-05-13; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.6262626262626,"warning":null},{"id":"4e82b51b3daf5aa1c8d98aee","modelId":"gpt-4o","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4o (March 2025, chatgpt-4o-latest)","configuration":"GPT-4o (March 2025, chatgpt-4o-latest); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt-03-25; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.4545454545455,"warning":null},{"id":"ba806794d6c0ebc1b2585e8f","modelId":"gpt-4o","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.1212121212121,"warning":null},{"id":"c88594bf8e16a074cc837dfb","modelId":"gpt-4o","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4o (ChatGPT)","configuration":"GPT-4o (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.11111111111109,"warning":null},{"id":"e0c41d17488dc4ec4ba0f0b7","modelId":"gpt-4o","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.3434343434343,"warning":null},{"id":"5b174d58d4c658ab3d23b8bd","modelId":"gpt-4o","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4o (ChatGPT)","configuration":"GPT-4o (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.73,"warning":null},{"id":"676bf2b6f6b8d425be537a91","modelId":"gpt-4o","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4o (March 2025, chatgpt-4o-latest)","configuration":"GPT-4o (March 2025, chatgpt-4o-latest); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt-03-25; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.9699999999999998,"warning":null},{"id":"c5b33db9dc0245a7fa091824","modelId":"gpt-4o","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.42,"warning":null},{"id":"c611da33ffcd9a9e78304e4b","modelId":"gpt-4o","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4o (May '24)","configuration":"GPT-4o (May '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-05-13; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.76,"warning":null},{"id":"fd73d709f3bb83bda7f56ac0","modelId":"gpt-4o","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.33,"warning":null},{"id":"adfbe98a4535a8b4dcbba8db","modelId":"gpt-4o","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.9863945578231,"warning":null},{"id":"e8a8172c365961fe52158eb8","modelId":"gpt-4o","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.285714285714306,"warning":null},{"id":"111a771f0c518f482817be18","modelId":"gpt-4o","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.7460317460317,"warning":null},{"id":"8d6af27176cb848c0da30e16","modelId":"gpt-4o","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4o (May '24)","configuration":"GPT-4o (May '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-05-13; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.4391534391534,"warning":null},{"id":"952674fc62e371988760c75e","modelId":"gpt-4o","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4o (March 2025, chatgpt-4o-latest)","configuration":"GPT-4o (March 2025, chatgpt-4o-latest); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt-03-25; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.5396825396825,"warning":null},{"id":"e26cd77447c8fea5c2954d73","modelId":"gpt-4o","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.8994708994709,"warning":null},{"id":"08df3f3cdd6b93731cef3bfe","modelId":"gpt-4o","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4o (March 2025, chatgpt-4o-latest)","configuration":"GPT-4o (March 2025, chatgpt-4o-latest); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt-03-25; Mode from evaluator isReasoning metadata; effort budget not specified","score":89.26666666666671,"warning":null},{"id":"51e4170d40d5c177fe0231db","modelId":"gpt-4o","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.9333333333333,"warning":null},{"id":"60f11419a7766054a4108a18","modelId":"gpt-4o","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4o (ChatGPT)","configuration":"GPT-4o (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.7333333333333,"warning":null},{"id":"67323d93d4710a1a34a75f0f","modelId":"gpt-4o","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.4666666666667,"warning":null},{"id":"93d23dcf1ef65f2e23f876f9","modelId":"gpt-4o","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4o (May '24)","configuration":"GPT-4o (May '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-05-13; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.1333333333333,"warning":null},{"id":"26a5dad1362342e4f74d3ddc","modelId":"gpt-4o","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4o (March 2025, chatgpt-4o-latest)","configuration":"GPT-4o (March 2025, chatgpt-4o-latest); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt-03-25; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.2942154255319,"warning":null},{"id":"8196e10e1bf2c6ab7a48896c","modelId":"gpt-4o","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.7672872340426,"warning":null},{"id":"893141e7225b21e42fa2ce7d","modelId":"gpt-4o","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4o (ChatGPT)","configuration":"GPT-4o (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.3354388297872,"warning":null},{"id":"ce9dd7a47db415076cf5f8f6","modelId":"gpt-4o","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4o (May '24)","configuration":"GPT-4o (May '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-05-13; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.9527925531915,"warning":null},{"id":"b8d0d0d98797b082257d92e1","modelId":"gpt-4o","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":56.300578034682104,"warning":null},{"id":"39324a10e110c667372e0e04","modelId":"gpt-4o","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":-10.5166666666667,"warning":null},{"id":"a9b3da7256fa223acaa958e2","modelId":"gpt-4o","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":-20.8666666666667,"warning":null},{"id":"750e957428961c84d14679b0","modelId":"gpt-4o","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.7,"warning":null},{"id":"b29b60a9d0d8925b03ed31d5","modelId":"gpt-4o","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.866666666666667,"warning":null},{"id":"007fbaa1f184a893e7a405b1","modelId":"gpt-4o","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.947368421052598,"warning":null},{"id":"79b4fb7fde5db57eeb15ccdc","modelId":"gpt-4o","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.1461988304094,"warning":null},{"id":"893788f0d9759cc5abf9ec76","modelId":"gpt-4o","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-4o (Nov '24)","configuration":"GPT-4o (Nov '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.33333333333333,"warning":null},{"id":"9d041de7b25693c5693bff38","modelId":"gpt-4o","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-4o (Aug '24)","configuration":"GPT-4o (Aug '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-2024-08-06; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.33333333333333,"warning":null},{"id":"454128a395b09cae70c46934","modelId":"gpt-4o-mini","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.6666666666667,"warning":null},{"id":"cabe088f2008b7789f50fbe3","modelId":"gpt-4o-mini","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":-22.31,"warning":null},{"id":"feeff42eebfedd28d3c20c28","modelId":"gpt-4o-mini","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.6262626262626,"warning":null},{"id":"40656865ba698a00d4904512","modelId":"gpt-4o-mini","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.2,"warning":null},{"id":"d07cd6e9ce1790b198fada31","modelId":"gpt-4o-mini","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.952380952381,"warning":null},{"id":"9211c553e48b0e7de26c2187","modelId":"gpt-4o-mini","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.3862433862434,"warning":null},{"id":"a918130cdb50b373f292baf0","modelId":"gpt-4o-mini","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.8666666666667,"warning":null},{"id":"89e1e3c13465dce088558ff9","modelId":"gpt-4o-mini","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"ebe5f26d2cff3c26c1e525dc","modelId":"gpt-4o-mini","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.77726063829789,"warning":null},{"id":"9efca0c777e99b96ac172fcc","modelId":"gpt-4o-mini","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.5028901734104,"warning":null},{"id":"cd920a26d682a51dea69c47c","modelId":"gpt-4o-mini","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.88659793814433,"warning":null},{"id":"0fd35df2bab242febe516b4d","modelId":"gpt-4o-mini","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GPT-4o mini","configuration":"GPT-4o mini; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-4o-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.61797752808989,"warning":null},{"id":"b8a7ebc653a427fb5d91d702","modelId":"gpt-5","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":94.3333333333333,"warning":null},{"id":"fd5387fb3404096f414a83b6","modelId":"gpt-5","benchmarkId":"aa-aime-2025","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":83,"warning":null},{"id":"62a27138a69e54269b2ced0d","modelId":"gpt-5","benchmarkId":"aa-aime-2025","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":91.6666666666667,"warning":null},{"id":"8e45b797da8beac9986e192f","modelId":"gpt-5","benchmarkId":"aa-aime-2025","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":31.6666666666667,"warning":null},{"id":"a69193c2658d703e3ab0d38f","modelId":"gpt-5","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.3333333333333,"warning":null},{"id":"750c968ff0446a482227e2b8","modelId":"gpt-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":78.2,"warning":null},{"id":"bac1e78739c1eb915445e9a6","modelId":"gpt-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":76,"warning":null},{"id":"886b2d802aa24414c17a9bc7","modelId":"gpt-5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":65,"warning":null},{"id":"526949b552b53e03b7d5a1fb","modelId":"gpt-5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":5.71428571428571,"warning":null},{"id":"5d4dcc5da420ddf20432d312","modelId":"gpt-5","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":1.1428571428571401,"warning":null},{"id":"5876f00ed1d82282ca87b0b9","modelId":"gpt-5","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":0,"warning":null},{"id":"b0e5c086b1f1f24d633cf203","modelId":"gpt-5","benchmarkId":"aa-critpt","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":0,"warning":null},{"id":"25d613e45d75b6f63c137be2","modelId":"gpt-5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":917.24,"warning":null},{"id":"19f8dadfaafe2174ffda6f14","modelId":"gpt-5","benchmarkId":"aa-global-mmlu-lite","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":90.66666666666667,"warning":null},{"id":"aa1c6942d4b8c7d2f605217f","modelId":"gpt-5","benchmarkId":"aa-global-mmlu-lite","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":90.39062499999999,"warning":null},{"id":"9b6573c67479dde64f04035d","modelId":"gpt-5","benchmarkId":"aa-global-mmlu-lite","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":84.421875,"warning":null},{"id":"019aa804c24b7642bbb4c9fe","modelId":"gpt-5","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":85.3535353535353,"warning":null},{"id":"223453038a985c31ffd6e4f8","modelId":"gpt-5","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":80.8080808080808,"warning":null},{"id":"9a39eae30636e7c9886cce67","modelId":"gpt-5","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":84.17508417508421,"warning":null},{"id":"b23e0251391efcf3b3a4bb40","modelId":"gpt-5","benchmarkId":"aa-gpqa-diamond","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":67.2727272727273,"warning":null},{"id":"ebf5588481b79006df61a812","modelId":"gpt-5","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.58585858585859,"warning":null},{"id":"8e803bc9d5b5c1ab1a7f9f24","modelId":"gpt-5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":28.498609823910996,"warning":null},{"id":"cd6b6e18fbfd95506365fd6e","modelId":"gpt-5","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":19.5551436515292,"warning":null},{"id":"8f4b11583819dbf74b843a4b","modelId":"gpt-5","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":25.3938832252085,"warning":null},{"id":"3eb032538fe120fe67ce3a7b","modelId":"gpt-5","benchmarkId":"aa-humanitys-last-exam","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":6.02409638554217,"warning":null},{"id":"de2fbae5009e0d6316cb4494","modelId":"gpt-5","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.62650602409639,"warning":null},{"id":"00227af149e9071d69f93aca","modelId":"gpt-5","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":73.06122448979589,"warning":null},{"id":"6f3d1c7bf90393fa9c17b8b8","modelId":"gpt-5","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":66.5986394557823,"warning":null},{"id":"154efb6f937bc36bf9ad232b","modelId":"gpt-5","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":70.6122448979592,"warning":null},{"id":"39f1d0850b0ebc50b216eaae","modelId":"gpt-5","benchmarkId":"aa-ifbench","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":45.578231292517,"warning":null},{"id":"76566dd933fc00aeb2f0db10","modelId":"gpt-5","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":45.034013605442205,"warning":null},{"id":"4f4d61af0c9b181e6c2c873c","modelId":"gpt-5","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":84.5502645502646,"warning":null},{"id":"6e9e875016ba9859a8f6134a","modelId":"gpt-5","benchmarkId":"aa-livecodebench","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":76.29629629629629,"warning":null},{"id":"a62167102cbf66237691c3f0","modelId":"gpt-5","benchmarkId":"aa-livecodebench","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":70.2645502645503,"warning":null},{"id":"f282d8c01d4dfa73b7c21cf6","modelId":"gpt-5","benchmarkId":"aa-livecodebench","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":55.7671957671958,"warning":null},{"id":"c65c8b6baa6f11b465f55b52","modelId":"gpt-5","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.285714285714306,"warning":null},{"id":"8879b6baa9e051ad62449770","modelId":"gpt-5","benchmarkId":"aa-math-500","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":99.4,"warning":null},{"id":"a0ddf05c4b0c98132dddb60a","modelId":"gpt-5","benchmarkId":"aa-math-500","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":98.73333333333329,"warning":null},{"id":"0eaef2c506d4f01adfa84574","modelId":"gpt-5","benchmarkId":"aa-math-500","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":99.1333333333333,"warning":null},{"id":"db0f5701f2e6a79689c6cea7","modelId":"gpt-5","benchmarkId":"aa-math-500","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":86.1333333333333,"warning":null},{"id":"7a42777f53f98dabc1ad53c8","modelId":"gpt-5","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":87.0678191489362,"warning":null},{"id":"79eb03e7805b182a2759793a","modelId":"gpt-5","benchmarkId":"aa-mmlu-pro","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":85.9790558510638,"warning":null},{"id":"21ec6c8059cfd230abba188d","modelId":"gpt-5","benchmarkId":"aa-mmlu-pro","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":86.7436835106383,"warning":null},{"id":"694430f2d7df7d6cdf682cb2","modelId":"gpt-5","benchmarkId":"aa-mmlu-pro","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":80.63497340425529,"warning":null},{"id":"5be07669413d5fbc82f7c02d","modelId":"gpt-5","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.0395611702128,"warning":null},{"id":"4e8e70956b7eae3fc7e4c102","modelId":"gpt-5","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":74.2196531791907,"warning":null},{"id":"b0e5a0d9eec214879841fe64","modelId":"gpt-5","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":73.757225433526,"warning":null},{"id":"a9ff5ecf1282ea87ec70832b","modelId":"gpt-5","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":74.3352601156069,"warning":null},{"id":"4217568857ca6459c186bcdc","modelId":"gpt-5","benchmarkId":"aa-mmmu-pro","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":62.080924855491304,"warning":null},{"id":"89547fadc0a82f235410b5f8","modelId":"gpt-5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":-8.73333333333333,"warning":null},{"id":"48c2795f7767156dfb87810e","modelId":"gpt-5","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":-10.7833333333333,"warning":null},{"id":"d269fd7d536cdc788d304da2","modelId":"gpt-5","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":-10.8666666666667,"warning":null},{"id":"83da1b801f784f99b059e2e3","modelId":"gpt-5","benchmarkId":"aa-omniscience","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":-33.8,"warning":null},{"id":"868c4193cee868de3bdc6889","modelId":"gpt-5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":40.31666666666667,"warning":null},{"id":"ee2531a02a7ffddc0fb01d8c","modelId":"gpt-5","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":38.11666666666667,"warning":null},{"id":"29adc8ded1a9c585293aa02c","modelId":"gpt-5","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":39.483333333333334,"warning":null},{"id":"51d11a89b1062b97c6c2afa5","modelId":"gpt-5","benchmarkId":"aa-omniscience-accuracy","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":29.633333333333333,"warning":null},{"id":"8737030f1ce70358f2e83364","modelId":"gpt-5","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":84.7953216374269,"warning":null},{"id":"b701276cc6546fc45822b865","modelId":"gpt-5","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":84.21052631578951,"warning":null},{"id":"a80c5c3b6ff92417077955ad","modelId":"gpt-5","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":86.5497076023392,"warning":null},{"id":"43f959c3e3f4b66e82a11070","modelId":"gpt-5","benchmarkId":"aa-tau2-bench","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":66.9590643274854,"warning":null},{"id":"2ed696e37ca20657d9df1361","modelId":"gpt-5","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f7e07fd86961e867896fb84b","modelId":"gpt-5","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":22.0618556701031,"warning":null},{"id":"b34ccd07a2ed0494dfd6a9d9","modelId":"gpt-5","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":32.5757575757576,"warning":null},{"id":"a6f7ea2ee9bcce410ee2baf5","modelId":"gpt-5","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-low","score":26.5151515151515,"warning":null},{"id":"3386dce9d4047ad04ffdb523","modelId":"gpt-5","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-medium","score":37.878787878787904,"warning":null},{"id":"5fae70015333e42ee80aabe2","modelId":"gpt-5","benchmarkId":"aa-terminalbench-hard","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-minimal","score":18.181818181818198,"warning":null},{"id":"292b0f2eaa1075a11794a0fc","modelId":"gpt-5","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-5 (ChatGPT)","configuration":"GPT-5 (ChatGPT); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/gpt-5-chatgpt; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.8787878787879,"warning":null},{"id":"0f34a353218d6fbf085641e3","modelId":"gpt-5","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Model source: https://artificialanalysis.ai/models/gpt-5","score":35.2059925093633,"warning":null},{"id":"cd810b3c8809c7e5dacdc516","modelId":"gpt-5","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5 (High)","configuration":"GPT-5 (High); Source model ID: gpt-5-2025-08-07-high; 120 tasks","score":9.86,"warning":null},{"id":"6d4dee0e98a30d128b276dd5","modelId":"gpt-5","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5 (Low)","configuration":"GPT-5 (Low); Source model ID: gpt-5-2025-08-07-low; 120 tasks","score":1.94,"warning":null},{"id":"88ed12ff4a36b415143266f3","modelId":"gpt-5","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5 (Medium)","configuration":"GPT-5 (Medium); Source model ID: gpt-5-2025-08-07-medium; 120 tasks","score":7.489999999999999,"warning":null},{"id":"3fcb070e5d72027bf895a4eb","modelId":"gpt-5","benchmarkId":"arc-v2semiprivate","level":"minimal","sourceLabel":"GPT-5 (Minimal)","configuration":"GPT-5 (Minimal); Source model ID: gpt-5-2025-08-07-minimal; 120 tasks","score":0,"warning":null},{"id":"9a4e17f7381dfa910e957c3d","modelId":"gpt-5-mini","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":428.44,"warning":null},{"id":"65f52ca925e988e6fd80eb13","modelId":"gpt-5-mini","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":90.6666666666667,"warning":null},{"id":"ed48d6c5fd48bdcb66ac91dc","modelId":"gpt-5-mini","benchmarkId":"aa-aime-2025","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":85,"warning":null},{"id":"bd865d0ac07d2c1f574d3789","modelId":"gpt-5-mini","benchmarkId":"aa-aime-2025","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":46.6666666666667,"warning":null},{"id":"cdb88e73a4762d4396c9a06e","modelId":"gpt-5-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":72.3333333333333,"warning":null},{"id":"da8ec0a0d77bf26ada37ff58","modelId":"gpt-5-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":72.3333333333333,"warning":null},{"id":"57449640277c3bbc502cab3c","modelId":"gpt-5-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":39.3333333333333,"warning":null},{"id":"27af7a5af07c19f4d88811d8","modelId":"gpt-5-mini","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":6.487466331916088,"warning":null},{"id":"3c2950d0606b9481aee47005","modelId":"gpt-5-mini","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":0,"warning":null},{"id":"807b3d447b2dc7d3638a79ab","modelId":"gpt-5-mini","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":1.42857142857143,"warning":null},{"id":"d33546c969eb3e944e163fdc","modelId":"gpt-5-mini","benchmarkId":"aa-critpt","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":0,"warning":null},{"id":"bdf30c3a0f09b77a90c44aa5","modelId":"gpt-5-mini","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":8.4,"warning":null},{"id":"a3ce8400a76c8589735a3202","modelId":"gpt-5-mini","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":766.11,"warning":null},{"id":"42ba50e4937f51066403a40a","modelId":"gpt-5-mini","benchmarkId":"aa-global-mmlu-lite","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":87.42708333333333,"warning":null},{"id":"780a661cb78b9c56e5125a5c","modelId":"gpt-5-mini","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":82.8282828282828,"warning":null},{"id":"ef0e8332739997b8c28476fb","modelId":"gpt-5-mini","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":80.3030303030303,"warning":null},{"id":"30e9619086bdee5dc9d5d3ed","modelId":"gpt-5-mini","benchmarkId":"aa-gpqa-diamond","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":68.68686868686869,"warning":null},{"id":"e3c571e8cc7c47693c7491ea","modelId":"gpt-5-mini","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":21.4550509731233,"warning":null},{"id":"e34db1c859849950b0b61ec0","modelId":"gpt-5-mini","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":15.940685820203901,"warning":null},{"id":"4950a461985bd98b66e025d5","modelId":"gpt-5-mini","benchmarkId":"aa-humanitys-last-exam","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":5.050973123262279,"warning":null},{"id":"1b233a3c4817c134297e6b90","modelId":"gpt-5-mini","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":75.4421768707483,"warning":null},{"id":"f1a1a99469c099dc3e7e35d9","modelId":"gpt-5-mini","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":71.15646258503399,"warning":null},{"id":"9cde08173b7b2e5fd0285002","modelId":"gpt-5-mini","benchmarkId":"aa-ifbench","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":45.578231292517,"warning":null},{"id":"3a8c6449356ba890c11867dd","modelId":"gpt-5-mini","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":83.8095238095238,"warning":null},{"id":"3154054b9776d3b25e8ebf72","modelId":"gpt-5-mini","benchmarkId":"aa-livecodebench","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":69.2063492063492,"warning":null},{"id":"794313967169b8d8abccba39","modelId":"gpt-5-mini","benchmarkId":"aa-livecodebench","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":54.4973544973545,"warning":null},{"id":"11c3fe71af8c1d04004c9911","modelId":"gpt-5-mini","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":83.72672872340429,"warning":null},{"id":"f9273c3b76c3e55a474b88e7","modelId":"gpt-5-mini","benchmarkId":"aa-mmlu-pro","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":82.7792553191489,"warning":null},{"id":"76561c07cfe5935c8448693d","modelId":"gpt-5-mini","benchmarkId":"aa-mmlu-pro","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":77.451795212766,"warning":null},{"id":"58af9776db89fef37b1a7aea","modelId":"gpt-5-mini","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":70.0578034682081,"warning":null},{"id":"b25b59c834da48070abde317","modelId":"gpt-5-mini","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":68.8439306358381,"warning":null},{"id":"2614340876670114b1f77074","modelId":"gpt-5-mini","benchmarkId":"aa-mmmu-pro","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":58.3815028901734,"warning":null},{"id":"27a8f43f659f15e583e7a1c2","modelId":"gpt-5-mini","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":-17.3166666666667,"warning":null},{"id":"5e4c27dfeaca364435205d77","modelId":"gpt-5-mini","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":-11.0333333333333,"warning":null},{"id":"d8b8696f5fe260c359a93780","modelId":"gpt-5-mini","benchmarkId":"aa-omniscience","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":-53.05,"warning":null},{"id":"f7a8c435629149e75efe37f0","modelId":"gpt-5-mini","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":24.96666666666667,"warning":null},{"id":"f9d87df08c158478b519dba3","modelId":"gpt-5-mini","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":23.066666666666666,"warning":null},{"id":"ba344ec24690481d28cec0ea","modelId":"gpt-5-mini","benchmarkId":"aa-omniscience-accuracy","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":19.166666666666668,"warning":null},{"id":"79c10f9e00a2b29918be9a1c","modelId":"gpt-5-mini","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":39.0046296296296,"warning":null},{"id":"10f9eba629d77facbe32f9b6","modelId":"gpt-5-mini","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":68.4210526315789,"warning":null},{"id":"dd3d255b3709f055d75b816b","modelId":"gpt-5-mini","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":71.0526315789474,"warning":null},{"id":"3f562cfc40d0ee39b5ca7783","modelId":"gpt-5-mini","benchmarkId":"aa-tau2-bench","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":31.8713450292398,"warning":null},{"id":"969afa46ac1ec6e14abac457","modelId":"gpt-5-mini","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":15.4639175257732,"warning":null},{"id":"43b87396047b7441da4e876a","modelId":"gpt-5-mini","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":33.3333333333333,"warning":null},{"id":"adc5b90744736975147841e8","modelId":"gpt-5-mini","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5 mini (Medium)","configuration":"GPT-5 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-mini-medium","score":28.7878787878788,"warning":null},{"id":"fe52f5262169e97e4e9bdf53","modelId":"gpt-5-mini","benchmarkId":"aa-terminalbench-hard","level":"minimal","sourceLabel":"GPT-5 mini (Minimal)","configuration":"GPT-5 mini (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-mini-minimal","score":14.3939393939394,"warning":null},{"id":"fbbbb64b85a5e18d3f24a5c7","modelId":"gpt-5-mini","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":3.7453183520599302,"warning":null},{"id":"135ea16317bc37b9f2fd8244","modelId":"gpt-5-mini","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-5 mini (High)","configuration":"GPT-5 mini (High); Model source: https://artificialanalysis.ai/models/gpt-5-mini","score":0,"warning":null},{"id":"849548a81dc92ff4714ec68d","modelId":"gpt-5-mini","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5 Mini (High)","configuration":"GPT-5 Mini (High); Source model ID: gpt-5-mini-2025-08-07-high; 120 tasks","score":4.44,"warning":null},{"id":"c50d5e16d51cce9801649bf5","modelId":"gpt-5-mini","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5 Mini (Low)","configuration":"GPT-5 Mini (Low); Source model ID: gpt-5-mini-2025-08-07-low; 120 tasks","score":0.83,"warning":null},{"id":"0bebc0e4edb23ab523a4bf08","modelId":"gpt-5-mini","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5 Mini (Medium)","configuration":"GPT-5 Mini (Medium); Source model ID: gpt-5-mini-2025-08-07-medium; 120 tasks","score":4.03,"warning":null},{"id":"6a02220558f7ab9b3d8a2e05","modelId":"gpt-5-mini","benchmarkId":"arc-v2semiprivate","level":"minimal","sourceLabel":"GPT-5 Mini (Minimal)","configuration":"GPT-5 Mini (Minimal); Source model ID: gpt-5-mini-2025-08-07-minimal; 120 tasks","score":1.67,"warning":null},{"id":"baae78f87fb06346c342f6dc","modelId":"gpt-5-nano","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":83.6666666666667,"warning":null},{"id":"38399b13b9de7e1d91f21876","modelId":"gpt-5-nano","benchmarkId":"aa-aime-2025","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":78.3333333333333,"warning":null},{"id":"f44ab80f5fba3db8c7f18446","modelId":"gpt-5-nano","benchmarkId":"aa-aime-2025","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":27.3333333333333,"warning":null},{"id":"beca8baccf9ea8527cbb05d0","modelId":"gpt-5-nano","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":45,"warning":null},{"id":"0cf6917b7fb6841cfa045d52","modelId":"gpt-5-nano","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":43.6666666666667,"warning":null},{"id":"1a932302aa9358c9850b009f","modelId":"gpt-5-nano","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":20,"warning":null},{"id":"12bb790421cfdc6cfdc5ba2c","modelId":"gpt-5-nano","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":0,"warning":null},{"id":"c4843a25f549d8b716b9d54c","modelId":"gpt-5-nano","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":0,"warning":null},{"id":"963319ee3a3ca101dbac6222","modelId":"gpt-5-nano","benchmarkId":"aa-critpt","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":0,"warning":null},{"id":"0c398d7ef4cbf729d318d8e7","modelId":"gpt-5-nano","benchmarkId":"aa-global-mmlu-lite","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":81.22916666666666,"warning":null},{"id":"f3a1f85787c45ea8d3745d7d","modelId":"gpt-5-nano","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":67.5757575757576,"warning":null},{"id":"2f2b5934711bd83c794bca3c","modelId":"gpt-5-nano","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":66.969696969697,"warning":null},{"id":"fecb40f15fac8f79d21c736a","modelId":"gpt-5-nano","benchmarkId":"aa-gpqa-diamond","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":42.8282828282828,"warning":null},{"id":"c4b74513315e89c6be419f27","modelId":"gpt-5-nano","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":9.49953660797034,"warning":null},{"id":"17febd32556cd8287d50e914","modelId":"gpt-5-nano","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":8.71177015755329,"warning":null},{"id":"40a6fa8fe27e1cb42352a013","modelId":"gpt-5-nano","benchmarkId":"aa-humanitys-last-exam","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":3.9851714550509696,"warning":null},{"id":"3c36c41ef8433083b77bbcae","modelId":"gpt-5-nano","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":67.5510204081633,"warning":null},{"id":"b1d7a280d371e1dfae991f0c","modelId":"gpt-5-nano","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":65.9183673469388,"warning":null},{"id":"c14d9a856130fce7ddc201ee","modelId":"gpt-5-nano","benchmarkId":"aa-ifbench","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":32.5170068027211,"warning":null},{"id":"8f27194e8f6eae6fbfe4c720","modelId":"gpt-5-nano","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":78.94179894179891,"warning":null},{"id":"33fb1a7cdfd7505524d236d7","modelId":"gpt-5-nano","benchmarkId":"aa-livecodebench","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":76.29629629629629,"warning":null},{"id":"70a4ddb203dc7067eb4e2368","modelId":"gpt-5-nano","benchmarkId":"aa-livecodebench","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":46.984126984127,"warning":null},{"id":"78da8d5c93ca3b33ea097baa","modelId":"gpt-5-nano","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":78.0252659574468,"warning":null},{"id":"091698af7c2cf7e4d2769572","modelId":"gpt-5-nano","benchmarkId":"aa-mmlu-pro","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":77.235704787234,"warning":null},{"id":"ecc43ffded9f3e459a4522d7","modelId":"gpt-5-nano","benchmarkId":"aa-mmlu-pro","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":55.6432845744681,"warning":null},{"id":"cec2ea912aaf272c90600116","modelId":"gpt-5-nano","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":60.982658959537595,"warning":null},{"id":"9c032e16f3e7d5a05305275c","modelId":"gpt-5-nano","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":58.2080924855491,"warning":null},{"id":"fe2adb9e4fe798077dc41728","modelId":"gpt-5-nano","benchmarkId":"aa-mmmu-pro","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":31.7919075144509,"warning":null},{"id":"495bc899315c1581ce854e28","modelId":"gpt-5-nano","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":-28.6666666666667,"warning":null},{"id":"43575cdae591da725c3b65b4","modelId":"gpt-5-nano","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":-25.7833333333333,"warning":null},{"id":"e0bdcd0f68066992a4330f39","modelId":"gpt-5-nano","benchmarkId":"aa-omniscience","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":-64.0666666666667,"warning":null},{"id":"6a9cc98dd3a7037f27758256","modelId":"gpt-5-nano","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":19.083333333333332,"warning":null},{"id":"b5368bdd72014baa092eeb50","modelId":"gpt-5-nano","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":17.583333333333336,"warning":null},{"id":"bd46f731ec6d1b7e92e3d9a6","modelId":"gpt-5-nano","benchmarkId":"aa-omniscience-accuracy","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":13.083333333333332,"warning":null},{"id":"f3fd664c0614fc9f79234584","modelId":"gpt-5-nano","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":36.5497076023392,"warning":null},{"id":"3daecf89a2538734c8e34de8","modelId":"gpt-5-nano","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":30.409356725146203,"warning":null},{"id":"9694dc8a1430273bd490da0e","modelId":"gpt-5-nano","benchmarkId":"aa-tau2-bench","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":25.730994152046797,"warning":null},{"id":"c569fa97c55a3ac0b4f8613c","modelId":"gpt-5-nano","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"GPT-5 nano (High)","configuration":"GPT-5 nano (High); Model source: https://artificialanalysis.ai/models/gpt-5-nano","score":12.1212121212121,"warning":null},{"id":"f54a607ff952e13aeae8d4ad","modelId":"gpt-5-nano","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5 nano (Medium)","configuration":"GPT-5 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-nano-medium","score":17.4242424242424,"warning":null},{"id":"005f4d23a6ce096654405de9","modelId":"gpt-5-nano","benchmarkId":"aa-terminalbench-hard","level":"minimal","sourceLabel":"GPT-5 nano (Minimal)","configuration":"GPT-5 nano (Minimal); Model source: https://artificialanalysis.ai/models/gpt-5-nano-minimal","score":6.81818181818182,"warning":null},{"id":"999b6bf8f366cec37f99e596","modelId":"gpt-5-nano","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5 Nano (High)","configuration":"GPT-5 Nano (High); Source model ID: gpt-5-nano-2025-08-07-high; 120 tasks","score":2.6100000000000003,"warning":null},{"id":"f05d176f9e5db89e675db70c","modelId":"gpt-5-nano","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5 Nano (Low)","configuration":"GPT-5 Nano (Low); Source model ID: gpt-5-nano-2025-08-07-low; 120 tasks","score":0,"warning":null},{"id":"b5db4793b8e4f1f4fac5f2e3","modelId":"gpt-5-nano","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5 Nano (Medium)","configuration":"GPT-5 Nano (Medium); Source model ID: gpt-5-nano-2025-08-07-medium; 120 tasks","score":0.88,"warning":null},{"id":"151137af98f4465de79cb3c1","modelId":"gpt-5-nano","benchmarkId":"arc-v2semiprivate","level":"minimal","sourceLabel":"GPT-5 Nano (Minimal)","configuration":"GPT-5 Nano (Minimal); Source model ID: gpt-5-nano-2025-08-07-minimal; 120 tasks","score":0,"warning":null},{"id":"11250576e9361fefb5ca3b64","modelId":"gpt-5.1","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":94,"warning":null},{"id":"6abab40890aa0c0372d0f089","modelId":"gpt-5.1","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":38,"warning":null},{"id":"6bcad467ad3fe69ff8cc3deb","modelId":"gpt-5.1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":80,"warning":null},{"id":"242467cf8ae03ca5317cf999","modelId":"gpt-5.1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":45,"warning":null},{"id":"9dd6f928e3fc2dd144bf9af5","modelId":"gpt-5.1","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":4.857142857142859,"warning":null},{"id":"9268c5b8746b04887ef9d764","modelId":"gpt-5.1","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":0,"warning":null},{"id":"6fdb7db289bccb88ce935fd5","modelId":"gpt-5.1","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":826.29,"warning":null},{"id":"0287f91fa9ecb659aa60c4a7","modelId":"gpt-5.1","benchmarkId":"aa-global-mmlu-lite","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":90.55729166666666,"warning":null},{"id":"baacbdf1ad1b1e4c82bb0f35","modelId":"gpt-5.1","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":87.27272727272731,"warning":null},{"id":"2fcc61e29b95a07ac78e5842","modelId":"gpt-5.1","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":64.3434343434343,"warning":null},{"id":"abc94a1fb2df11a57957513d","modelId":"gpt-5.1","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":28.498609823910996,"warning":null},{"id":"7cf4aa2e67a6c7ef11ed452d","modelId":"gpt-5.1","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":5.28266913809083,"warning":null},{"id":"d7e7090c6f375f2da1bee8f0","modelId":"gpt-5.1","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":72.8571428571429,"warning":null},{"id":"3cf752361e0a75cc3c1dc615","modelId":"gpt-5.1","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":43.1972789115646,"warning":null},{"id":"ed5c9c1f5fc146f01be5bfb3","modelId":"gpt-5.1","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":86.7724867724868,"warning":null},{"id":"f1596ea955bb7e7d0db23d9a","modelId":"gpt-5.1","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":49.417989417989396,"warning":null},{"id":"4b01031f2fb3cafc944fec0a","modelId":"gpt-5.1","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":87.04288563829789,"warning":null},{"id":"15ccd3e2bad00dbe4d513d94","modelId":"gpt-5.1","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":80.1279920212766,"warning":null},{"id":"e3b8ab065a5b4025535d762a","modelId":"gpt-5.1","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":75.4913294797688,"warning":null},{"id":"6ecc1a5e9d3e0bc1554f7e5a","modelId":"gpt-5.1","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":62.3699421965318,"warning":null},{"id":"671d0f66976d744d141136a9","modelId":"gpt-5.1","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":5.4,"warning":null},{"id":"ad964ce0ca054e729c02f3c7","modelId":"gpt-5.1","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":-34.45,"warning":null},{"id":"fac3ffd101175424a69bcc4b","modelId":"gpt-5.1","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":37.71666666666667,"warning":null},{"id":"dcb6c1cfb154e6dfd47677c2","modelId":"gpt-5.1","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":29.46666666666667,"warning":null},{"id":"5ff9f965317ff34501845ba5","modelId":"gpt-5.1","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":81.87134502923979,"warning":null},{"id":"301d76c19099e4fcffbb57c5","modelId":"gpt-5.1","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":46.4912280701754,"warning":null},{"id":"830f557b2cf11822bf07e107","modelId":"gpt-5.1","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":15.876288659793799,"warning":null},{"id":"8a953d660cb5bda8f9879649","modelId":"gpt-5.1","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":45.454545454545496,"warning":null},{"id":"f32ac719f9f42f00e7041db0","modelId":"gpt-5.1","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-5.1 (Non-reasoning)","configuration":"GPT-5.1 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-1-non-reasoning","score":22.727272727272698,"warning":null},{"id":"8f8490ceebcfe23747f231db","modelId":"gpt-5.1","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-5.1 (High)","configuration":"GPT-5.1 (High); Model source: https://artificialanalysis.ai/models/gpt-5-1","score":52.4344569288389,"warning":null},{"id":"0fbbe06ee7a41a332409a02a","modelId":"gpt-5.1","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.1 (Thinking, High)","configuration":"GPT-5.1 (Thinking, High); Source model ID: gpt-5-1-2025-11-13-thinking-high; 120 tasks","score":17.64,"warning":null},{"id":"0f99ee8a4e9e3a03a96cb7c2","modelId":"gpt-5.1","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.1 (Thinking, Low)","configuration":"GPT-5.1 (Thinking, Low); Source model ID: gpt-5-1-2025-11-13-thinking-low; 120 tasks","score":1.94,"warning":null},{"id":"be4f45a2511f61842abfae20","modelId":"gpt-5.1","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.1 (Thinking, Medium)","configuration":"GPT-5.1 (Thinking, Medium); Source model ID: gpt-5-1-2025-11-13-thinking-medium; 120 tasks","score":6.529999999999999,"warning":null},{"id":"29c08eeadae8b2d685b7f079","modelId":"gpt-5.1","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"GPT-5.1 (Thinking, None)","configuration":"GPT-5.1 (Thinking, None); Source model ID: gpt-5-1-2025-11-13-thinking-none; 120 tasks","score":0.42,"warning":null},{"id":"c6025ca3320d8c4928328b2b","modelId":"gpt-5.2","benchmarkId":"aa-aime-2025","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":96.6666666666667,"warning":null},{"id":"43f5bf080d66ec470f227436","modelId":"gpt-5.2","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":51,"warning":null},{"id":"dc0488e261bd5a31286821f7","modelId":"gpt-5.2","benchmarkId":"aa-aime-2025","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":99,"warning":null},{"id":"30f1effd53d1e31524ad55d8","modelId":"gpt-5.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":70.3333333333333,"warning":null},{"id":"97b188f9a482b61245c487bf","modelId":"gpt-5.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":45.6666666666667,"warning":null},{"id":"4389b41ae30766d84439d9d1","modelId":"gpt-5.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":82.6666666666667,"warning":null},{"id":"2e6c0b28d53c156bb01e107b","modelId":"gpt-5.2","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":7.85714285714285,"warning":null},{"id":"39e4db6a2d4b286593f81a2f","modelId":"gpt-5.2","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":0.571428571428571,"warning":null},{"id":"270b13c88ee3289bf1e374c0","modelId":"gpt-5.2","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":11.5510204081633,"warning":null},{"id":"27e740b88688dc7ba4618798","modelId":"gpt-5.2","benchmarkId":"aa-global-mmlu-lite","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":89.75520833333334,"warning":null},{"id":"8bcbd556ed005557ce8a7b3c","modelId":"gpt-5.2","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":86.80208333333333,"warning":null},{"id":"c9ea3ddcdf8abc8b74728938","modelId":"gpt-5.2","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":86.3636363636364,"warning":null},{"id":"edd0345b8eecf02194512beb","modelId":"gpt-5.2","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":71.2121212121212,"warning":null},{"id":"a63b906883d72482322ae8ae","modelId":"gpt-5.2","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":90.3030303030303,"warning":null},{"id":"1b9125ee38b1b279002320b1","modelId":"gpt-5.2","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":26.6913809082484,"warning":null},{"id":"8bf3ce28a17c78a01b889176","modelId":"gpt-5.2","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":8.01668211306766,"warning":null},{"id":"299ca2a6d6badc28c370ac43","modelId":"gpt-5.2","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":37.6737720111214,"warning":null},{"id":"3a4ed47518fbc9057c2c31ab","modelId":"gpt-5.2","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":65.2380952380953,"warning":null},{"id":"93f2eb1c55112966a4b0b422","modelId":"gpt-5.2","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":47.414965986394606,"warning":null},{"id":"5649ef2737dcff8b99c635e9","modelId":"gpt-5.2","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":75.4421768707483,"warning":null},{"id":"59b1daf784997f39edc8287a","modelId":"gpt-5.2","benchmarkId":"aa-livecodebench","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":89.41798941798939,"warning":null},{"id":"4fdab89ccf0e68eb33edd46b","modelId":"gpt-5.2","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":66.8783068783069,"warning":null},{"id":"c76eb6da2681e2ee97d971b0","modelId":"gpt-5.2","benchmarkId":"aa-livecodebench","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":88.8888888888889,"warning":null},{"id":"905fae76bc3338a4b1456688","modelId":"gpt-5.2","benchmarkId":"aa-mmlu-pro","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":85.85438829787229,"warning":null},{"id":"902ccf2fb17b5ce2a099b850","modelId":"gpt-5.2","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":81.358045212766,"warning":null},{"id":"73a1ebeaf1c6dff290e29d72","modelId":"gpt-5.2","benchmarkId":"aa-mmlu-pro","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":87.358710106383,"warning":null},{"id":"48df26e0495ecb16072f51f0","modelId":"gpt-5.2","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":74.5664739884393,"warning":null},{"id":"a3a15fb476b3147892ff7991","modelId":"gpt-5.2","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":65.8381502890173,"warning":null},{"id":"79c7838d7d3a6bb22b1ce139","modelId":"gpt-5.2","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":0.266666666666667,"warning":null},{"id":"818ef12ce3bfce5ac2c89f3b","modelId":"gpt-5.2","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":-12.25,"warning":null},{"id":"d01377d4aa4c240dca9f6a14","modelId":"gpt-5.2","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":-0.883333333333333,"warning":null},{"id":"b46eaf2b1f09b51bbf3647f6","modelId":"gpt-5.2","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":38.28333333333334,"warning":null},{"id":"1e7c4454251c39b4487c0012","modelId":"gpt-5.2","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":30.85,"warning":null},{"id":"c8fab29fc5407305d0f2ce16","modelId":"gpt-5.2","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":44.31666666666666,"warning":null},{"id":"1c1c6becf1972e34a043f8d6","modelId":"gpt-5.2","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":74.2690058479532,"warning":null},{"id":"c1395f4f5ccb6546d770922c","modelId":"gpt-5.2","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":46.4912280701754,"warning":null},{"id":"5a5011b42ee184885124646d","modelId":"gpt-5.2","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":84.7953216374269,"warning":null},{"id":"7064152db45f868fe5c922ea","modelId":"gpt-5.2","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":11.1340206185567,"warning":null},{"id":"e141c2a4b97814b006db03e6","modelId":"gpt-5.2","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-2-medium","score":43.1818181818182,"warning":null},{"id":"8a2cead721972b6b2ce7905a","modelId":"gpt-5.2","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-5.2 (Non-reasoning)","configuration":"GPT-5.2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-2-non-reasoning","score":31.818181818181802,"warning":null},{"id":"6a4f12a1c49afe58431fc0b3","modelId":"gpt-5.2","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.2 (Xhigh)","configuration":"GPT-5.2 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-2","score":46.969696969697,"warning":null},{"id":"b725135e8e60193d69dd2960","modelId":"gpt-5.2","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.2 (High)","configuration":"GPT-5.2 (High); Source model ID: gpt-5-2-2025-12-11-thinking-high; 120 tasks","score":43.33,"warning":null},{"id":"bdca95afb303a9bc58cb0fba","modelId":"gpt-5.2","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.2 (Low)","configuration":"GPT-5.2 (Low); Source model ID: gpt-5-2-2025-12-11-thinking-low; 120 tasks","score":9.719999999999999,"warning":null},{"id":"17b466b9eb5c749ece0b0106","modelId":"gpt-5.2","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.2 (Medium)","configuration":"GPT-5.2 (Medium); Source model ID: gpt-5-2-2025-12-11-thinking-medium; 120 tasks","score":26.669999999999998,"warning":null},{"id":"5967fa8ce31cc61440245a31","modelId":"gpt-5.2","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.2 (XHigh)","configuration":"GPT-5.2 (XHigh); Source model ID: gpt-5-2-2025-12-11-thinking-xhigh; 120 tasks","score":52.910000000000004,"warning":null},{"id":"26501f216454a4be44cafa23","modelId":"gpt-5.2-pro","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.2 Pro (High)","configuration":"GPT-5.2 Pro (High); Source model ID: gpt-5-2-pro-2025-12-11-high; 120 tasks","score":54.16,"warning":null},{"id":"c69d8d070ed466b0864fd277","modelId":"gpt-5.2-pro","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.2 Pro (Medium)","configuration":"GPT-5.2 Pro (Medium); Source model ID: gpt-5-2-pro-2025-12-11-medium; 120 tasks","score":38.47,"warning":null},{"id":"e134a7165dc311e60a10a59c","modelId":"gpt-5.3-codex","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":868.34,"warning":null},{"id":"d7701bab5c95acf22bfcb60a","modelId":"gpt-5.3-codex","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":83.3333333333333,"warning":null},{"id":"07b18dfea8a920cbb139c5b9","modelId":"gpt-5.3-codex","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":16.8571428571429,"warning":null},{"id":"045aef1a50daf39f6a1f6caf","modelId":"gpt-5.3-codex","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":91.5151515151515,"warning":null},{"id":"0840e875b7982341b208381c","modelId":"gpt-5.3-codex","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":42.493049119555096,"warning":null},{"id":"2c6f394ead006aaf6d42a2e2","modelId":"gpt-5.3-codex","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":75.3741496598639,"warning":null},{"id":"5456601870b518c709bb8764","modelId":"gpt-5.3-codex","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":78.4971098265896,"warning":null},{"id":"540b1d5498c3e0a301b1538c","modelId":"gpt-5.3-codex","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":10.8666666666667,"warning":null},{"id":"b44083a5a2aa4f26c30336f8","modelId":"gpt-5.3-codex","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":52.88333333333334,"warning":null},{"id":"e3d276610a33af6ec9ea0502","modelId":"gpt-5.3-codex","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":85.9649122807018,"warning":null},{"id":"b12c136f4b585aba8b52aec9","modelId":"gpt-5.3-codex","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.3 Codex (Xhigh)","configuration":"GPT-5.3 Codex (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-3-codex","score":53.030303030303,"warning":null},{"id":"8d150951509aabb2d05a87ff","modelId":"gpt-5.4","benchmarkId":"aa-apex-agents-aa","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":33.259587020649,"warning":null},{"id":"a5a3c88a6ef0e73b0d4b3c6b","modelId":"gpt-5.4","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":76.6666666666667,"warning":null},{"id":"95d78edd094271c8788986f1","modelId":"gpt-5.4","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":58.3333333333333,"warning":null},{"id":"764f953119825af98f18d356","modelId":"gpt-5.4","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":82,"warning":null},{"id":"7b2a25d7d5163219e7710897","modelId":"gpt-5.4","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":7.428571428571431,"warning":null},{"id":"2d984aba50e03049fd01ac4a","modelId":"gpt-5.4","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":0.571428571428571,"warning":null},{"id":"6c4fbe89fce7543c08625116","modelId":"gpt-5.4","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":23.42857143,"warning":null},{"id":"985716ec48075a5e66d619f1","modelId":"gpt-5.4","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":1246.23,"warning":null},{"id":"8cfb3855724b94948bece37c","modelId":"gpt-5.4","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":87.0707070707071,"warning":null},{"id":"39ae139dd203eb5f9760d5fe","modelId":"gpt-5.4","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":74.8484848484848,"warning":null},{"id":"c033cbce1b1f0ca4167fbd4e","modelId":"gpt-5.4","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":92.020202020202,"warning":null},{"id":"f05bf634d2eb678049d1cb04","modelId":"gpt-5.4","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":30.8155699721965,"warning":null},{"id":"b486a1fb6c5b341875d5225f","modelId":"gpt-5.4","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":11.2604263206673,"warning":null},{"id":"905144b36da3e71991cae33f","modelId":"gpt-5.4","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":43.7442075996293,"warning":null},{"id":"74502e659a17e8d3e5d5ec2b","modelId":"gpt-5.4","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":65.9183673469388,"warning":null},{"id":"df4e5119a87760b18053846e","modelId":"gpt-5.4","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":48.4353741496599,"warning":null},{"id":"2f1ec46058962a926609946a","modelId":"gpt-5.4","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":73.9455782312925,"warning":null},{"id":"c4d3c886f134af9a7091f0fd","modelId":"gpt-5.4","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":77.9768786127168,"warning":null},{"id":"3aea57a33fca8a2445fb7b9b","modelId":"gpt-5.4","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":70.5780346820809,"warning":null},{"id":"8b0420cc6bf11cd5fa24d863","modelId":"gpt-5.4","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":78.4393063583815,"warning":null},{"id":"8327c80a5c4f5dae3387a82f","modelId":"gpt-5.4","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":4.78333333333333,"warning":null},{"id":"0f3e47626675965f1c173e90","modelId":"gpt-5.4","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":-15.6666666666667,"warning":null},{"id":"f8d4a4b076f89d356e4af279","modelId":"gpt-5.4","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":5.78333333333333,"warning":null},{"id":"343892b2592f94ca09f096e9","modelId":"gpt-5.4","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":47.85,"warning":null},{"id":"a69fa647e71e7ea817df5065","modelId":"gpt-5.4","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":37.38333333333333,"warning":null},{"id":"0811ccb8f11abd0363b4d874","modelId":"gpt-5.4","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":50.849999999999994,"warning":null},{"id":"3937aa46e1ff32f1979abfe7","modelId":"gpt-5.4","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":74.56140350877189,"warning":null},{"id":"d29f2bc805408ffd2df1fcd8","modelId":"gpt-5.4","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":35.964912280701796,"warning":null},{"id":"33d3ddb29daf735d811d1107","modelId":"gpt-5.4","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":87.1345029239766,"warning":null},{"id":"a0bfa2a97f718615ed92100f","modelId":"gpt-5.4","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":39.587628865979404,"warning":null},{"id":"4262485dd1a36c6c772deb78","modelId":"gpt-5.4","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low","score":43.1818181818182,"warning":null},{"id":"9c16de1fecb482726fe8e501","modelId":"gpt-5.4","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-5.4 (Non-reasoning)","configuration":"GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning","score":37.878787878787904,"warning":null},{"id":"f0794fc8dd57dc2e00cdd8e9","modelId":"gpt-5.4","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":57.5757575757576,"warning":null},{"id":"f6121dbb26ebe1a9f0f5ad61","modelId":"gpt-5.4","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-5.4 (Xhigh)","configuration":"GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4","score":78.2771535580524,"warning":null},{"id":"b0856c22135884a80af840b3","modelId":"gpt-5.4","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.4 (High)","configuration":"GPT-5.4 (High); Source model ID: gpt-5-4-high; 120 tasks","score":67.5,"warning":null},{"id":"061ce8db3e09b74846381514","modelId":"gpt-5.4","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.4 (Low)","configuration":"GPT-5.4 (Low); Source model ID: gpt-5-4-low; 120 tasks","score":29.17,"warning":null},{"id":"f3c4874853691ebdb76d0676","modelId":"gpt-5.4","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.4 (Medium)","configuration":"GPT-5.4 (Medium); Source model ID: gpt-5-4-medium; 120 tasks","score":55.42,"warning":null},{"id":"5a6430a8d8f773ee0e4b369f","modelId":"gpt-5.4","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.4 (XHigh)","configuration":"GPT-5.4 (XHigh); Source model ID: gpt-5-4-xhigh; 120 tasks","score":73.95,"warning":null},{"id":"38295b12f6b510281d185208","modelId":"gpt-5.4","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"gpt-5-4","configuration":"gpt-5-4; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":51.8,"warning":null},{"id":"62c54d2819ec816195f7efee","modelId":"gpt-5.4","benchmarkId":"swepro-minisweagentstarredrows","level":"xhigh","sourceLabel":"gpt-5.4 (xHigh)*","configuration":"gpt-5.4 (xHigh)*; 731 tasks; source date 2026-04-08","score":59.1,"warning":null},{"id":"f4dcea334aae2dcb30b8753a","modelId":"gpt-5.4-mini","benchmarkId":"aa-aa-analyst-agent","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":15,"warning":null},{"id":"f3a670e32dc8bb7984855bb1","modelId":"gpt-5.4-mini","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":719.36,"warning":null},{"id":"67119219cf8635997e799f9f","modelId":"gpt-5.4-mini","benchmarkId":"aa-apex-agents-aa","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":28.1710914454277,"warning":null},{"id":"27f5e98d058de81cde243036","modelId":"gpt-5.4-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":67,"warning":null},{"id":"6c351ee13bd1f68934f58e90","modelId":"gpt-5.4-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":37,"warning":null},{"id":"90ff56149d01f09ea467bff3","modelId":"gpt-5.4-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":77,"warning":null},{"id":"5db5144008f8fd161f30920b","modelId":"gpt-5.4-mini","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":26.204887225711754,"warning":null},{"id":"168d4bb18b4ce6329b82875e","modelId":"gpt-5.4-mini","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":2.85714285714286,"warning":null},{"id":"79db0631bb8678befd1788bd","modelId":"gpt-5.4-mini","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":0,"warning":null},{"id":"ba497276202b46af9aaa9665","modelId":"gpt-5.4-mini","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":10,"warning":null},{"id":"679cf4bef4b9c7fec77fefa4","modelId":"gpt-5.4-mini","benchmarkId":"aa-enterprise-ops-gym-aa","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":34.61653237839451,"warning":null},{"id":"097b929efc58ff2f2c1b7d6e","modelId":"gpt-5.4-mini","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":13.4,"warning":null},{"id":"fbfb2fb210ec7cab45eadd48","modelId":"gpt-5.4-mini","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":589.8,"warning":null},{"id":"156d4c59162ba0c2f00f6d6c","modelId":"gpt-5.4-mini","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":1011.97,"warning":null},{"id":"c25fa22a8d22e9426e0e485e","modelId":"gpt-5.4-mini","benchmarkId":"aa-global-mmlu-lite","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":89.41145833333333,"warning":null},{"id":"8ff89c80ec855a2e1d71b572","modelId":"gpt-5.4-mini","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":82.3232323232323,"warning":null},{"id":"e76efd53f24c393a43e37889","modelId":"gpt-5.4-mini","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":60.606060606060595,"warning":null},{"id":"0d45f6a7bbb709c9c03a3d4f","modelId":"gpt-5.4-mini","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":87.4747474747475,"warning":null},{"id":"db6209ca9020f3dba7b70e8c","modelId":"gpt-5.4-mini","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":18.5820203892493,"warning":null},{"id":"9cc5343437a2443eea18334e","modelId":"gpt-5.4-mini","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":5.88507877664504,"warning":null},{"id":"961955935944d72739e414fc","modelId":"gpt-5.4-mini","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":28.127896200185397,"warning":null},{"id":"9df3b6e0118ce4c15b2d76fb","modelId":"gpt-5.4-mini","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":64.7619047619048,"warning":null},{"id":"5c1919a46d84b469ff8a026e","modelId":"gpt-5.4-mini","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":38.843537414966,"warning":null},{"id":"5b1f4258b7f0f468fec42b11","modelId":"gpt-5.4-mini","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":73.265306122449,"warning":null},{"id":"dae8251d27fe092deb17c0c8","modelId":"gpt-5.4-mini","benchmarkId":"aa-itbench-aa","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":18.6440677966102,"warning":null},{"id":"c66a4b8f9074d949c975af30","modelId":"gpt-5.4-mini","benchmarkId":"aa-itbench-aa","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":35.2165725047081,"warning":null},{"id":"109a3b874e68bd6343d2b61d","modelId":"gpt-5.4-mini","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":3.88888888888889,"warning":null},{"id":"efc87b8ffeabd7b6a2d5eeaf","modelId":"gpt-5.4-mini","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":71.1560693641618,"warning":null},{"id":"ae20ddcff96df7b14339a3f1","modelId":"gpt-5.4-mini","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":60.462427745664705,"warning":null},{"id":"c053fd4c43af856625108877","modelId":"gpt-5.4-mini","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":73.2947976878613,"warning":null},{"id":"2a3ca10746890e5987da723d","modelId":"gpt-5.4-mini","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":-20.6166666666667,"warning":null},{"id":"bb219decdc8cf7631a71f8aa","modelId":"gpt-5.4-mini","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":-45.4166666666667,"warning":null},{"id":"0df0c44f9dbd70c6d17a04aa","modelId":"gpt-5.4-mini","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":-18.9166666666667,"warning":null},{"id":"8feda783451d9249dcd37ef7","modelId":"gpt-5.4-mini","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":36.46666666666666,"warning":null},{"id":"b799fd73c716a16511ae6576","modelId":"gpt-5.4-mini","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":25.650000000000002,"warning":null},{"id":"eb04020587808233dab7c0f6","modelId":"gpt-5.4-mini","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":37.46666666666666,"warning":null},{"id":"faea98938a0a1f3c02346703","modelId":"gpt-5.4-mini","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":52.0833333333333,"warning":null},{"id":"8cfeb12d5ddccc5b75616261","modelId":"gpt-5.4-mini","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":36.5497076023392,"warning":null},{"id":"1680fa8b0c83fb8b292db2b7","modelId":"gpt-5.4-mini","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":23.3918128654971,"warning":null},{"id":"535e70954096d3be78fe68ac","modelId":"gpt-5.4-mini","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":83.3333333333333,"warning":null},{"id":"375c2967c2e20dc5932dbc08","modelId":"gpt-5.4-mini","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":25.5670103092783,"warning":null},{"id":"c7c874f90aba1735babc8ca6","modelId":"gpt-5.4-mini","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5.4 mini (Medium)","configuration":"GPT-5.4 mini (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-medium","score":34.0909090909091,"warning":null},{"id":"677fb8830d339b917cdff24d","modelId":"gpt-5.4-mini","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-5.4 mini (Non-reasoning)","configuration":"GPT-5.4 mini (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini-non-reasoning","score":18.181818181818198,"warning":null},{"id":"ea3c4baefc826e4cf0489915","modelId":"gpt-5.4-mini","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":52.2727272727273,"warning":null},{"id":"094a88e931ed8cb79e03e9e1","modelId":"gpt-5.4-mini","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":59.17602996254679,"warning":null},{"id":"8c534f14278a145657c92624","modelId":"gpt-5.4-mini","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-5.4 mini (Xhigh)","configuration":"GPT-5.4 mini (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-mini","score":2.02020202020202,"warning":null},{"id":"c23a7086cb8ecb85426614bd","modelId":"gpt-5.4-mini","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.4 Mini (High)","configuration":"GPT-5.4 Mini (High); Source model ID: gpt-5-4-mini-high; 120 tasks","score":13.19,"warning":null},{"id":"57762c5e8137a2a5daeba07d","modelId":"gpt-5.4-mini","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.4 Mini (Low)","configuration":"GPT-5.4 Mini (Low); Source model ID: gpt-5-4-mini-low; 120 tasks","score":1.11,"warning":null},{"id":"4da6e2b1da2186d54e3ce949","modelId":"gpt-5.4-mini","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.4 Mini (Medium)","configuration":"GPT-5.4 Mini (Medium); Source model ID: gpt-5-4-mini-medium; 120 tasks","score":4.44,"warning":null},{"id":"ebaca6d5960cef29ea2c77b1","modelId":"gpt-5.4-mini","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.4 Mini (XHigh)","configuration":"GPT-5.4 Mini (XHigh); Source model ID: gpt-5-4-mini-xhigh; 120 tasks","score":18.9,"warning":null},{"id":"2f5685afb6976287ee0b81c4","modelId":"gpt-5.4-nano","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":673.76,"warning":null},{"id":"c28732cd3202beee5519a723","modelId":"gpt-5.4-nano","benchmarkId":"aa-apex-agents-aa","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":24.9262536873156,"warning":null},{"id":"71a837acf7220d2d13abcd92","modelId":"gpt-5.4-nano","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":67.3333333333333,"warning":null},{"id":"4bb499f59408c5504907eeb9","modelId":"gpt-5.4-nano","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":29.666666666666703,"warning":null},{"id":"2d68f8bae21d36203a74c9eb","modelId":"gpt-5.4-nano","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":76.6666666666667,"warning":null},{"id":"038446f4f13088790def1885","modelId":"gpt-5.4-nano","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":5.9606130748688075,"warning":null},{"id":"7e12c82bf23acdaba510ebce","modelId":"gpt-5.4-nano","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":5.14285714285714,"warning":null},{"id":"c6e39241376d11151c4c491b","modelId":"gpt-5.4-nano","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":0,"warning":null},{"id":"91b5e4498893331f3c28839c","modelId":"gpt-5.4-nano","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":9.253061224489791,"warning":null},{"id":"349fa58171965483166888a6","modelId":"gpt-5.4-nano","benchmarkId":"aa-enterprise-ops-gym-aa","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":31.48313936138466,"warning":null},{"id":"138fde6fd3c91059ef85fe5d","modelId":"gpt-5.4-nano","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":7.8,"warning":null},{"id":"fc312817e4a743b969a50245","modelId":"gpt-5.4-nano","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":947.98,"warning":null},{"id":"af8a3988a2099d1bb8a0dc14","modelId":"gpt-5.4-nano","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":76.06060606060609,"warning":null},{"id":"f6262573cc28431e97fa1b2a","modelId":"gpt-5.4-nano","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":55.75757575757579,"warning":null},{"id":"671a524c59ee3135dfa9d88d","modelId":"gpt-5.4-nano","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":81.7171717171717,"warning":null},{"id":"bc47ef6088fca871d3605b27","modelId":"gpt-5.4-nano","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":15.940685820203901,"warning":null},{"id":"3ae2d833698c7e29b9838d81","modelId":"gpt-5.4-nano","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":4.07784986098239,"warning":null},{"id":"43df3efc42a8173597271f0d","modelId":"gpt-5.4-nano","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":28.2669138090825,"warning":null},{"id":"bfd66e4cd9770885e1c30c9c","modelId":"gpt-5.4-nano","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":64.421768707483,"warning":null},{"id":"3bb8de4bc88e68f472f250d2","modelId":"gpt-5.4-nano","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":32.7210884353741,"warning":null},{"id":"e3977c10d3d8a7d3d3731cee","modelId":"gpt-5.4-nano","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":75.9183673469388,"warning":null},{"id":"7e5c2c47c6c26f6be5cd751f","modelId":"gpt-5.4-nano","benchmarkId":"aa-itbench-aa","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":12.909604519774001,"warning":null},{"id":"f2eef5b9356421bd7029b6b9","modelId":"gpt-5.4-nano","benchmarkId":"aa-itbench-aa","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":24.3879472693032,"warning":null},{"id":"2a20eb32bca909faaa3ce8cc","modelId":"gpt-5.4-nano","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":5,"warning":null},{"id":"9ef8f4e9d62d606ed0745df4","modelId":"gpt-5.4-nano","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":59.5375722543353,"warning":null},{"id":"a4e2392412324897608ab870","modelId":"gpt-5.4-nano","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":43.757225433525996,"warning":null},{"id":"9b693f166694281a4fad6d5f","modelId":"gpt-5.4-nano","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":65.3757225433526,"warning":null},{"id":"41d8f84c096cece5209bb1f6","modelId":"gpt-5.4-nano","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":-18.0333333333333,"warning":null},{"id":"23051d6c0cf6193d2f9162bf","modelId":"gpt-5.4-nano","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":-37.4833333333333,"warning":null},{"id":"e86607181dc76cf771cc5a79","modelId":"gpt-5.4-nano","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":-29.4666666666667,"warning":null},{"id":"722798e2d210d540a3f7047a","modelId":"gpt-5.4-nano","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":21.9,"warning":null},{"id":"095fceb144c27b6d16b3e41e","modelId":"gpt-5.4-nano","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":14.883333333333335,"warning":null},{"id":"717fed908bdc4410bd07a602","modelId":"gpt-5.4-nano","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":25.666666666666664,"warning":null},{"id":"0052c5d0008eeb0968eb331d","modelId":"gpt-5.4-nano","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":47.2222222222222,"warning":null},{"id":"b6946c3f504304da43b0570a","modelId":"gpt-5.4-nano","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":52.631578947368396,"warning":null},{"id":"1fac58694796aca7430344f4","modelId":"gpt-5.4-nano","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":34.7953216374269,"warning":null},{"id":"c422b3a37e83a97394f36bb3","modelId":"gpt-5.4-nano","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":76.0233918128655,"warning":null},{"id":"e455a7d9cfac8748d05d0617","modelId":"gpt-5.4-nano","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":27.422680412371097,"warning":null},{"id":"9150027624e96121757dd6f9","modelId":"gpt-5.4-nano","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5.4 nano (Medium)","configuration":"GPT-5.4 nano (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-medium","score":33.3333333333333,"warning":null},{"id":"d5a1edb8823f03366085f611","modelId":"gpt-5.4-nano","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-5.4 nano (Non-reasoning)","configuration":"GPT-5.4 nano (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano-non-reasoning","score":24.2424242424242,"warning":null},{"id":"c473e412a77f5243debc0e49","modelId":"gpt-5.4-nano","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":42.4242424242424,"warning":null},{"id":"260fdde6b12aaec0d3c19fd3","modelId":"gpt-5.4-nano","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":60.674157303370805,"warning":null},{"id":"283ad2f31381bc489e150b0b","modelId":"gpt-5.4-nano","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-5.4 nano (Xhigh)","configuration":"GPT-5.4 nano (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-nano","score":0.505050505050505,"warning":null},{"id":"42f871c2b906d38c584a352f","modelId":"gpt-5.4-nano","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.4 Nano (High)","configuration":"GPT-5.4 Nano (High); Source model ID: gpt-5-4-nano-high; 120 tasks","score":3.61,"warning":null},{"id":"469ea5fe99e4ed4a10b50a81","modelId":"gpt-5.4-nano","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.4 Nano (Low)","configuration":"GPT-5.4 Nano (Low); Source model ID: gpt-5-4-nano-low; 120 tasks","score":1.53,"warning":null},{"id":"04abf124fc7c57207ee66816","modelId":"gpt-5.4-nano","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.4 Nano (Medium)","configuration":"GPT-5.4 Nano (Medium); Source model ID: gpt-5-4-nano-medium; 120 tasks","score":1.94,"warning":null},{"id":"a7b73b0067ddb6090993cb88","modelId":"gpt-5.4-nano","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.4 Nano (XHigh)","configuration":"GPT-5.4 Nano (XHigh); Source model ID: gpt-5-4-nano-xhigh; 120 tasks","score":5.6899999999999995,"warning":null},{"id":"398390040a395779ee1ab3e6","modelId":"gpt-5.4-pro","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.4 Pro (Xhigh)","configuration":"GPT-5.4 Pro (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4-pro","score":30,"warning":null},{"id":"eccabedcc721af9865bf90fa","modelId":"gpt-5.4-pro","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.4 Pro (XHigh)","configuration":"GPT-5.4 Pro (XHigh); Source model ID: gpt-5-4-pro-xhigh; 120 tasks","score":83.33,"warning":null},{"id":"6e174c25c1c54ebe5c182868","modelId":"gpt-5.5","benchmarkId":"aa-aa-analyst-agent","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":7.5,"warning":null},{"id":"8447b308cb52643f76659cb9","modelId":"gpt-5.5","benchmarkId":"aa-aa-analyst-agent","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":50,"warning":null},{"id":"ac7e4412596ec3269be37dfe","modelId":"gpt-5.5","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":1088.3,"warning":null},{"id":"5be2f3c5084cf188e2ad12f0","modelId":"gpt-5.5","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":1000,"warning":null},{"id":"f76606eb8665efe0cc94acdc","modelId":"gpt-5.5","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":1136.33,"warning":null},{"id":"4b7eadb3a41fa9e51738d293","modelId":"gpt-5.5","benchmarkId":"aa-apex-agents-aa","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":37.6843657817109,"warning":null},{"id":"c1f0381bd4dccf70ce8ad09b","modelId":"gpt-5.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":84.3333333333333,"warning":null},{"id":"9ff8665dd14c491fa507f93d","modelId":"gpt-5.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":81,"warning":null},{"id":"2aef968da0b7dcbe9b2a4f8a","modelId":"gpt-5.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":83,"warning":null},{"id":"5cc54c913f0c3c5fd1bfbe8e","modelId":"gpt-5.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":64,"warning":null},{"id":"8e3b4209492370f0ce174222","modelId":"gpt-5.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":84.3333333333333,"warning":null},{"id":"0527fab8e3f73649c1f07aec","modelId":"gpt-5.5","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":44.346163232163846,"warning":null},{"id":"788043295090ff8d1982b75b","modelId":"gpt-5.5","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":40.068834155236424,"warning":null},{"id":"756e89ee5a334f2879707255","modelId":"gpt-5.5","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":47.300565872933134,"warning":null},{"id":"a66efdcf3800246072abaf48","modelId":"gpt-5.5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":25.428571428571402,"warning":null},{"id":"cbc5aca2bc66670d8df90d81","modelId":"gpt-5.5","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":8,"warning":null},{"id":"39a9a43d4cbda096eb0e8bfe","modelId":"gpt-5.5","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":18.5714285714286,"warning":null},{"id":"c0135d32f920a750ca394238","modelId":"gpt-5.5","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":1.42857142857143,"warning":null},{"id":"d1caebb3d01c1e166bf0a53f","modelId":"gpt-5.5","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":27.142857142857103,"warning":null},{"id":"a0e936628db6f66a24b348d6","modelId":"gpt-5.5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":44.106236944195764,"warning":null},{"id":"e2b5f5642554c2ba630ceed6","modelId":"gpt-5.5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":46.64279319606088,"warning":null},{"id":"4f33080b5a24001a5b507280","modelId":"gpt-5.5","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":22.6,"warning":null},{"id":"7319032a9f99ea123027e144","modelId":"gpt-5.5","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":20.599999999999998,"warning":null},{"id":"eebad02b5627a3a6eff4688a","modelId":"gpt-5.5","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":21.2,"warning":null},{"id":"4db2c5d76189192d038130cf","modelId":"gpt-5.5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":1326.23,"warning":null},{"id":"35ad6daf35695cb216c54d06","modelId":"gpt-5.5","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":1044.57,"warning":null},{"id":"8dfe5c2c12597c2f148dc302","modelId":"gpt-5.5","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":1231,"warning":null},{"id":"5bb88901ae3a8781fbb78df4","modelId":"gpt-5.5","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":969.37,"warning":null},{"id":"1b44253c0ccde4008ded6177","modelId":"gpt-5.5","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":1350.71,"warning":null},{"id":"84b7c962e5ffc44d0064684a","modelId":"gpt-5.5","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":93.2323232323232,"warning":null},{"id":"e0c4aedfe3d9ed97841ea7c3","modelId":"gpt-5.5","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":91.010101010101,"warning":null},{"id":"365fca9f546e38e7608ece96","modelId":"gpt-5.5","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":92.6262626262626,"warning":null},{"id":"b28eee79d55650b860a152f1","modelId":"gpt-5.5","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":76.76767676767679,"warning":null},{"id":"277ed24c33c9f8e294e5be88","modelId":"gpt-5.5","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":93.53535353535351,"warning":null},{"id":"35f87856e4369e227ac225e6","modelId":"gpt-5.5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":45.0417052826691,"warning":null},{"id":"6ecbb1cb2c873550d0944fde","modelId":"gpt-5.5","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":32.6691380908248,"warning":null},{"id":"e2da5db70fe490f3a53fd6b9","modelId":"gpt-5.5","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":42.354031510658004,"warning":null},{"id":"be62a02da138b9b14310aa5d","modelId":"gpt-5.5","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":13.7164040778499,"warning":null},{"id":"6efefa9ee78eb42ac69860e8","modelId":"gpt-5.5","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":45.783132530120504,"warning":null},{"id":"aa809194fea064d6c44de156","modelId":"gpt-5.5","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":71.6326530612245,"warning":null},{"id":"0ddb8c7ac1883cd3e93ba625","modelId":"gpt-5.5","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":64.3537414965986,"warning":null},{"id":"0161c4026c859ad4135dccd2","modelId":"gpt-5.5","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":70.952380952381,"warning":null},{"id":"3e3eb1a94f0b4480f8138f70","modelId":"gpt-5.5","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":46.1224489795918,"warning":null},{"id":"d188208aca54d3b2df08c431","modelId":"gpt-5.5","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":75.8503401360544,"warning":null},{"id":"6382ecb472fba5f6ad8b1b96","modelId":"gpt-5.5","benchmarkId":"aa-itbench-aa","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":45.8097928436911,"warning":null},{"id":"5bbb35a2e804421c4a250b8f","modelId":"gpt-5.5","benchmarkId":"aa-mlcr-aa","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":11.1111111111111,"warning":null},{"id":"70f15d54df23ca3059f38070","modelId":"gpt-5.5","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":15,"warning":null},{"id":"de75f017388afc38d5fb596e","modelId":"gpt-5.5","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":17.7777777777778,"warning":null},{"id":"d94e16c27774d955afda494d","modelId":"gpt-5.5","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":81.0982658959538,"warning":null},{"id":"8e8b4883ee2fb9a44ea1adbb","modelId":"gpt-5.5","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":79.0173410404624,"warning":null},{"id":"fda66f410c38ba9b423e2b91","modelId":"gpt-5.5","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":81.1560693641619,"warning":null},{"id":"f3711a13a54eefb292d276f3","modelId":"gpt-5.5","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":71.38728323699421,"warning":null},{"id":"7a48f7dd0aa92254fa8e6e87","modelId":"gpt-5.5","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":79.8843930635838,"warning":null},{"id":"3c451d8f6634f505e51773a4","modelId":"gpt-5.5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":18.75,"warning":null},{"id":"b83d04c88c92bb3d49ebe682","modelId":"gpt-5.5","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":15.1,"warning":null},{"id":"1913dc520d4a89f0ff9dcb21","modelId":"gpt-5.5","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":18.1,"warning":null},{"id":"631dd98cd9c042f570e09cdd","modelId":"gpt-5.5","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":-4.75,"warning":null},{"id":"a19566dbb3e95e4e2873ca6e","modelId":"gpt-5.5","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":20.5166666666667,"warning":null},{"id":"4407296f514428b54d51f49e","modelId":"gpt-5.5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":57.03333333333334,"warning":null},{"id":"784f99dc176a7c47701f0fa0","modelId":"gpt-5.5","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":54.85,"warning":null},{"id":"1f595f6a19ed9d3a265db7d3","modelId":"gpt-5.5","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":56.8,"warning":null},{"id":"390acde9b4dd20f84fcd6e76","modelId":"gpt-5.5","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":45.63333333333333,"warning":null},{"id":"46829027426de35ce07d1743","modelId":"gpt-5.5","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":57.95,"warning":null},{"id":"e7f175b942e3a8b257d84b73","modelId":"gpt-5.5","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":56.1342592592593,"warning":null},{"id":"f655094357b4e6bb3de7e139","modelId":"gpt-5.5","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":54.51388888888889,"warning":null},{"id":"7091723aa1a23c7293d6469e","modelId":"gpt-5.5","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":55.787037037037,"warning":null},{"id":"2df219a851c5f3fea09eb195","modelId":"gpt-5.5","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":92.9824561403509,"warning":null},{"id":"98d49812f963462146542c3c","modelId":"gpt-5.5","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":83.9181286549708,"warning":null},{"id":"d8949f43a254fce3d60c82c8","modelId":"gpt-5.5","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":91.812865497076,"warning":null},{"id":"e02f2d421477234c78032860","modelId":"gpt-5.5","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":69.2982456140351,"warning":null},{"id":"21548eb9e992b982e304f90a","modelId":"gpt-5.5","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":93.859649122807,"warning":null},{"id":"f96023916aee097e66012d15","modelId":"gpt-5.5","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":36.701030927835,"warning":null},{"id":"ba56046afb9c6b00c25603de","modelId":"gpt-5.5","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":24.9484536082474,"warning":null},{"id":"54e5c007f6222ad68f3dc703","modelId":"gpt-5.5","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":29.8969072164948,"warning":null},{"id":"3588eb89e953dc5abe08662f","modelId":"gpt-5.5","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":14.8453608247423,"warning":null},{"id":"73415c5eb78a63d5c5049fe5","modelId":"gpt-5.5","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":38.9690721649485,"warning":null},{"id":"6965d3a99836b2daaea2f5de","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":59.848484848484894,"warning":null},{"id":"b7563525357509b58d5825ea","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":52.2727272727273,"warning":null},{"id":"db1f5b3ae21cdd7aba252f4c","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":57.5757575757576,"warning":null},{"id":"7d2fcfb4c90f4384817d28c2","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":49.2424242424242,"warning":null},{"id":"c12cc75753862e4839c1ae41","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":60.606060606060595,"warning":null},{"id":"83f097bf99d86c8df8d2f214","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":79.40074906367039,"warning":null},{"id":"b9db01a719a53679d36d93b9","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-5-low","score":65.5430711610487,"warning":null},{"id":"87a08256a80867ff57bc8c92","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":80.5243445692884,"warning":null},{"id":"111185a60b03a16824fda31d","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GPT-5.5 (Non-reasoning)","configuration":"GPT-5.5 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-5-non-reasoning","score":61.0486891385768,"warning":null},{"id":"1852189c1ef21608411b7478","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":84.2696629213483,"warning":null},{"id":"25694e0ba1d922c8b1c8c566","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Model source: https://artificialanalysis.ai/models/gpt-5-5-high","score":9.09090909090909,"warning":null},{"id":"b2dba53b441d0e54397d2ba9","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-5-medium","score":5.05050505050505,"warning":null},{"id":"f42ad83363dfb6ba7ee246b6","modelId":"gpt-5.5","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-5.5 (Xhigh)","configuration":"GPT-5.5 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5","score":14.6464646464646,"warning":null},{"id":"bc45f7c6e1d0aab8b73e7aec","modelId":"gpt-5.5","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.5 (High)","configuration":"GPT-5.5 (High); Source model ID: gpt-5-5-2026-04-22-thinking-high; 120 tasks","score":83.33,"warning":null},{"id":"205dbfe9537d5cfd1c612317","modelId":"gpt-5.5","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.5 (Low)","configuration":"GPT-5.5 (Low); Source model ID: gpt-5-5-2026-04-22-thinking-low; 120 tasks","score":33.33,"warning":null},{"id":"58c732552823494e3e219dc6","modelId":"gpt-5.5","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.5 (Medium)","configuration":"GPT-5.5 (Medium); Source model ID: gpt-5-5-2026-04-22-thinking-medium; 120 tasks","score":70.42,"warning":null},{"id":"e082fe244b558a6479f48c03","modelId":"gpt-5.5","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.5 (XHigh)","configuration":"GPT-5.5 (XHigh); Source model ID: gpt-5-5-2026-04-22-thinking-xhigh; 120 tasks","score":85,"warning":null},{"id":"073ccc861e30db90aaf28ba0","modelId":"gpt-5.5","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gpt-5-5","configuration":"gpt-5-5; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":64.4,"warning":null},{"id":"552921ea13a22c9abdd154aa","modelId":"gpt-5.5","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"gpt-5-5","configuration":"gpt-5-5; Reported effort: low; 113 unique tasks; 452 attempts; 4 runs","score":27,"warning":null},{"id":"83db8e5e910b00269a48317e","modelId":"gpt-5.5","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"gpt-5-5","configuration":"gpt-5-5; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":54,"warning":null},{"id":"2171f2252c414c3ccf229887","modelId":"gpt-5.5","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"gpt-5-5","configuration":"gpt-5-5; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":67,"warning":null},{"id":"e58e8485712a3d724ce4a879","modelId":"gpt-5.5","benchmarkId":"terminal-codex","level":"xhigh","sourceLabel":"GPT-5.5","configuration":"GPT-5.5; Reported effort: xhigh; 445 trials; recorded 2026-04-23","score":83.15,"warning":null},{"id":"084e9dff682a86761c6ade08","modelId":"gpt-5.5","benchmarkId":"terminal-terminus2","level":"xhigh","sourceLabel":"GPT-5.5","configuration":"GPT-5.5; Reported effort: xhigh; 445 trials; recorded 2026-04-23","score":77.98,"warning":null},{"id":"54794aa73a19cc681202d4e2","modelId":"gpt-5.5-pro","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.5 Pro (Xhigh)","configuration":"GPT-5.5 Pro (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-5-pro","score":30.571428571428598,"warning":null},{"id":"4e24cf58ae552e5e1cac37b5","modelId":"gpt-5.5-pro","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.5 Pro (High)","configuration":"GPT-5.5 Pro (High); Source model ID: gpt-5-5-pro-2026-04-23-high; 120 tasks","score":84.58,"warning":null},{"id":"4e4b0ebdcce2198efad4878e","modelId":"gpt-5.5-pro","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.5 Pro (XHigh)","configuration":"GPT-5.5 Pro (XHigh); Source model ID: gpt-5-5-pro-2026-04-23-xhigh; 120 tasks","score":84.16,"warning":null},{"id":"08aa0fc7cbca475beb3f974c","modelId":"gpt-5.6-luna","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":1170.08,"warning":null},{"id":"0fedb2309ba96b9256510144","modelId":"gpt-5.6-luna","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":740.69,"warning":null},{"id":"8f4bb6a39d9752cd476c93ae","modelId":"gpt-5.6-luna","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":1342.18,"warning":null},{"id":"7c0d1942bdee5013c3a8b09d","modelId":"gpt-5.6-luna","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":934.59,"warning":null},{"id":"21902bcc1072ab9edef4a340","modelId":"gpt-5.6-luna","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":706.04,"warning":null},{"id":"d860331acf0d88b156a0caf4","modelId":"gpt-5.6-luna","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":1267.93,"warning":null},{"id":"6204fb699cf808ff33c15137","modelId":"gpt-5.6-luna","benchmarkId":"aa-apex-agents-aa","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":35.8407079646018,"warning":null},{"id":"f873a94e10faa93ad7969449","modelId":"gpt-5.6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":80.3333333333333,"warning":null},{"id":"df3e2998117eaf50d14576ab","modelId":"gpt-5.6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":70,"warning":null},{"id":"f07c3470a2b3eecbe9404152","modelId":"gpt-5.6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":83.6666666666667,"warning":null},{"id":"74cdd7d809ccfdef6ac33c29","modelId":"gpt-5.6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":75,"warning":null},{"id":"86b4187194a5b9f269e16b79","modelId":"gpt-5.6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":42.6666666666667,"warning":null},{"id":"40ba3b930e45bdde10698b65","modelId":"gpt-5.6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":81.6666666666667,"warning":null},{"id":"4cce229e4e804916c3dab249","modelId":"gpt-5.6-luna","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":35.58794293060383,"warning":null},{"id":"50da7cba304a4fc4eee85e6f","modelId":"gpt-5.6-luna","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":11.692162258271626,"warning":null},{"id":"b3de4ac001936b757080707e","modelId":"gpt-5.6-luna","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":50.208617637561936,"warning":null},{"id":"64c0381c899748e1da7f3c5f","modelId":"gpt-5.6-luna","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":23.074538803075942,"warning":null},{"id":"90813cb1780cc300fd82336b","modelId":"gpt-5.6-luna","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":4.970838599882015,"warning":null},{"id":"02f90ea3ac2e8f0f420b0956","modelId":"gpt-5.6-luna","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":42.597767374285944,"warning":null},{"id":"1b6109901dfeb427f2188d86","modelId":"gpt-5.6-luna","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":16.5714285714286,"warning":null},{"id":"eb2892b4aa5b5a6f31c8815e","modelId":"gpt-5.6-luna","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":2.57142857142857,"warning":null},{"id":"cb7fbf8730236ac18d53738f","modelId":"gpt-5.6-luna","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":20.571428571428598,"warning":null},{"id":"895f630765dd9e748563ce85","modelId":"gpt-5.6-luna","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":4.857142857142859,"warning":null},{"id":"db6c8eebd3a0ec57c7656dd0","modelId":"gpt-5.6-luna","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"c9fc885eb1ff9ed5623036cb","modelId":"gpt-5.6-luna","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":20.571428571428598,"warning":null},{"id":"f0c47d410dab8bc168714164","modelId":"gpt-5.6-luna","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":40.82363473589973,"warning":null},{"id":"50e3b87352c4319addeb32cf","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":22.2,"warning":null},{"id":"67706ca27c52fa69c458aac4","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":14.2,"warning":null},{"id":"9b4e143d5c7b4834f34e7766","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":24,"warning":null},{"id":"42adc09bb396e1d056940208","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":15.2,"warning":null},{"id":"64e7c821e87e87e0d6ef3a9e","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":6,"warning":null},{"id":"82e0f29b5e292046023abf03","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":23.799999999999997,"warning":null},{"id":"167e957b4ef4690d6c6d333b","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":1330.1,"warning":null},{"id":"000286c76c81d5044c73521f","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":995.33,"warning":null},{"id":"479122ab39199257813ffcf1","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":1459.2,"warning":null},{"id":"d9e31923f53e7a8cf8579328","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":1129.07,"warning":null},{"id":"7f8f2987d382c7616999da3c","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":919.68,"warning":null},{"id":"879b5c0cf80e8e458e144bda","modelId":"gpt-5.6-luna","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":1399.92,"warning":null},{"id":"719891b90b3ab5d162d55c96","modelId":"gpt-5.6-luna","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":89.1919191919192,"warning":null},{"id":"ee2471d6441caa758f861e91","modelId":"gpt-5.6-luna","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":83.5353535353535,"warning":null},{"id":"252336018d7d21ed5c89686d","modelId":"gpt-5.6-luna","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":91.1111111111111,"warning":null},{"id":"b8c5132b4259c9b6b9a14740","modelId":"gpt-5.6-luna","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":85.8585858585859,"warning":null},{"id":"c15d04aa1dd9a38f265a487b","modelId":"gpt-5.6-luna","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":64.54545454545459,"warning":null},{"id":"c6ab252112bb290154fb98ed","modelId":"gpt-5.6-luna","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":89.49494949494951,"warning":null},{"id":"0aaf98457fc038ca823735a9","modelId":"gpt-5.6-luna","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":33.4105653382762,"warning":null},{"id":"ca3884f250021f03228f0cc2","modelId":"gpt-5.6-luna","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":19.8331788693234,"warning":null},{"id":"df49f4b1845c75c4f4cf7c0b","modelId":"gpt-5.6-luna","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":39.4810009267841,"warning":null},{"id":"6e79d47137a9f94197aad4c0","modelId":"gpt-5.6-luna","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":25.7645968489342,"warning":null},{"id":"47864eccd9ca9b39975cabb8","modelId":"gpt-5.6-luna","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":7.228915662650599,"warning":null},{"id":"3ed4fd66623c20d30d99f05e","modelId":"gpt-5.6-luna","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":36.9786839666358,"warning":null},{"id":"353b90d7ae0b3cecda59d5a4","modelId":"gpt-5.6-luna","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":40.3201506591337,"warning":null},{"id":"c42bcc1a3297a9a594553544","modelId":"gpt-5.6-luna","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":12.2222222222222,"warning":null},{"id":"8b4b25cefb540449881e504f","modelId":"gpt-5.6-luna","benchmarkId":"aa-mlcr-aa","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":6.11111111111111,"warning":null},{"id":"35325e3b3863b7610f9365a2","modelId":"gpt-5.6-luna","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":19.4444444444444,"warning":null},{"id":"d567cb24df745d3868299233","modelId":"gpt-5.6-luna","benchmarkId":"aa-mlcr-aa","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":8.33333333333333,"warning":null},{"id":"75d195a2b4e88614b35d9b73","modelId":"gpt-5.6-luna","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":16.1111111111111,"warning":null},{"id":"7efca93edb1e688a87e88fb1","modelId":"gpt-5.6-luna","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":77.5722543352601,"warning":null},{"id":"2b78b930ad5a95938742261c","modelId":"gpt-5.6-luna","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":73.9884393063584,"warning":null},{"id":"419ed89c76800e7e0148bfee","modelId":"gpt-5.6-luna","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":78.5549132947977,"warning":null},{"id":"34f045ba84c6d8752d46de97","modelId":"gpt-5.6-luna","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":75.8381502890173,"warning":null},{"id":"0dfef61c8d999d26f5ea98c9","modelId":"gpt-5.6-luna","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":60.346820809248605,"warning":null},{"id":"cf82863603e1709cf7467526","modelId":"gpt-5.6-luna","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":78.5549132947977,"warning":null},{"id":"f266a9323e42c26d29c56798","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":-12,"warning":null},{"id":"ef6a561e2c83715a872c7f7a","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":-14.65,"warning":null},{"id":"02de7c94d0ab8b2ceba4512c","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":-10.2833333333333,"warning":null},{"id":"b13a8bac5c637ce921467dbb","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":-13.1833333333333,"warning":null},{"id":"150c85bb1557c00c7a6d1ffe","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":-24.95,"warning":null},{"id":"05676a264b60533c21673971","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":-10.7666666666667,"warning":null},{"id":"765f9760ebc1e5891c0325c9","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":41.78333333333333,"warning":null},{"id":"ba34cfb3e575aca0db7d5ec7","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":39.61666666666667,"warning":null},{"id":"a359f1bf8301451e36017e73","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":42.733333333333334,"warning":null},{"id":"011c1228b7cd04048cf72e29","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":40.699999999999996,"warning":null},{"id":"d46bbeaeae22d52a50c1fecc","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":28.599999999999998,"warning":null},{"id":"f152eb3ee53db8d1016475f2","modelId":"gpt-5.6-luna","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":42.449999999999996,"warning":null},{"id":"f7a065b0663ae0a319240f74","modelId":"gpt-5.6-luna","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":51.6203703703704,"warning":null},{"id":"46a2ab6ca51572c3af625d9f","modelId":"gpt-5.6-luna","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":46.064814814814795,"warning":null},{"id":"ffd31be8a2c5dfaad4c42a10","modelId":"gpt-5.6-luna","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":53.587962962963,"warning":null},{"id":"a1e6634a8ad09dda7a7f2e0d","modelId":"gpt-5.6-luna","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":46.7592592592593,"warning":null},{"id":"a26faa4f2758b30066d00e5c","modelId":"gpt-5.6-luna","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":40.3935185185185,"warning":null},{"id":"8c33016fdd4bbc68eb016673","modelId":"gpt-5.6-luna","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":50.462962962963,"warning":null},{"id":"27d42c4057d81fe9fde8718b","modelId":"gpt-5.6-luna","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":25.1546391752577,"warning":null},{"id":"6bf2923f73cdd8b5a14eee4b","modelId":"gpt-5.6-luna","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":12.7835051546392,"warning":null},{"id":"1f573add80b4ffbb5c561656","modelId":"gpt-5.6-luna","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":31.1340206185567,"warning":null},{"id":"223d4f4d43bbb11c9f4f48cb","modelId":"gpt-5.6-luna","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":17.7319587628866,"warning":null},{"id":"34aa71f8070e9a1e6ef3977f","modelId":"gpt-5.6-luna","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":9.69072164948454,"warning":null},{"id":"205af12b1ce13f08393cce88","modelId":"gpt-5.6-luna","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":28.659793814432998,"warning":null},{"id":"033e22dfc34709d054844db8","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":3.33333333333333,"warning":null},{"id":"bdfd86dfae6f8a1edf289a8c","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":69.6629213483146,"warning":null},{"id":"565264603a1349a94c3faca5","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":43.4456928838951,"warning":null},{"id":"9bb0ee056335a7479d3a2d36","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":80.8988764044944,"warning":null},{"id":"84b0115dee18016a84dd76e3","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":53.1835205992509,"warning":null},{"id":"4d5b3002c73d51ef146aa99d","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":38.9513108614232,"warning":null},{"id":"b634f6cd42e7e240acfc27d4","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":77.9026217228464,"warning":null},{"id":"cf30810d32c1ace255555908","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-high","score":2.52525252525253,"warning":null},{"id":"27ccfd0b69711c209adf87c2","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-low","score":0,"warning":null},{"id":"45d3c21755a4aaab0fb5ad1f","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna","score":11.6161616161616,"warning":null},{"id":"86746d003c1cf11eb233e512","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-medium","score":0.505050505050505,"warning":null},{"id":"c1c16566cefc5721416052dd","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"GPT-5.6 Luna (Non-reasoning)","configuration":"GPT-5.6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-non-reasoning","score":1.01010101010101,"warning":null},{"id":"67bc4f2a4b9cf0a0931d3aa7","modelId":"gpt-5.6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-5.6 Luna (Xhigh)","configuration":"GPT-5.6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-luna-xhigh","score":3.53535353535354,"warning":null},{"id":"5d5408de712e8c20684fe7d9","modelId":"gpt-5.6-luna","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.6 Luna (High)","configuration":"GPT-5.6 Luna (High); Source model ID: openai-gpt-5-6-luna-high; 120 tasks","score":29.310000000000002,"warning":null},{"id":"6ffbc8e19c59582d10c22554","modelId":"gpt-5.6-luna","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.6 Luna (Low)","configuration":"GPT-5.6 Luna (Low); Source model ID: openai-gpt-5-6-luna-low; 120 tasks","score":5.140000000000001,"warning":null},{"id":"b2854ee06891af72b895e0a6","modelId":"gpt-5.6-luna","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GPT-5.6 Luna (Max)","configuration":"GPT-5.6 Luna (Max); Source model ID: openai-gpt-5-6-luna-max; 120 tasks","score":59.540000000000006,"warning":null},{"id":"58b2a2ac3a81857ae37fac17","modelId":"gpt-5.6-luna","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.6 Luna (Medium)","configuration":"GPT-5.6 Luna (Medium); Source model ID: openai-gpt-5-6-luna-medium; 120 tasks","score":7.359999999999999,"warning":null},{"id":"d97b14ae64e526fca71b0f14","modelId":"gpt-5.6-luna","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.6 Luna (XHigh)","configuration":"GPT-5.6 Luna (XHigh); Source model ID: openai-gpt-5-6-luna-xhigh; 120 tasks","score":47.64,"warning":null},{"id":"d581184225be8abecad4b42a","modelId":"gpt-5.6-luna","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gpt-5-6-luna","configuration":"gpt-5-6-luna; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":44.2,"warning":null},{"id":"b56c7ef4cb20e19925c7bcbd","modelId":"gpt-5.6-luna","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"gpt-5-6-luna","configuration":"gpt-5-6-luna; Reported effort: low; 113 unique tasks; 452 attempts; 4 runs","score":1.5,"warning":null},{"id":"41d96fb8508ef9bd7eca956c","modelId":"gpt-5.6-luna","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"gpt-5-6-luna","configuration":"gpt-5-6-luna; Reported effort: max; 113 unique tasks; 448 attempts; 4 runs","score":67.2,"warning":null},{"id":"ce146c74890ffa309e9e6ac3","modelId":"gpt-5.6-luna","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"gpt-5-6-luna","configuration":"gpt-5-6-luna; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":11.3,"warning":null},{"id":"2b08fc13fce99f9456b958dd","modelId":"gpt-5.6-luna","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"gpt-5-6-luna","configuration":"gpt-5-6-luna; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":56.9,"warning":null},{"id":"d6552d55b1ffb5a8f8cd5cc5","modelId":"gpt-5.6-luna","benchmarkId":"terminal-codex","level":"max","sourceLabel":"GPT-5.6 Luna","configuration":"GPT-5.6 Luna; Reported effort: max; 445 trials; recorded 2026-06-26","score":75.73,"warning":null},{"id":"9974c0ea34e19312c78d5770","modelId":"gpt-5.6-sol","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":47.5,"warning":null},{"id":"81b9d211ae78c0cad059745a","modelId":"gpt-5.6-sol","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":1368.7,"warning":null},{"id":"fc8f0c47484181ea78848108","modelId":"gpt-5.6-sol","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":1043.52,"warning":null},{"id":"68faa35879d24af6ad12b0d6","modelId":"gpt-5.6-sol","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":1480.04,"warning":null},{"id":"933190fe7f455d8cf2c6f19e","modelId":"gpt-5.6-sol","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":1241.93,"warning":null},{"id":"d45d83ea3e76e9c9d4c75e6a","modelId":"gpt-5.6-sol","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":1008.14,"warning":null},{"id":"3c577e9bddf8069b0e8e9a96","modelId":"gpt-5.6-sol","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":1437.45,"warning":null},{"id":"d84a692d121b29795bbe004c","modelId":"gpt-5.6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":81.6666666666667,"warning":null},{"id":"9d4a45654187bb1f2e546cba","modelId":"gpt-5.6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":78,"warning":null},{"id":"0949add7de44daff00be3446","modelId":"gpt-5.6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":84,"warning":null},{"id":"7f0e79993e50fe12fe784d93","modelId":"gpt-5.6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":80.3333333333333,"warning":null},{"id":"5b477cc013715702362456cb","modelId":"gpt-5.6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":62.33333333333329,"warning":null},{"id":"c9ee66f9097d1fd5c16728c3","modelId":"gpt-5.6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":82.3333333333333,"warning":null},{"id":"3784e6ba61fee5c664f55ba7","modelId":"gpt-5.6-sol","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":55.31220357393729,"warning":null},{"id":"bf6b0b3bc6a82eaf37609ef3","modelId":"gpt-5.6-sol","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":40.985670113448045,"warning":null},{"id":"51211e7755dad5690c7acb4b","modelId":"gpt-5.6-sol","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":60.08114996276329,"warning":null},{"id":"80e219cdeaff5bb42cb63eeb","modelId":"gpt-5.6-sol","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":51.327511588639865,"warning":null},{"id":"c6241986485aa5caf72f6404","modelId":"gpt-5.6-sol","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":22.656758038944595,"warning":null},{"id":"55315b8aeb43698d12e9c8d4","modelId":"gpt-5.6-sol","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":55.311672693269884,"warning":null},{"id":"58e8ffc37ade324e60ee76c5","modelId":"gpt-5.6-sol","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":25.7142857142857,"warning":null},{"id":"4d4a237de764fc96a2af3767","modelId":"gpt-5.6-sol","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":14.857142857142899,"warning":null},{"id":"ee873572b5016f32a17adf6a","modelId":"gpt-5.6-sol","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":32.2857142857143,"warning":null},{"id":"48d16f18482b65e6db3ed884","modelId":"gpt-5.6-sol","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":22.8571428571429,"warning":null},{"id":"9b9dbe4777c1a4c323eb5d01","modelId":"gpt-5.6-sol","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":5.14285714285714,"warning":null},{"id":"03544089b78511ac2a3d4880","modelId":"gpt-5.6-sol","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":28.571428571428598,"warning":null},{"id":"5cbcdffcde150028bb82c601","modelId":"gpt-5.6-sol","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":42.9125634139063,"warning":null},{"id":"905e88fb59f70e49b2f25b8a","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":27.800000000000004,"warning":null},{"id":"1cce76de0b10f21ef2931606","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":21,"warning":null},{"id":"34fc58a4912299d9ddca6389","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":27.200000000000003,"warning":null},{"id":"94672efadcafeff83ea21b8d","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":26.200000000000003,"warning":null},{"id":"2fc96191e6f022b6e6ef264b","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":15.2,"warning":null},{"id":"420ccc50bf25589411df39f9","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":27.6,"warning":null},{"id":"89096765cda066fca4b9b456","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":1503.99,"warning":null},{"id":"423b077dbc952dfbfb035080","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":1303.39,"warning":null},{"id":"d0036b767ccd39f82a1fc5cb","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":1608.63,"warning":null},{"id":"ed133e797ea85e0d21b4caf4","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":1421.47,"warning":null},{"id":"511c156bf87c77e46f1d890c","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":1241.59,"warning":null},{"id":"d9a54e1ee449fe8e25fb0687","modelId":"gpt-5.6-sol","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":1568.48,"warning":null},{"id":"99a80542e4d6a00085fd5b39","modelId":"gpt-5.6-sol","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":92.82828282828281,"warning":null},{"id":"76f3825292ae839cd4487be0","modelId":"gpt-5.6-sol","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":89.79797979797979,"warning":null},{"id":"24abdda42bb57668a30cbc8b","modelId":"gpt-5.6-sol","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":94.1414141414141,"warning":null},{"id":"5663f564b9b58369d0393d1b","modelId":"gpt-5.6-sol","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":92.6262626262626,"warning":null},{"id":"5c89f4aedb58325a99416195","modelId":"gpt-5.6-sol","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":78.989898989899,"warning":null},{"id":"42c55b74f85d8f93a7ac01e3","modelId":"gpt-5.6-sol","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":93.13131313131309,"warning":null},{"id":"15d5e2915db8e15020f5dbb2","modelId":"gpt-5.6-sol","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":46.014828544949,"warning":null},{"id":"7e0b30f72b916805da2bc9c3","modelId":"gpt-5.6-sol","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":39.3883225208526,"warning":null},{"id":"8eaca8e31c2a7d563051a5fb","modelId":"gpt-5.6-sol","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":49.4902687673772,"warning":null},{"id":"93cac4d1e442b4c0c54f4a46","modelId":"gpt-5.6-sol","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":42.2150139017609,"warning":null},{"id":"46dab6b30ddf4f6a6277b5bb","modelId":"gpt-5.6-sol","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":16.682113067655198,"warning":null},{"id":"787b66d51ef091b22ac10ffb","modelId":"gpt-5.6-sol","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":47.3123262279889,"warning":null},{"id":"83ee7bbaa5032f9f8a963ad7","modelId":"gpt-5.6-sol","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":69.1836734693878,"warning":null},{"id":"504a14bfe0bb0c5dfef8d170","modelId":"gpt-5.6-sol","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":66.53061224489801,"warning":null},{"id":"4e2d790094a23ad881bce27c","modelId":"gpt-5.6-sol","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":72.6530612244898,"warning":null},{"id":"961f8f1d764728d0e1e39a20","modelId":"gpt-5.6-sol","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":69.5918367346939,"warning":null},{"id":"f6219d909dac73cd86792da7","modelId":"gpt-5.6-sol","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":71.0204081632653,"warning":null},{"id":"592a9019653ee03656236054","modelId":"gpt-5.6-sol","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":56.2146892655367,"warning":null},{"id":"b51320fe644ad06aacca006d","modelId":"gpt-5.6-sol","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":17.2222222222222,"warning":null},{"id":"3dcab0edb14495d99f66c983","modelId":"gpt-5.6-sol","benchmarkId":"aa-mlcr-aa","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":12.7777777777778,"warning":null},{"id":"5fe9f9ad2696d4dc6a484ed9","modelId":"gpt-5.6-sol","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":26.1111111111111,"warning":null},{"id":"12493385b9c269d4ac32fd4a","modelId":"gpt-5.6-sol","benchmarkId":"aa-mlcr-aa","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":15,"warning":null},{"id":"697d92fb507d15f7cf0d1651","modelId":"gpt-5.6-sol","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":18.3333333333333,"warning":null},{"id":"e8111dd565adfca1c96d9aa9","modelId":"gpt-5.6-sol","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":81.84971098265899,"warning":null},{"id":"7d5d07ae05a49610c5fc7b40","modelId":"gpt-5.6-sol","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":80.9826589595376,"warning":null},{"id":"dff2cbcb0b8f095c8a943017","modelId":"gpt-5.6-sol","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":83.4104046242775,"warning":null},{"id":"ffbb3fbc2aed5615f0458c6c","modelId":"gpt-5.6-sol","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":81.38728323699421,"warning":null},{"id":"eaf5836dc0d624ab48f5d238","modelId":"gpt-5.6-sol","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":71.90751445086701,"warning":null},{"id":"3e9584bda97637ed86511f62","modelId":"gpt-5.6-sol","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":82.6589595375723,"warning":null},{"id":"ee678562d47d22a6875558b3","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":20.3666666666667,"warning":null},{"id":"a63ab95a8a408b289abbb46d","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":18.8666666666667,"warning":null},{"id":"d23f488c99b2190b4e0982c2","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":21.9666666666667,"warning":null},{"id":"f646f68c571f26eb0fc17d00","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":19.4,"warning":null},{"id":"52cbe7096d25c4e59ae6c392","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":1.1,"warning":null},{"id":"b79e0ff46dc455df5138ba5f","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":20.9833333333333,"warning":null},{"id":"abacf66c74669ec1a527df88","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":58.35,"warning":null},{"id":"7a2cf256d98c551ae91c077c","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":57.166666666666664,"warning":null},{"id":"ef0a95ab0ef161b90b153d47","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":59.4,"warning":null},{"id":"63ba00757b6dd3d0a0aa7083","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":57.75,"warning":null},{"id":"d8c39b5b1f8cffe20bc488a4","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":48.699999999999996,"warning":null},{"id":"eccf78b96d6ca3cbcab2e725","modelId":"gpt-5.6-sol","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":58.81666666666666,"warning":null},{"id":"709e81a0a33a47d464af44aa","modelId":"gpt-5.6-sol","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":57.7546296296296,"warning":null},{"id":"c2177dc76be61711d1a61b06","modelId":"gpt-5.6-sol","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":56.3657407407407,"warning":null},{"id":"b155134d83a92a8e34c71988","modelId":"gpt-5.6-sol","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":57.0601851851852,"warning":null},{"id":"3136590de59224790e6fd89a","modelId":"gpt-5.6-sol","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":57.4074074074074,"warning":null},{"id":"b1479cb06719eda54907a9d9","modelId":"gpt-5.6-sol","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":47.685185185185205,"warning":null},{"id":"c4809acc90ab2c62f55cab04","modelId":"gpt-5.6-sol","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":57.0601851851852,"warning":null},{"id":"a1c30d3ea06981b21ed080e5","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":83.3333333333333,"warning":null},{"id":"471380da766f31b223124640","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":76.0233918128655,"warning":null},{"id":"259abf76039a82a12f80e1f4","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":85.0877192982456,"warning":null},{"id":"dde83904b2d2b88feda43a37","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":80.9941520467836,"warning":null},{"id":"4ca23b7218126b16b7bf2ac0","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":84.7953216374269,"warning":null},{"id":"382f6444647348a835f17c84","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":36.701030927835,"warning":null},{"id":"19a67bf1b22e11db9e9cd2cf","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":29.0721649484536,"warning":null},{"id":"7be2c174e04c3515a19514b2","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":44.3298969072165,"warning":null},{"id":"de86e8c22bb32644cd0c16bb","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":36.4948453608247,"warning":null},{"id":"08b420cc86ad300f73cb6ece","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":19.5876288659794,"warning":null},{"id":"b3520655c539ee36840e925e","modelId":"gpt-5.6-sol","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":38.144329896907195,"warning":null},{"id":"8da12db08f7c8df6f94bcb90","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":22.3809523809524,"warning":null},{"id":"6dcfaffdcd935d110c67e98e","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":62.121212121212096,"warning":null},{"id":"48dd8babcc5f09f92136ca87","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":60.606060606060595,"warning":null},{"id":"0aa1b482d07a39f7fe8363bb","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":65.90909090909089,"warning":null},{"id":"0296fff15d0b6f4671e5fd71","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":62.878787878787904,"warning":null},{"id":"07a4e6a430d6464d700d4bac","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":61.363636363636395,"warning":null},{"id":"96cc9128c8a64604039fac1d","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":87.2659176029963,"warning":null},{"id":"d4a631e8346c3b21752e1988","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":76.77902621722849,"warning":null},{"id":"5279247b2f384817941391c9","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":88.0149812734082,"warning":null},{"id":"17e0b62b4409266667ff8091","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":86.1423220973783,"warning":null},{"id":"3dd6efbb347ef62f98104dbb","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GPT-5.6 Sol (Non-reasoning)","configuration":"GPT-5.6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-non-reasoning","score":74.1573033707865,"warning":null},{"id":"f07dada8b8d9767f715e9127","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":89.5131086142322,"warning":null},{"id":"4ba1c52bab8baf294722e4c9","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-high","score":20.7070707070707,"warning":null},{"id":"ccb3770434f9f22b728e6c50","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-low","score":1.01010101010101,"warning":null},{"id":"453f566fa9b1e074f7c85a8c","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol","score":39.8989898989899,"warning":null},{"id":"2587843de6896ad5f6f3db1b","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-medium","score":14.6464646464646,"warning":null},{"id":"6d0e086f2b4f9dfc32336e52","modelId":"gpt-5.6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-5.6 Sol (Xhigh)","configuration":"GPT-5.6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-sol-xhigh","score":24.747474747474698,"warning":null},{"id":"44d14513877192a9ab52ca19","modelId":"gpt-5.6-sol","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.6 Sol (High)","configuration":"GPT-5.6 Sol (High); Source model ID: openai-gpt-5-6-sol-high; 120 tasks","score":85.42,"warning":null},{"id":"807ea1d13e6eff4620060049","modelId":"gpt-5.6-sol","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.6 Sol (Low)","configuration":"GPT-5.6 Sol (Low); Source model ID: openai-gpt-5-6-sol-low; 120 tasks","score":42.5,"warning":null},{"id":"2b83cb97a7ce702b83c5f1a1","modelId":"gpt-5.6-sol","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GPT-5.6 Sol (Max)","configuration":"GPT-5.6 Sol (Max); Source model ID: openai-gpt-5-6-sol-max; 120 tasks","score":92.5,"warning":null},{"id":"c14d5d98e2ac0c35e12373b0","modelId":"gpt-5.6-sol","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.6 Sol (Medium)","configuration":"GPT-5.6 Sol (Medium); Source model ID: openai-gpt-5-6-sol-medium; 120 tasks","score":67.08,"warning":null},{"id":"b9714ddbf13fb44d8cbb4989","modelId":"gpt-5.6-sol","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.6 Sol (XHigh)","configuration":"GPT-5.6 Sol (XHigh); Source model ID: openai-gpt-5-6-sol-xhigh; 120 tasks","score":90,"warning":null},{"id":"3289fff293acd220d866101d","modelId":"gpt-5.6-sol","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gpt-5-6-sol","configuration":"gpt-5-6-sol; Reported effort: high; 113 unique tasks; 451 attempts; 4 runs","score":69.4,"warning":null},{"id":"f7154b7752b50075526308bc","modelId":"gpt-5.6-sol","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"gpt-5-6-sol","configuration":"gpt-5-6-sol; Reported effort: low; 113 unique tasks; 452 attempts; 4 runs","score":45.4,"warning":null},{"id":"8ee05b294998d24ac869adf8","modelId":"gpt-5.6-sol","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"gpt-5-6-sol","configuration":"gpt-5-6-sol; Reported effort: max; 113 unique tasks; 450 attempts; 4 runs","score":72.7,"warning":null},{"id":"fd0b93e69454c6e0b78551ea","modelId":"gpt-5.6-sol","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"gpt-5-6-sol","configuration":"gpt-5-6-sol; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":61.1,"warning":null},{"id":"045b52fd58198f20f15b3ce9","modelId":"gpt-5.6-sol","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"gpt-5-6-sol","configuration":"gpt-5-6-sol; Reported effort: xhigh; 113 unique tasks; 451 attempts; 4 runs","score":70.7,"warning":null},{"id":"c0222b1f409ef8c270d05753","modelId":"gpt-5.6-terra","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":1202.17,"warning":null},{"id":"c9542d99cde7168563fda1c8","modelId":"gpt-5.6-terra","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":1014.15,"warning":null},{"id":"4eb3693e28b5c639480c123e","modelId":"gpt-5.6-terra","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":1334.22,"warning":null},{"id":"05bb2ffc8f323b804ff3a9d8","modelId":"gpt-5.6-terra","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":1039.24,"warning":null},{"id":"2faac4960c311646d5d8cdea","modelId":"gpt-5.6-terra","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":947.53,"warning":null},{"id":"86d83394e5c1dedc4e073325","modelId":"gpt-5.6-terra","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":1332.76,"warning":null},{"id":"96c3a3a2f035810e06f7c504","modelId":"gpt-5.6-terra","benchmarkId":"aa-apex-agents-aa","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":38.9380530973451,"warning":null},{"id":"da72f35326ec924994432ae2","modelId":"gpt-5.6-terra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":77.6666666666667,"warning":null},{"id":"976d7f8165eb9485a175b69d","modelId":"gpt-5.6-terra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":71.3333333333333,"warning":null},{"id":"10cbe5fa43e014bed382086d","modelId":"gpt-5.6-terra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":83,"warning":null},{"id":"5f53c2771eb757d3cf48a96a","modelId":"gpt-5.6-terra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":74,"warning":null},{"id":"558071cbc1fad71c429098b9","modelId":"gpt-5.6-terra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":58.6666666666667,"warning":null},{"id":"2c432d2c8b7a766c8e6fa9ee","modelId":"gpt-5.6-terra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":79,"warning":null},{"id":"81e98433d0c877616e6c3c0e","modelId":"gpt-5.6-terra","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":42.02038295182785,"warning":null},{"id":"bfe40ff055da82f7bc414895","modelId":"gpt-5.6-terra","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":29.106376138008166,"warning":null},{"id":"9c2bd279a0064b7c6f9a8dc4","modelId":"gpt-5.6-terra","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":59.64995802534495,"warning":null},{"id":"9ff5213c647e57fdbbba21de","modelId":"gpt-5.6-terra","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":33.536040384184965,"warning":null},{"id":"9b21e4ecbf840dbe6fe463a9","modelId":"gpt-5.6-terra","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":12.5407685673699,"warning":null},{"id":"cde8f1d0912a8d8d7b53a2fc","modelId":"gpt-5.6-terra","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":47.13631340597769,"warning":null},{"id":"11c39f6e675b8536e7610db3","modelId":"gpt-5.6-terra","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":22.8571428571429,"warning":null},{"id":"8b88cbb811b02401ffdc78d9","modelId":"gpt-5.6-terra","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":9.42857142857143,"warning":null},{"id":"482b224a42ef2c0ebab6349a","modelId":"gpt-5.6-terra","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":30,"warning":null},{"id":"025b9123f3de9ba5613f746d","modelId":"gpt-5.6-terra","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":17.4285714285714,"warning":null},{"id":"4c1370b3f65b8ad3d1b78deb","modelId":"gpt-5.6-terra","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":2,"warning":null},{"id":"3f5411918c81e8417d609437","modelId":"gpt-5.6-terra","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":27.142857142857103,"warning":null},{"id":"808e972a6513da410559e288","modelId":"gpt-5.6-terra","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":38.49597135183527,"warning":null},{"id":"432b733a92206e0ae0ec8d2c","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":20.8,"warning":null},{"id":"e2c632bb7e806838d4f81fb0","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":17.599999999999998,"warning":null},{"id":"e91edbaf5cd42bdff700000a","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":24,"warning":null},{"id":"5b0dec8e5666bee78f17d15a","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":17,"warning":null},{"id":"624d491d473a7a1bb1f45baf","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":11.200000000000001,"warning":null},{"id":"c21d9ad9948b19bd911c878b","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":24.6,"warning":null},{"id":"e061c950398d771035224919","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":1375.94,"warning":null},{"id":"8033faafc5f1e6a7c96955b3","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":1108.75,"warning":null},{"id":"84ede5e6045c0ddb02194b7e","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":1451.58,"warning":null},{"id":"abfa176ea4c9a7cb4f6ebb29","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":1265.44,"warning":null},{"id":"a305fd24000f9f954201a4f0","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":1098.01,"warning":null},{"id":"9b3b37fb91a24c4cff4cc9ad","modelId":"gpt-5.6-terra","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":1446.98,"warning":null},{"id":"bde949be3c6c4af41df9040b","modelId":"gpt-5.6-terra","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":89.5959595959596,"warning":null},{"id":"509a0b5bf39aeae551ccd5e7","modelId":"gpt-5.6-terra","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":84.3434343434343,"warning":null},{"id":"9a14552947e63008bb4144d5","modelId":"gpt-5.6-terra","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":92.5252525252525,"warning":null},{"id":"b52c98ccf19f855316aee8e6","modelId":"gpt-5.6-terra","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":87.1717171717172,"warning":null},{"id":"33c3d386b31ffb1ebcf6d165","modelId":"gpt-5.6-terra","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":74.64646464646461,"warning":null},{"id":"fa42c089a53fd555317dbe29","modelId":"gpt-5.6-terra","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":90.8080808080808,"warning":null},{"id":"cf5579b04a842106d8dd426a","modelId":"gpt-5.6-terra","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":38.5078776645042,"warning":null},{"id":"4ceb28cde6b1596087a57567","modelId":"gpt-5.6-terra","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":29.1936978683967,"warning":null},{"id":"7bd616b3ad913e37ff76098c","modelId":"gpt-5.6-terra","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":42.9101019462465,"warning":null},{"id":"cc0fd5140539d980b4e6aca0","modelId":"gpt-5.6-terra","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":33.271547729379094,"warning":null},{"id":"38cef7b62412ad7e99c933fd","modelId":"gpt-5.6-terra","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":11.3994439295644,"warning":null},{"id":"a6dda5e47ee22a10d0d528f3","modelId":"gpt-5.6-terra","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":41.8906394810009,"warning":null},{"id":"8a2a1a43e76c332bed156cfd","modelId":"gpt-5.6-terra","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":64.421768707483,"warning":null},{"id":"7e104039289d7fc14ef9f586","modelId":"gpt-5.6-terra","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":59.659863945578195,"warning":null},{"id":"d390f50c6bf94cf1e469c067","modelId":"gpt-5.6-terra","benchmarkId":"aa-ifbench","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":71.2244897959184,"warning":null},{"id":"9bf9d80a85c8034bca770ecd","modelId":"gpt-5.6-terra","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":62.176870748299294,"warning":null},{"id":"11ee9137ca06723a61832a4a","modelId":"gpt-5.6-terra","benchmarkId":"aa-ifbench","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":66.2585034013605,"warning":null},{"id":"551516755f69c30b82218d1b","modelId":"gpt-5.6-terra","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":51.035781544256096,"warning":null},{"id":"6d67fb89e5fa98246db38905","modelId":"gpt-5.6-terra","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":12.2222222222222,"warning":null},{"id":"ae160ce0eff28148df829ba2","modelId":"gpt-5.6-terra","benchmarkId":"aa-mlcr-aa","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":8.888888888888891,"warning":null},{"id":"844b7cf56ba64216d5afd86a","modelId":"gpt-5.6-terra","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":31.6666666666667,"warning":null},{"id":"c7acdae5a149d8091e2337d7","modelId":"gpt-5.6-terra","benchmarkId":"aa-mlcr-aa","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":12.2222222222222,"warning":null},{"id":"8e8e407c3e017b07dc1ef99b","modelId":"gpt-5.6-terra","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":21.6666666666667,"warning":null},{"id":"d1ceaef42c6c2282d75cc061","modelId":"gpt-5.6-terra","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":79.0751445086705,"warning":null},{"id":"7ae3fdf8f41e4de36323c7d0","modelId":"gpt-5.6-terra","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":76.1271676300578,"warning":null},{"id":"3c69d65f18eb996cf86b6710","modelId":"gpt-5.6-terra","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":80.6936416184971,"warning":null},{"id":"137871edba9c2527a9cc0a2d","modelId":"gpt-5.6-terra","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":76.7630057803468,"warning":null},{"id":"ea99a5d3d1870424b5f5826f","modelId":"gpt-5.6-terra","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":66.7052023121387,"warning":null},{"id":"ce8f5b1c4ce56d1658212785","modelId":"gpt-5.6-terra","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":79.47976878612721,"warning":null},{"id":"6ba5406295cfbfee0aeeb688","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":-3.46666666666667,"warning":null},{"id":"fb08cf5faee0f46c6fec58ca","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":-6.83333333333333,"warning":null},{"id":"5378a72f622a354a37276342","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":0.05,"warning":null},{"id":"6ad0c4bdf9b9ec84c4ee0711","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":-5.13333333333333,"warning":null},{"id":"acb6eaf0788afbef68320415","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":-23.2166666666667,"warning":null},{"id":"4525fa025f453c3f7e18ba1f","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":-2.98333333333333,"warning":null},{"id":"80397099e3664b43170a185c","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":45.483333333333334,"warning":null},{"id":"b6a77b521be14c9f243e9567","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":43.733333333333334,"warning":null},{"id":"42e8b0b41d2c2a6905c39a44","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":46.800000000000004,"warning":null},{"id":"04540512a2d8e2ea11478d89","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":44.6,"warning":null},{"id":"0cf0c4cefc218ea3fcfed10b","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":36.81666666666666,"warning":null},{"id":"b62ff97ce03382d2ffb23d24","modelId":"gpt-5.6-terra","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":45.516666666666666,"warning":null},{"id":"8440f260de487af3659ac24e","modelId":"gpt-5.6-terra","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":52.4305555555556,"warning":null},{"id":"cad30981cda759eddf24482f","modelId":"gpt-5.6-terra","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":49.8842592592593,"warning":null},{"id":"722175536c6573787a0a82b7","modelId":"gpt-5.6-terra","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":54.976851851851805,"warning":null},{"id":"24599291513403a2120872ee","modelId":"gpt-5.6-terra","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":50.462962962963,"warning":null},{"id":"200cba6f59d9e1ed30007fc7","modelId":"gpt-5.6-terra","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":45.1388888888889,"warning":null},{"id":"84bbd4f88c501a6b5c239ee2","modelId":"gpt-5.6-terra","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":52.3148148148148,"warning":null},{"id":"4dd442f313a479d5e90cd322","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":78.3625730994152,"warning":null},{"id":"825e79632273448f16218dc8","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":60.52631578947369,"warning":null},{"id":"e5e477a801e611893a224b38","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau2-bench","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":86.2573099415205,"warning":null},{"id":"ed610bdab58d272989a6b7f9","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":72.8070175438596,"warning":null},{"id":"c8b0c50f6cebd6baf77fe47a","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau2-bench","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":80.4093567251462,"warning":null},{"id":"f576e5c881272a9151254be0","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":28.659793814432998,"warning":null},{"id":"9c90acc79272ab988a0c26dd","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":18.7628865979381,"warning":null},{"id":"043d1293688d8b31aa878216","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":40.2061855670103,"warning":null},{"id":"6c8ab0b52a7c4aec1c308cd5","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":25.5670103092783,"warning":null},{"id":"cc2e0c6125fc70488c1fc96a","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":15.6701030927835,"warning":null},{"id":"ce1da51dd767cfb113a95de3","modelId":"gpt-5.6-terra","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":29.6907216494845,"warning":null},{"id":"aa45a0adb91ce3f068658a53","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":13.8095238095238,"warning":null},{"id":"e87fe7de90b1bfe016e14ec9","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":57.5757575757576,"warning":null},{"id":"c1659dd6967ce3e4b921efe8","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":43.9393939393939,"warning":null},{"id":"d89971379e97e976f16034b5","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-hard","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":57.5757575757576,"warning":null},{"id":"e878d0cba9034a989af2ce79","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-hard","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":62.878787878787904,"warning":null},{"id":"c3c74bdc2b7601fdf0148cd6","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":75.6554307116105,"warning":null},{"id":"1993793c853b925a767d702e","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":62.546816479400704,"warning":null},{"id":"28feac946c1190fb690e46f5","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":88.0149812734082,"warning":null},{"id":"b471f88e4106eb6883be1958","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":72.2846441947566,"warning":null},{"id":"3ec836beaabc8142f018c103","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":56.1797752808989,"warning":null},{"id":"0a83c8ef0e3c0b400210b1ab","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":80.1498127340824,"warning":null},{"id":"ff6c519e7533c9a3a9448d79","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-high","score":1.51515151515152,"warning":null},{"id":"06663dcafe05cdc14f74403f","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-low","score":1.51515151515152,"warning":null},{"id":"f0d9727936a52201f01fc6e9","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra","score":35.3535353535354,"warning":null},{"id":"56fc2c38a6ac97ecdb0c78ba","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-medium","score":1.01010101010101,"warning":null},{"id":"daec7972c49960bddeda4c94","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"GPT-5.6 Terra (Non-reasoning)","configuration":"GPT-5.6 Terra (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-non-reasoning","score":0.505050505050505,"warning":null},{"id":"056b08fb9ab01edd59f46622","modelId":"gpt-5.6-terra","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-5.6 Terra (Xhigh)","configuration":"GPT-5.6 Terra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-6-terra-xhigh","score":10.1010101010101,"warning":null},{"id":"a3ab01caf6d0687478a0ce49","modelId":"gpt-5.6-terra","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-5.6 Terra (High)","configuration":"GPT-5.6 Terra (High); Source model ID: openai-gpt-5-6-terra-high; 120 tasks","score":67.08,"warning":null},{"id":"b818dce06c0fb558e7d7167a","modelId":"gpt-5.6-terra","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-5.6 Terra (Low)","configuration":"GPT-5.6 Terra (Low); Source model ID: openai-gpt-5-6-terra-low; 120 tasks","score":18.75,"warning":null},{"id":"853bc246193f33df5eea54f8","modelId":"gpt-5.6-terra","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GPT-5.6 Terra (Max)","configuration":"GPT-5.6 Terra (Max); Source model ID: openai-gpt-5-6-terra-max; 120 tasks","score":83.89999999999999,"warning":null},{"id":"0a8c913b486b6fffe98e5acb","modelId":"gpt-5.6-terra","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-5.6 Terra (Medium)","configuration":"GPT-5.6 Terra (Medium); Source model ID: openai-gpt-5-6-terra-medium; 120 tasks","score":37.5,"warning":null},{"id":"e4620c3a04edefbb34c222e8","modelId":"gpt-5.6-terra","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-5.6 Terra (XHigh)","configuration":"GPT-5.6 Terra (XHigh); Source model ID: openai-gpt-5-6-terra-xhigh; 120 tasks","score":74.17,"warning":null},{"id":"d998a419297ba93aa26c6ef6","modelId":"gpt-5.6-terra","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gpt-5-6-terra","configuration":"gpt-5-6-terra; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":53.8,"warning":null},{"id":"b2a027bcbbe350e83bb9379f","modelId":"gpt-5.6-terra","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"gpt-5-6-terra","configuration":"gpt-5-6-terra; Reported effort: low; 113 unique tasks; 449 attempts; 4 runs","score":24.1,"warning":null},{"id":"8af3e2d0812a9e3c0e993ce1","modelId":"gpt-5.6-terra","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"gpt-5-6-terra","configuration":"gpt-5-6-terra; Reported effort: max; 113 unique tasks; 451 attempts; 4 runs","score":69.6,"warning":null},{"id":"5b1b840c7e90fab89305a587","modelId":"gpt-5.6-terra","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"gpt-5-6-terra","configuration":"gpt-5-6-terra; Reported effort: medium; 113 unique tasks; 450 attempts; 4 runs","score":35.1,"warning":null},{"id":"2ca3b5887af025e030617db6","modelId":"gpt-5.6-terra","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"gpt-5-6-terra","configuration":"gpt-5-6-terra; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":60.2,"warning":null},{"id":"e0e6da3e67b23e17eb3133aa","modelId":"gpt-5.6-terra","benchmarkId":"terminal-codex","level":"max","sourceLabel":"GPT-5.6 Terra","configuration":"GPT-5.6 Terra; Reported effort: max; 445 trials; recorded 2026-06-26","score":78.43,"warning":null},{"id":"a15775f54335adef61419215","modelId":"gpt-6-astra","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":51.24999999999999,"warning":null},{"id":"3306472236bd0265432d14a9","modelId":"gpt-6-astra","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":1505.91,"warning":null},{"id":"5de5714c117be311ab3e0b6c","modelId":"gpt-6-astra","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":1257.01,"warning":null},{"id":"85b8ecd429880d200e00a398","modelId":"gpt-6-astra","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":1568.99,"warning":null},{"id":"0e8636e6c5f0104452cf8d91","modelId":"gpt-6-astra","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":1460.92,"warning":null},{"id":"67ca6196001a106d64ef5e74","modelId":"gpt-6-astra","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":1546.7,"warning":null},{"id":"39409fae57302ca9faeb927a","modelId":"gpt-6-astra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":80,"warning":null},{"id":"5df46d86b02fd0842c7cf4c9","modelId":"gpt-6-astra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":80,"warning":null},{"id":"dd5479697cf074b77c8a02ce","modelId":"gpt-6-astra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":80.6666666666667,"warning":null},{"id":"4670870f099d246b275c7068","modelId":"gpt-6-astra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":79.6666666666667,"warning":null},{"id":"012b7edd52e1e9f92f97051c","modelId":"gpt-6-astra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":80,"warning":null},{"id":"144782ec8089c6effc04478e","modelId":"gpt-6-astra","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":66.6122174975011,"warning":null},{"id":"e6be8f14e6825f1bc99851ff","modelId":"gpt-6-astra","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":59.09672610445402,"warning":null},{"id":"d80dccedbe3f2ce8236cc39d","modelId":"gpt-6-astra","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":68.49174806688876,"warning":null},{"id":"3e1e53285f08696c89010311","modelId":"gpt-6-astra","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":64.59498850561177,"warning":null},{"id":"8a2ed96d0591ee82c77b05c0","modelId":"gpt-6-astra","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":67.17778020215393,"warning":null},{"id":"9c1c24ee17bdace54d039a55","modelId":"gpt-6-astra","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":28.857142857142897,"warning":null},{"id":"3868a1cc6fd696e3bf4fe80f","modelId":"gpt-6-astra","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":26.285714285714302,"warning":null},{"id":"770de22cf6837ba9567ed41e","modelId":"gpt-6-astra","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":31.7142857142857,"warning":null},{"id":"e3543abe0620e3ac9004d0ed","modelId":"gpt-6-astra","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":29.1428571428571,"warning":null},{"id":"94820e4d0d454921c6aba0d3","modelId":"gpt-6-astra","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":31.4285714285714,"warning":null},{"id":"93a8756804f6409d796a8241","modelId":"gpt-6-astra","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":31,"warning":null},{"id":"ca2b834660f47252bea4b1d1","modelId":"gpt-6-astra","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":30.4,"warning":null},{"id":"8738686d0dc2a17d182e3113","modelId":"gpt-6-astra","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":31,"warning":null},{"id":"c065230d455e7d9c70fca9f2","modelId":"gpt-6-astra","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":30.4,"warning":null},{"id":"fddde6e8de94311bc7d22534","modelId":"gpt-6-astra","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":32.2,"warning":null},{"id":"d64f2c40cdd744f12f18f7d8","modelId":"gpt-6-astra","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":1522.92,"warning":null},{"id":"f96791a890a01181c2f0cbae","modelId":"gpt-6-astra","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":1389.25,"warning":null},{"id":"5ccdb77a6f9a3a8d02d57dcc","modelId":"gpt-6-astra","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":1572.84,"warning":null},{"id":"b066683df9acd8d74175e6b3","modelId":"gpt-6-astra","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":1486.4,"warning":null},{"id":"341a0eb0df533ba3257468c1","modelId":"gpt-6-astra","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":1555.91,"warning":null},{"id":"f6e00130e48af2515bb29d78","modelId":"gpt-6-astra","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":94.94949494949499,"warning":null},{"id":"2c11b614575baa3edd4ea255","modelId":"gpt-6-astra","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":93.13131313131309,"warning":null},{"id":"7f05b02e6f808515a93f167f","modelId":"gpt-6-astra","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":96.0606060606061,"warning":null},{"id":"fb3a15a9cc5d60fb95bba7ac","modelId":"gpt-6-astra","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":93.9393939393939,"warning":null},{"id":"ffa596ff99ba203489b0b172","modelId":"gpt-6-astra","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":96.2626262626263,"warning":null},{"id":"16d220d928175ee74b8d0492","modelId":"gpt-6-astra","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":8.611111111111109,"warning":null},{"id":"5e90689cbb2c5df59c046e99","modelId":"gpt-6-astra","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":53.058387395736794,"warning":null},{"id":"ac62fed233935e7dfe90b5dd","modelId":"gpt-6-astra","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":49.2122335495829,"warning":null},{"id":"faafe666826d6591d79ddb5e","modelId":"gpt-6-astra","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":54.6802594995366,"warning":null},{"id":"32d0295251d0c38765dc01e3","modelId":"gpt-6-astra","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":52.73401297497679,"warning":null},{"id":"13b2b6216d3dcfdde1b27627","modelId":"gpt-6-astra","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":54.587581093605195,"warning":null},{"id":"f861154ba82953aeb3b450b6","modelId":"gpt-6-astra","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":48.5875706214689,"warning":null},{"id":"9f4ba53c8465e3b7e7fd2c5b","modelId":"gpt-6-astra","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":35,"warning":null},{"id":"77f5198b6ddc73d1f79de3b8","modelId":"gpt-6-astra","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":86.4161849710983,"warning":null},{"id":"549bfd068100b5edcacf8b19","modelId":"gpt-6-astra","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":84.6242774566474,"warning":null},{"id":"25c47bbff1f4c6c57d50efb0","modelId":"gpt-6-astra","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":86.878612716763,"warning":null},{"id":"5ce7afa0c29c8f933b6b5672","modelId":"gpt-6-astra","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":85.08670520231209,"warning":null},{"id":"6ce5ac431457726574b076dd","modelId":"gpt-6-astra","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":86.242774566474,"warning":null},{"id":"c2ed4078e8723069cad94348","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":43.7333333333333,"warning":null},{"id":"0bd39379aa60980372b1a483","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":40.55,"warning":null},{"id":"936b4694f2d3e110c96c7d1d","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":43.4,"warning":null},{"id":"29809c33eceeba5622b144bb","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":42.2166666666667,"warning":null},{"id":"104d966ee15689cf157d5e7f","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":43.4166666666667,"warning":null},{"id":"551992e2d945d529e608999b","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":61.133333333333326,"warning":null},{"id":"7e68b2d7b1355f504488f4b5","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":59.53333333333334,"warning":null},{"id":"264cafe65edbf0954b13f92e","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":62.6,"warning":null},{"id":"e7ea8a458aaf2c4ea3f92273","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":60.56666666666667,"warning":null},{"id":"a967b8315c90c4f938e482a4","modelId":"gpt-6-astra","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":61.85000000000001,"warning":null},{"id":"8b41b672e41052e0cd3069a0","modelId":"gpt-6-astra","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":55.4398148148148,"warning":null},{"id":"976f352750456195751d0a56","modelId":"gpt-6-astra","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":54.050925925925895,"warning":null},{"id":"62be10449eb661358f9cb609","modelId":"gpt-6-astra","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":56.4814814814815,"warning":null},{"id":"d8d8078f1b862e001d46f606","modelId":"gpt-6-astra","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":54.1666666666667,"warning":null},{"id":"d9e2313727f7a8e1ad52e120","modelId":"gpt-6-astra","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":55.671296296296305,"warning":null},{"id":"951efb176288d047c3edb93c","modelId":"gpt-6-astra","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":40,"warning":null},{"id":"e3a3f60aa8ea54c9a186c1ed","modelId":"gpt-6-astra","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":31.958762886597903,"warning":null},{"id":"9716086f04cc09344b507629","modelId":"gpt-6-astra","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":41.4432989690722,"warning":null},{"id":"2fc4a5319dcd44b6132e21c3","modelId":"gpt-6-astra","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":35.4639175257732,"warning":null},{"id":"e0d529ed3d02b7e8d02a83e1","modelId":"gpt-6-astra","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":43.0927835051546,"warning":null},{"id":"af914cf2b8698f09e21fa009","modelId":"gpt-6-astra","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":63.3333333333333,"warning":null},{"id":"b6c3a059c111937b25cbb981","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":89.8876404494382,"warning":null},{"id":"6c8093f44c06ee5133fb47b0","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":88.0149812734082,"warning":null},{"id":"5b213107007dc966fa21bdbf","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":88.38951310861421,"warning":null},{"id":"be6fb47b80b00f737c4a8e45","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":89.5131086142322,"warning":null},{"id":"ea7e719bd1a0713c5581a76b","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":89.1385767790262,"warning":null},{"id":"b4a051abe87f5047b2a9d4c0","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Model source: https://artificialanalysis.ai/models/gpt-6-astra-high","score":54.040404040404,"warning":null},{"id":"e3ba9b92c5c6f13ff9f8916b","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Model source: https://artificialanalysis.ai/models/gpt-6-astra-low","score":41.9191919191919,"warning":null},{"id":"905559652cdb38bdf511bcc4","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Model source: https://artificialanalysis.ai/models/gpt-6-astra","score":59.09090909090911,"warning":null},{"id":"b70756c954a318a876c4fc81","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-astra-medium","score":49.494949494949495,"warning":null},{"id":"92f0ff4a5a6e52e0c239bcd5","modelId":"gpt-6-astra","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-6 Astra (Xhigh)","configuration":"GPT-6 Astra (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-astra-xhigh","score":59.5959595959596,"warning":null},{"id":"0c5432a88a6429cd53ba80de","modelId":"gpt-6-astra","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-6 Astra (High)","configuration":"GPT-6 Astra (High); Source model ID: openai-gpt-6-astra-high; 120 tasks","score":92.08333333333333,"warning":null},{"id":"aff9b666a1cca3b3ee07c2ea","modelId":"gpt-6-astra","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-6 Astra (Low)","configuration":"GPT-6 Astra (Low); Source model ID: openai-gpt-6-astra-low; 120 tasks","score":85.41666666666666,"warning":null},{"id":"34add3a000b32e26d7ba251e","modelId":"gpt-6-astra","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"GPT-6 Astra (Max); Source model ID: openai-gpt-6-astra-max; 120 tasks","score":95,"warning":null},{"id":"6f7f64f65d746e4beebd2d85","modelId":"gpt-6-astra","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-6 Astra (Medium)","configuration":"GPT-6 Astra (Medium); Source model ID: openai-gpt-6-astra-medium; 120 tasks","score":92.08333333333333,"warning":null},{"id":"b5faeccd54fcda2954357a39","modelId":"gpt-6-astra","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"GPT-6 Astra (None)","configuration":"GPT-6 Astra (None); Source model ID: openai-gpt-6-astra-none; 120 tasks","score":59.583333333333336,"warning":"Source-reported label. OpenAI documents Low through Max for Astra; this does not establish a supported non-reasoning API setting."},{"id":"8a7b9cc900bc1e48522b2d77","modelId":"gpt-6-astra","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-6 Astra (XHigh)","configuration":"GPT-6 Astra (XHigh); Source model ID: openai-gpt-6-astra-xhigh; 120 tasks","score":93.33333333333333,"warning":null},{"id":"192a0a52134befbac5912c90","modelId":"gpt-6-astra","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"gpt-6-astra","configuration":"gpt-6-astra; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":73.2,"warning":null},{"id":"eb733075f994115168315353","modelId":"gpt-6-astra","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"gpt-6-astra","configuration":"gpt-6-astra; Reported effort: low; 113 unique tasks; 452 attempts; 4 runs","score":67,"warning":null},{"id":"fa893735776a86c9eaeeab33","modelId":"gpt-6-astra","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"gpt-6-astra","configuration":"gpt-6-astra; Reported effort: max; 113 unique tasks; 452 attempts; 4 runs","score":73.2,"warning":null},{"id":"eba425fa2243963df9b630e9","modelId":"gpt-6-astra","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"gpt-6-astra","configuration":"gpt-6-astra; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":72.8,"warning":null},{"id":"49b1b70f581245c54e071b3a","modelId":"gpt-6-astra","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"gpt-6-astra","configuration":"gpt-6-astra; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":74.1,"warning":null},{"id":"434544966d1431c1926d52f5","modelId":"gpt-6-astra","benchmarkId":"terminal-codex","level":"high","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; Reported effort: high; 445 trials; recorded 2026-09-03","score":87.42,"warning":null},{"id":"91bc2d06c970323599005b34","modelId":"gpt-6-astra","benchmarkId":"terminal-codex","level":"low","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; Reported effort: low; 445 trials; recorded 2026-09-03","score":86.74,"warning":null},{"id":"bf597f8d4309561326a1ef1f","modelId":"gpt-6-astra","benchmarkId":"terminal-codex","level":"max","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; Reported effort: max; 445 trials; recorded 2026-09-03","score":86.74,"warning":null},{"id":"bb6111da7a05efb610da52fd","modelId":"gpt-6-astra","benchmarkId":"terminal-codex","level":"medium","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; Reported effort: medium; 445 trials; recorded 2026-09-03","score":86.97,"warning":null},{"id":"cefb2b8c7589ea007cb6a16b","modelId":"gpt-6-astra","benchmarkId":"terminal-codex","level":"xhigh","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; Reported effort: xhigh; 445 trials; recorded 2026-09-03","score":85.84,"warning":null},{"id":"85cdbce7a287aedba059fbef","modelId":"gpt-6-luna","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":43.75,"warning":null},{"id":"6df099fcab3bc59df1a2848e","modelId":"gpt-6-luna","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":1186.42,"warning":null},{"id":"6e7bd695aa23bf06ff18d157","modelId":"gpt-6-luna","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":741.34,"warning":null},{"id":"ecbc90fb7fe042e052ba1753","modelId":"gpt-6-luna","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":1334.83,"warning":null},{"id":"07b2e07748cef846e9f0817f","modelId":"gpt-6-luna","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":1098.75,"warning":null},{"id":"46d79399ee441894914a8f1a","modelId":"gpt-6-luna","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":903.18,"warning":null},{"id":"7a82437d9f707352e49c5dda","modelId":"gpt-6-luna","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":1236.97,"warning":null},{"id":"4add5bece4540a198cb2a579","modelId":"gpt-6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":79.3333333333333,"warning":null},{"id":"57dcd7df3e4da472eeb09e87","modelId":"gpt-6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":74,"warning":null},{"id":"c8bf4ee9e05436bbf3c463c3","modelId":"gpt-6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":83.3333333333333,"warning":null},{"id":"7d8566afcdef04eb1769a040","modelId":"gpt-6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":78.3333333333333,"warning":null},{"id":"7616e33171a1a060a1c26c5b","modelId":"gpt-6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":39.6666666666667,"warning":null},{"id":"eccd70ba4b9cd043c9a657a0","modelId":"gpt-6-luna","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":80,"warning":null},{"id":"6c9b637c2631b2c8a3972ddc","modelId":"gpt-6-luna","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":47.82449139916615,"warning":null},{"id":"818ef24865c84783fc39375b","modelId":"gpt-6-luna","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":12.049187272690505,"warning":null},{"id":"fcd6b9de9db634e2c4a9844b","modelId":"gpt-6-luna","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":53.16814034483175,"warning":null},{"id":"8d93d178aa6105b5644b1587","modelId":"gpt-6-luna","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":40.46136879135791,"warning":null},{"id":"8c4dcbe13c8ef2ffadfe1e89","modelId":"gpt-6-luna","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":9.141771078680867,"warning":null},{"id":"90b73130e61aba8283ac4c32","modelId":"gpt-6-luna","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":47.800686767566575,"warning":null},{"id":"52976d5bd4fd450f6e70bb8f","modelId":"gpt-6-luna","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":15.4285714285714,"warning":null},{"id":"414447fbe6d878f1a0a8f204","modelId":"gpt-6-luna","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":2.57142857142857,"warning":null},{"id":"4ee05767675070ace8ff4a10","modelId":"gpt-6-luna","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":19.428571428571402,"warning":null},{"id":"39c58ccfce992c4098e15db3","modelId":"gpt-6-luna","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":10.5714285714286,"warning":null},{"id":"be3e5d200a8ef22d2e7d908b","modelId":"gpt-6-luna","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":1.1428571428571401,"warning":null},{"id":"842bda61fdf4a03695e85fc0","modelId":"gpt-6-luna","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":17.4285714285714,"warning":null},{"id":"35037581a38fca7108a8ff15","modelId":"gpt-6-luna","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":18.2,"warning":null},{"id":"f166779c35e15578e2d20011","modelId":"gpt-6-luna","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":9,"warning":null},{"id":"e76d489e380fde49262e3e7a","modelId":"gpt-6-luna","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":22.8,"warning":null},{"id":"ed69ba3033796ee33cfbba7f","modelId":"gpt-6-luna","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":13.8,"warning":null},{"id":"7bdc20c361b224dd8d5ac840","modelId":"gpt-6-luna","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":5.800000000000001,"warning":null},{"id":"d7e2e92d7df76f060438837a","modelId":"gpt-6-luna","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":18.6,"warning":null},{"id":"f9610d53f3227c80d1acc510","modelId":"gpt-6-luna","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":1339.93,"warning":null},{"id":"1fc182b7287f42dad1a05465","modelId":"gpt-6-luna","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":1033.12,"warning":null},{"id":"34f4c0efa1fd06fc30de9556","modelId":"gpt-6-luna","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":1431.86,"warning":null},{"id":"e424a52bc46cd95f4a4008b3","modelId":"gpt-6-luna","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":1262.27,"warning":null},{"id":"87b55df4b09954ea311ec452","modelId":"gpt-6-luna","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":1045.45,"warning":null},{"id":"2324d8cba91fb427eed089e9","modelId":"gpt-6-luna","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":1363.87,"warning":null},{"id":"4c0b35ba0b0dea0d2ec65470","modelId":"gpt-6-luna","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":3.3333333333333335,"warning":null},{"id":"dae9bf3e3d4b5670d3893f15","modelId":"gpt-6-luna","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":32.9471733086191,"warning":null},{"id":"16510b56f5d5fa75b2b71352","modelId":"gpt-6-luna","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":20.2502316960148,"warning":null},{"id":"86869e43d3ba2fd62a2efaa9","modelId":"gpt-6-luna","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":38.5078776645042,"warning":null},{"id":"f701423cfe5ba7df2b56aa37","modelId":"gpt-6-luna","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":28.2669138090825,"warning":null},{"id":"f20be376e35e699e68422716","modelId":"gpt-6-luna","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":8.57275254865616,"warning":null},{"id":"7fe3b1afa7a09cc26986c6c3","modelId":"gpt-6-luna","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":34.3373493975904,"warning":null},{"id":"d359d2dbc04c1e7574e2cf32","modelId":"gpt-6-luna","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":16.1111111111111,"warning":null},{"id":"92aeb221c4bb583e17bbc377","modelId":"gpt-6-luna","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":77.514450867052,"warning":null},{"id":"69e7261a2f652eca41752cbc","modelId":"gpt-6-luna","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":75.9537572254335,"warning":null},{"id":"d6daa4658a4816df33eb8a68","modelId":"gpt-6-luna","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":79.6531791907514,"warning":null},{"id":"22dc8988bd2fdbf4d396764a","modelId":"gpt-6-luna","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":77.8612716763006,"warning":null},{"id":"7c5470194e5dd81e14991a33","modelId":"gpt-6-luna","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":62.080924855491304,"warning":null},{"id":"7970dfff3e9c437c790e082d","modelId":"gpt-6-luna","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":79.7109826589595,"warning":null},{"id":"20e3ef7fbf64984566407e4e","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":-5.5,"warning":null},{"id":"9ea2580ef691d710d9cc2dd5","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":-9.16666666666667,"warning":null},{"id":"9f2ebaa525d1bf0a2dcd8fc9","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":0.65,"warning":null},{"id":"2f88e138bda31c9cfa8ab817","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":-5.05,"warning":null},{"id":"63298170bbb423389ae897e0","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":-21.9,"warning":null},{"id":"f663cb94aed0c489b8cd081e","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":-1.83333333333333,"warning":null},{"id":"2e7e07707092204b5dba5f48","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":42.8,"warning":null},{"id":"235e213364000a65a590a20f","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":40.766666666666666,"warning":null},{"id":"97809e7f09d0767b3d0f6154","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":43.78333333333334,"warning":null},{"id":"b615d1750e6c87a9bf6f1db3","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":43.13333333333333,"warning":null},{"id":"038cd005e04bbd9da5b2a39b","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":31.85,"warning":null},{"id":"fd535b33980376b6b1b2cd33","modelId":"gpt-6-luna","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":44.166666666666664,"warning":null},{"id":"0fa34aea17e35cf7ae3beeb9","modelId":"gpt-6-luna","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":50.3472222222222,"warning":null},{"id":"2d547df4fe61b3c4b08abfc2","modelId":"gpt-6-luna","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":46.875,"warning":null},{"id":"f4c807a7e3a01a657b5dca3d","modelId":"gpt-6-luna","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":54.6296296296296,"warning":null},{"id":"12f4aaf84c8045b8e65eeaa1","modelId":"gpt-6-luna","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":50.925925925925895,"warning":null},{"id":"7f3ec19d60fc54e1792d8e6a","modelId":"gpt-6-luna","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":43.0555555555556,"warning":null},{"id":"d5d2b5f5d6082482db5e6acf","modelId":"gpt-6-luna","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":51.73611111111111,"warning":null},{"id":"bad4886aa53b9782b1daf1f5","modelId":"gpt-6-luna","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":8.57142857142857,"warning":null},{"id":"aae07df2a2545ea42cd10c66","modelId":"gpt-6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Model source: https://artificialanalysis.ai/models/gpt-6-luna-high","score":4.545454545454549,"warning":null},{"id":"59995116df23e68ce652da78","modelId":"gpt-6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Model source: https://artificialanalysis.ai/models/gpt-6-luna-low","score":0,"warning":null},{"id":"8e39763d3c72f894ba76fae2","modelId":"gpt-6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Model source: https://artificialanalysis.ai/models/gpt-6-luna","score":12.6262626262626,"warning":null},{"id":"e1accbba79c31b0f64fcc569","modelId":"gpt-6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-luna-medium","score":2.52525252525253,"warning":null},{"id":"ae9cd41cfe460d23e2ba5708","modelId":"gpt-6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"GPT-6 Luna (Non-reasoning)","configuration":"GPT-6 Luna (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-luna-non-reasoning","score":1.51515151515152,"warning":null},{"id":"4ce516f57cf7ad82a9643b56","modelId":"gpt-6-luna","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-6 Luna (Xhigh)","configuration":"GPT-6 Luna (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-luna-xhigh","score":8.08080808080808,"warning":null},{"id":"a8d7dca0504bb948aeb158e3","modelId":"gpt-6-luna","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-6 Luna (High)","configuration":"GPT-6 Luna (High); Source model ID: openai-gpt-6-luna-high; 120 tasks","score":31.38888888888889,"warning":null},{"id":"79c498606e7843c499358ae4","modelId":"gpt-6-luna","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-6 Luna (Low)","configuration":"GPT-6 Luna (Low); Source model ID: openai-gpt-6-luna-low; 120 tasks","score":4.583333333333333,"warning":null},{"id":"b4f0c67df2c799bb9967528e","modelId":"gpt-6-luna","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"GPT-6 Luna (Max); Source model ID: openai-gpt-6-luna-max; 120 tasks","score":59.30555555555556,"warning":null},{"id":"ddaf72bfd88bc1448399808c","modelId":"gpt-6-luna","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-6 Luna (Medium)","configuration":"GPT-6 Luna (Medium); Source model ID: openai-gpt-6-luna-medium; 120 tasks","score":18.055555555555554,"warning":null},{"id":"5849e34f56744250ee2ebc46","modelId":"gpt-6-luna","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"GPT-6 Luna (None)","configuration":"GPT-6 Luna (None); Source model ID: openai-gpt-6-luna-none; 120 tasks","score":0,"warning":null},{"id":"706ee66952f7ced5735443bf","modelId":"gpt-6-luna","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-6 Luna (XHigh)","configuration":"GPT-6 Luna (XHigh); Source model ID: openai-gpt-6-luna-xhigh; 120 tasks","score":41.94444444444444,"warning":null},{"id":"98b42a9c860fb80ff6f4d382","modelId":"gpt-6-sol","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":1266.41,"warning":null},{"id":"29d28140ad3c106c00f471a4","modelId":"gpt-6-sol","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":886.44,"warning":null},{"id":"3874a1f1a69d8772d7aaf5ab","modelId":"gpt-6-sol","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":1480.92,"warning":null},{"id":"350e87d54192d48c964c1205","modelId":"gpt-6-sol","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":1149.3,"warning":null},{"id":"450b0b738e836e36c8fd5b56","modelId":"gpt-6-sol","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":1103.23,"warning":null},{"id":"27c8c75ae34f350da099dfcc","modelId":"gpt-6-sol","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":1357.36,"warning":null},{"id":"d8dbdc56f80af68220f9bc80","modelId":"gpt-6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":83.6666666666667,"warning":null},{"id":"c8cf364304f7f3f1e7b6c215","modelId":"gpt-6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":79.3333333333333,"warning":null},{"id":"ec95ea6172b70ad59049ce05","modelId":"gpt-6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":83.6666666666667,"warning":null},{"id":"b9beaeb2d5579bd5aaea6061","modelId":"gpt-6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":82.3333333333333,"warning":null},{"id":"4ec8e704c2c823cb11b5388b","modelId":"gpt-6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":64,"warning":null},{"id":"75987472ff41b7dbea74c08f","modelId":"gpt-6-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":81.3333333333333,"warning":null},{"id":"2e486240163922f265ea86a0","modelId":"gpt-6-sol","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":60.12584091127188,"warning":null},{"id":"048baebd12c938c93c4760ad","modelId":"gpt-6-sol","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":53.88999278467668,"warning":null},{"id":"5475871f97ffc6a337855afa","modelId":"gpt-6-sol","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":61.63369295959864,"warning":null},{"id":"ad1ac307eabf4d8b593bb402","modelId":"gpt-6-sol","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":57.97223510768895,"warning":null},{"id":"86fb0f33b786e6d8c299ac63","modelId":"gpt-6-sol","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":34.180173483579395,"warning":null},{"id":"7347911755ca938562a34fd6","modelId":"gpt-6-sol","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":61.694235115864096,"warning":null},{"id":"f067747277aeded9564f53df","modelId":"gpt-6-sol","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":25.428571428571402,"warning":null},{"id":"ef6a217b2f2b992ca1783bce","modelId":"gpt-6-sol","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":16.2857142857143,"warning":null},{"id":"9b6a88e1da04914125bd0771","modelId":"gpt-6-sol","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":30.8571428571429,"warning":null},{"id":"c0080ebf64853b9d38d88e9b","modelId":"gpt-6-sol","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":24.571428571428598,"warning":null},{"id":"acb47cda766f9adcbd502018","modelId":"gpt-6-sol","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":4,"warning":null},{"id":"30e043165f9edcd99a02e13b","modelId":"gpt-6-sol","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":28.000000000000004,"warning":null},{"id":"38892c085d6eecf2cebd5c8f","modelId":"gpt-6-sol","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":24.4,"warning":null},{"id":"0909e2cf145d2043b08f7372","modelId":"gpt-6-sol","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":24.2,"warning":null},{"id":"a3eae8d4a26969cdbba052a4","modelId":"gpt-6-sol","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":25.2,"warning":null},{"id":"2095ada63a849fc5de43fa43","modelId":"gpt-6-sol","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":23.799999999999997,"warning":null},{"id":"6b8f37d7e3d5b51b767810bb","modelId":"gpt-6-sol","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":17,"warning":null},{"id":"590664e4ce67f3e910a09137","modelId":"gpt-6-sol","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":24.6,"warning":null},{"id":"656559fc83ff67aaff743eae","modelId":"gpt-6-sol","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":1394.08,"warning":null},{"id":"dc64d11df946c9266008a9ac","modelId":"gpt-6-sol","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":1200,"warning":null},{"id":"6ab8303578aff13b2bdd9f8d","modelId":"gpt-6-sol","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":1508.23,"warning":null},{"id":"a530be32d03db88edd19c8f0","modelId":"gpt-6-sol","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":1345.82,"warning":null},{"id":"62a985ae5e9767a7ee019e35","modelId":"gpt-6-sol","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":1248.98,"warning":null},{"id":"01397847b95699a967a3d819","modelId":"gpt-6-sol","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":1455.23,"warning":null},{"id":"cfa6111b396204ce86edf1e3","modelId":"gpt-6-sol","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":3.6111111111111107,"warning":null},{"id":"b10e09ec9a484e79d05da8d1","modelId":"gpt-6-sol","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":44.114921223355005,"warning":null},{"id":"2d5d5f301f1e2ee16ad2bb5e","modelId":"gpt-6-sol","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":34.9397590361446,"warning":null},{"id":"2dcd27aa0b60624312e45908","modelId":"gpt-6-sol","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":47.914735866543104,"warning":null},{"id":"855fa14ccf7c37cf0afd8e00","modelId":"gpt-6-sol","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":40.9638554216867,"warning":null},{"id":"d42c6f5647236d5c0bf1e421","modelId":"gpt-6-sol","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":18.396663577386498,"warning":null},{"id":"12e3cb8c3f650c504767af49","modelId":"gpt-6-sol","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":46.2928637627433,"warning":null},{"id":"9c02d2e039e7542a491d73d6","modelId":"gpt-6-sol","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":49.4350282485876,"warning":null},{"id":"b7aea959dd9231579d9dcc0f","modelId":"gpt-6-sol","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":16.1111111111111,"warning":null},{"id":"4f96a6b4084ca01d8674b3d1","modelId":"gpt-6-sol","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":81.9653179190751,"warning":null},{"id":"dffa90b77a2844423ea867e2","modelId":"gpt-6-sol","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":80.1734104046243,"warning":null},{"id":"df6af471f097ec031232e29c","modelId":"gpt-6-sol","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":82.94797687861269,"warning":null},{"id":"52324c8bd3220fd1d2ed5019","modelId":"gpt-6-sol","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":80.4046242774566,"warning":null},{"id":"2574bc448482ab1eafc301d2","modelId":"gpt-6-sol","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":70.3468208092486,"warning":null},{"id":"4ad8011ccfb97221b686dba1","modelId":"gpt-6-sol","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":82.5433526011561,"warning":null},{"id":"1154bdd990076a6f598e4ea3","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":26.8166666666667,"warning":null},{"id":"94cfcf715e69e4cc2280424b","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":26.5166666666667,"warning":null},{"id":"f86dc82f576fec44f2d4cccf","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":27.1166666666667,"warning":null},{"id":"34432444ac91998692a575ea","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":27.0166666666667,"warning":null},{"id":"eef671de884b38fd10c41087","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":-0.85,"warning":null},{"id":"72685ddb1ea0dff79cde3146","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":26.6666666666667,"warning":null},{"id":"78e0be78ce3758451826c570","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":53.71666666666667,"warning":null},{"id":"aa090fe0381ab3fc0e867219","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":51.24999999999999,"warning":null},{"id":"463d57efc495213abd110bdd","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":54.48333333333333,"warning":null},{"id":"ebb5431a52a2685db93ec30e","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":53.449999999999996,"warning":null},{"id":"30d83050aa0d748594cdf574","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":45.18333333333333,"warning":null},{"id":"50b4823c2092983d29b42145","modelId":"gpt-6-sol","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":53.849999999999994,"warning":null},{"id":"8212c304db0a4fdfcf231c60","modelId":"gpt-6-sol","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":54.86111111111111,"warning":null},{"id":"1a0703351660c70d1889457a","modelId":"gpt-6-sol","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":50.2314814814815,"warning":null},{"id":"2a4c60137e91918664141ffa","modelId":"gpt-6-sol","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":57.63888888888889,"warning":null},{"id":"561b959d2c87575ae98baabb","modelId":"gpt-6-sol","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":53.8194444444444,"warning":null},{"id":"f17dd5a9a9a4bbb5248c51e1","modelId":"gpt-6-sol","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":47.337962962963,"warning":null},{"id":"ecef52afc6967c3cb896841f","modelId":"gpt-6-sol","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":55.0925925925926,"warning":null},{"id":"acb9caf9ee703de30b9231da","modelId":"gpt-6-sol","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":17.1428571428571,"warning":null},{"id":"40d900586fdf5979a0c29d55","modelId":"gpt-6-sol","benchmarkId":"aa-terminal-bench-science","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":3.33333333333333,"warning":null},{"id":"7aa964f9245acce58db84785","modelId":"gpt-6-sol","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":30,"warning":null},{"id":"aa6ef65aaad68669ffccd796","modelId":"gpt-6-sol","benchmarkId":"aa-terminal-bench-science","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":8.57142857142857,"warning":null},{"id":"ceca3b04b35d861f9d5c33ad","modelId":"gpt-6-sol","benchmarkId":"aa-terminal-bench-science","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":5.71428571428571,"warning":null},{"id":"bef41a807968eaba154d7d25","modelId":"gpt-6-sol","benchmarkId":"aa-terminal-bench-science","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":17.6190476190476,"warning":null},{"id":"72a43fba42e0926927f1bf5e","modelId":"gpt-6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-sol-high","score":26.2626262626263,"warning":null},{"id":"de4817c92f4bb68295a9aaef","modelId":"gpt-6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-sol-low","score":9.09090909090909,"warning":null},{"id":"594142495acc5f2492b7e7ac","modelId":"gpt-6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-sol","score":43.9393939393939,"warning":null},{"id":"3da9ac10f9618218ac7ebc12","modelId":"gpt-6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-sol-medium","score":18.6868686868687,"warning":null},{"id":"6ce189d994550c6b0782154a","modelId":"gpt-6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"GPT-6 Sol (Non-reasoning)","configuration":"GPT-6 Sol (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-6-sol-non-reasoning","score":13.1313131313131,"warning":null},{"id":"f0239ed164f14d4210cfe69b","modelId":"gpt-6-sol","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-6 Sol (Xhigh)","configuration":"GPT-6 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-sol-xhigh","score":30.303030303030297,"warning":null},{"id":"0ed90f472fa6fcd20b3f44f4","modelId":"gpt-6-sol","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-6 Sol (High)","configuration":"GPT-6 Sol (High); Source model ID: openai-gpt-6-sol-high; 120 tasks","score":68.88888888888887,"warning":null},{"id":"64c804965cb22828aa731d12","modelId":"gpt-6-sol","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-6 Sol (Low)","configuration":"GPT-6 Sol (Low); Source model ID: openai-gpt-6-sol-low; 120 tasks","score":31.52777777777777,"warning":null},{"id":"8893871195c4275354b68816","modelId":"gpt-6-sol","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"GPT-6 Sol (Max); Source model ID: openai-gpt-6-sol-max; 120 tasks","score":89.58333333333334,"warning":null},{"id":"e37cfc02ce7c52019f9a3f46","modelId":"gpt-6-sol","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-6 Sol (Medium)","configuration":"GPT-6 Sol (Medium); Source model ID: openai-gpt-6-sol-medium; 120 tasks","score":57.777777777777786,"warning":null},{"id":"6824b8899439016d089b9c2b","modelId":"gpt-6-sol","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"GPT-6 Sol (None)","configuration":"GPT-6 Sol (None); Source model ID: openai-gpt-6-sol-none; 120 tasks","score":1.6666666666666667,"warning":null},{"id":"f6230374a382c253de51b3d6","modelId":"gpt-6-sol","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-6 Sol (XHigh)","configuration":"GPT-6 Sol (XHigh); Source model ID: openai-gpt-6-sol-xhigh; 120 tasks","score":78.05555555555556,"warning":null},{"id":"3cd5ffa0af0a02d00875f995","modelId":"gpt-6.1-sol","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":50,"warning":null},{"id":"f03006b520978bf3c24f99da","modelId":"gpt-6.1-sol","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":1465.25,"warning":null},{"id":"2609c5cfdfd05688f687ca40","modelId":"gpt-6.1-sol","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":1116.74,"warning":null},{"id":"aa56776633010d47a0f9d44e","modelId":"gpt-6.1-sol","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":1557,"warning":null},{"id":"2e385e660dd6a446c9e7f123","modelId":"gpt-6.1-sol","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":1361.16,"warning":null},{"id":"c46a80f217bde09c3c6798f7","modelId":"gpt-6.1-sol","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":1503.6,"warning":null},{"id":"0594b1ba8329f0c608e25a00","modelId":"gpt-6.1-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":82.3333333333333,"warning":null},{"id":"186b979ca9c564ecb0fc397a","modelId":"gpt-6.1-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":84,"warning":null},{"id":"bdc25bd6f48e029af6be763a","modelId":"gpt-6.1-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":83,"warning":null},{"id":"0a574bb316c4409afabbf2ec","modelId":"gpt-6.1-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":83.3333333333333,"warning":null},{"id":"ad6adcfae88d17c02a260723","modelId":"gpt-6.1-sol","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":79.6666666666667,"warning":null},{"id":"a99e81d0301ecb84ae40215f","modelId":"gpt-6.1-sol","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":64.49523732184466,"warning":null},{"id":"eaa63db76aff6dc401cc1e7d","modelId":"gpt-6.1-sol","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":52.57088964747538,"warning":null},{"id":"bead7cc623b5c67c3764e537","modelId":"gpt-6.1-sol","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":64.86844882398171,"warning":null},{"id":"d265f0613e786b9c4c0fbbae","modelId":"gpt-6.1-sol","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":62.573509704356866,"warning":null},{"id":"a787f23d809b41df18b662a1","modelId":"gpt-6.1-sol","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":66.57372141364344,"warning":null},{"id":"cdb0cdc178b48a3fa1cbd27e","modelId":"gpt-6.1-sol","benchmarkId":"aa-critpt","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":30,"warning":null},{"id":"2e501ec3aa84e9017a6ea8bb","modelId":"gpt-6.1-sol","benchmarkId":"aa-critpt","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":24.8571428571429,"warning":null},{"id":"82ceeab1edf55e2721cbabe1","modelId":"gpt-6.1-sol","benchmarkId":"aa-critpt","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":31.7142857142857,"warning":null},{"id":"3b62ed798fdedd9180acac61","modelId":"gpt-6.1-sol","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":27.7142857142857,"warning":null},{"id":"0a5f7cd42f3ba7fd46dded96","modelId":"gpt-6.1-sol","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":31.7142857142857,"warning":null},{"id":"dbadf041bd9535c6a44dad3b","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":32,"warning":null},{"id":"d2fb5d79934d0805a4aab601","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":27,"warning":null},{"id":"89645671976a89c8c393ad0b","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":31,"warning":null},{"id":"bbe9f6ce36d0956fb0383d92","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":30,"warning":null},{"id":"96e84979cd4c9d68561332ac","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":31.8,"warning":null},{"id":"96dd08e2c22e61583e6b1179","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":1511.35,"warning":null},{"id":"fd60d0d9979f7f5c5359ee3b","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":1314.67,"warning":null},{"id":"a9421e2ad943574cceda5ba1","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":1593.85,"warning":null},{"id":"620edaad9daf5c1f00277437","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":1450.1,"warning":null},{"id":"79185b29d71d8ff55a770faf","modelId":"gpt-6.1-sol","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":1538.05,"warning":null},{"id":"ba6fc06f2b712fecf3094d9b","modelId":"gpt-6.1-sol","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":6.944444444444444,"warning":null},{"id":"ecae7a56059af0e1b44ea5bb","modelId":"gpt-6.1-sol","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":51.390176088971295,"warning":null},{"id":"72ad8c0de9853e14527f9d4b","modelId":"gpt-6.1-sol","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":47.3586654309546,"warning":null},{"id":"d99148ab1b49ecd806978224","modelId":"gpt-6.1-sol","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":52.919369786839695,"warning":null},{"id":"eb67b13ba47ce09622d71f7c","modelId":"gpt-6.1-sol","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":49.8609823911029,"warning":null},{"id":"310385827f9524f8744d0596","modelId":"gpt-6.1-sol","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":52.6413345690454,"warning":null},{"id":"2c3bcfca102abc0025a7c83c","modelId":"gpt-6.1-sol","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":33.8888888888889,"warning":null},{"id":"92d34fdca8840182cb2210ca","modelId":"gpt-6.1-sol","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":84.9132947976879,"warning":null},{"id":"d03eb34e4a90dde4ea96c573","modelId":"gpt-6.1-sol","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":83.121387283237,"warning":null},{"id":"56f52aac7cf62029873b4b28","modelId":"gpt-6.1-sol","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":85.9537572254335,"warning":null},{"id":"1b800b4de20f1d6ece9fa0bd","modelId":"gpt-6.1-sol","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":83.93063583815031,"warning":null},{"id":"47ff1810755166a61a00ec3c","modelId":"gpt-6.1-sol","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":85.08670520231209,"warning":null},{"id":"64f3fac7e6fe92bc9f326811","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":41.45,"warning":null},{"id":"f8cc520aa71efc9d2cca2ed5","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":37.6333333333333,"warning":null},{"id":"e386ccb4642f088a69b91e74","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":41.5333333333333,"warning":null},{"id":"c0c872113c362c920957fa03","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":39.95,"warning":null},{"id":"03bdd4a2b19a8e8b281dfc5d","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":40.8666666666667,"warning":null},{"id":"9d58f13b1d1723d9227d1cec","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":60.8,"warning":null},{"id":"dab323260123be447a6905d6","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":58.86666666666667,"warning":null},{"id":"3dd192ee7883e912f32fe4ef","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":62.1,"warning":null},{"id":"404a074773a6d7d353c39c48","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":60.4,"warning":null},{"id":"e9a9526a6aa0000835e3810e","modelId":"gpt-6.1-sol","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":60.81666666666666,"warning":null},{"id":"39da558f9bc667508c72560e","modelId":"gpt-6.1-sol","benchmarkId":"aa-scicode","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":55.787037037037,"warning":null},{"id":"9ed1c59bf4a943cdd58181e0","modelId":"gpt-6.1-sol","benchmarkId":"aa-scicode","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":53.2407407407407,"warning":null},{"id":"74d95ec2338ee9f4688df1b0","modelId":"gpt-6.1-sol","benchmarkId":"aa-scicode","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":54.1666666666667,"warning":null},{"id":"def63ce73e46d175b4c3e3cf","modelId":"gpt-6.1-sol","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":53.2407407407407,"warning":null},{"id":"0d166f1c77c1d87e8fbd3608","modelId":"gpt-6.1-sol","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":55.671296296296305,"warning":null},{"id":"d3801dffbe393dd5b2fbea8b","modelId":"gpt-6.1-sol","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":58.0952380952381,"warning":null},{"id":"b418f646b7b7bc49a83cf5b9","modelId":"gpt-6.1-sol","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-high","score":51.5151515151515,"warning":null},{"id":"c0a2f93dfb74de3d75b15bb0","modelId":"gpt-6.1-sol","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-low","score":30.808080808080803,"warning":null},{"id":"1bc6b4a566ebb8b1b9a5d5fa","modelId":"gpt-6.1-sol","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol","score":56.0606060606061,"warning":null},{"id":"fda82aabe8d4043e2eec41d0","modelId":"gpt-6.1-sol","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-medium","score":47.979797979798,"warning":null},{"id":"4c41bdfde4a7cc3f17da8208","modelId":"gpt-6.1-sol","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"GPT-6.1 Sol (Xhigh)","configuration":"GPT-6.1 Sol (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-6-1-sol-xhigh","score":54.040404040404,"warning":null},{"id":"36c9e6284ae54fa051152ee0","modelId":"gpt-6.1-sol","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"GPT-6.1 Sol (High)","configuration":"GPT-6.1 Sol (High); Source model ID: openai-gpt-6-1-sol-high; 120 tasks","score":91.66666666666666,"warning":null},{"id":"cbdf066b748ed67709eff6f2","modelId":"gpt-6.1-sol","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"GPT-6.1 Sol (Low)","configuration":"GPT-6.1 Sol (Low); Source model ID: openai-gpt-6-1-sol-low; 120 tasks","score":76.66666666666667,"warning":null},{"id":"9d1e9dd6a984d63a866082f2","modelId":"gpt-6.1-sol","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"GPT-6.1 Sol (Max)","configuration":"GPT-6.1 Sol (Max); Source model ID: openai-gpt-6-1-sol-max; 120 tasks","score":94.16666666666667,"warning":null},{"id":"c7c6753faaae2aea4f3f4b48","modelId":"gpt-6.1-sol","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"GPT-6.1 Sol (Medium)","configuration":"GPT-6.1 Sol (Medium); Source model ID: openai-gpt-6-1-sol-medium; 120 tasks","score":86.66666666666667,"warning":null},{"id":"94656c0c34c7669c56667a80","modelId":"gpt-6.1-sol","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"GPT-6.1 Sol (XHigh)","configuration":"GPT-6.1 Sol (XHigh); Source model ID: openai-gpt-6-1-sol-xhigh; 120 tasks","score":91.66666666666666,"warning":null},{"id":"ced8d549a0ea37753468659a","modelId":"gpt-oss-120b","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":0,"warning":null},{"id":"0309d9fed1f28ab58e39b794","modelId":"gpt-oss-120b","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":93.4416666666667,"warning":null},{"id":"12b0aeab0a719995d6dbcbd1","modelId":"gpt-oss-120b","benchmarkId":"aa-aime-2025","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":66.6666666666667,"warning":null},{"id":"e33b40dca705cbeafd7d840b","modelId":"gpt-oss-120b","benchmarkId":"aa-apex-agents-aa","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":3.09734513274336,"warning":null},{"id":"5c8030dafb1cbd9a32a23db4","modelId":"gpt-oss-120b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":52,"warning":null},{"id":"96f097fe0259f0b576796ffe","modelId":"gpt-oss-120b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":46,"warning":null},{"id":"fcd236260ec6f006182ba708","modelId":"gpt-oss-120b","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":0.19906990691605594,"warning":null},{"id":"c89833eb6da11665f2d19ab9","modelId":"gpt-oss-120b","benchmarkId":"aa-critpt","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":1.1428571428571401,"warning":null},{"id":"4558d0f0fcad6c3e998424d5","modelId":"gpt-oss-120b","benchmarkId":"aa-critpt","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":0,"warning":null},{"id":"16f38aa48e7dc0ee4bd51431","modelId":"gpt-oss-120b","benchmarkId":"aa-enterprise-ops-gym-aa","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":25.544613548194565,"warning":null},{"id":"490466894285657d38c73135","modelId":"gpt-oss-120b","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":4,"warning":null},{"id":"ef1fd133b2fab48488c25177","modelId":"gpt-oss-120b","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":608.65,"warning":null},{"id":"2cf568b0d5adbb030fb1dce9","modelId":"gpt-oss-120b","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":172.35,"warning":null},{"id":"52c707dbfd8788a88f152bf4","modelId":"gpt-oss-120b","benchmarkId":"aa-global-mmlu-lite","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":82.78125,"warning":null},{"id":"dac688a867617389db82e42d","modelId":"gpt-oss-120b","benchmarkId":"aa-global-mmlu-lite","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":80.11979166666667,"warning":null},{"id":"1b180de766a2c5aa1fb1a986","modelId":"gpt-oss-120b","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":78.1818181818182,"warning":null},{"id":"1bb03f07e109dd0f57c1902d","modelId":"gpt-oss-120b","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":67.1717171717172,"warning":null},{"id":"39ac165948c35c2353884a42","modelId":"gpt-oss-120b","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":19.6014828544949,"warning":null},{"id":"e25562e5afdd6439ca02afeb","modelId":"gpt-oss-120b","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":5.88507877664504,"warning":null},{"id":"57747d23e0d816188c9ed265","modelId":"gpt-oss-120b","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":68.9795918367347,"warning":null},{"id":"6c417fa36a33efe614acbd2c","modelId":"gpt-oss-120b","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":58.299319727891195,"warning":null},{"id":"548bcfe974fb30a055b080a1","modelId":"gpt-oss-120b","benchmarkId":"aa-itbench-aa","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":5.649717514124291,"warning":null},{"id":"b6f271deef58f4c1c242de6a","modelId":"gpt-oss-120b","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":87.83068783068781,"warning":null},{"id":"05c5f6af598d4e4c83190896","modelId":"gpt-oss-120b","benchmarkId":"aa-livecodebench","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":70.6878306878307,"warning":null},{"id":"61f0f16826ed3fff69a9f3c7","modelId":"gpt-oss-120b","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":1.1111111111111098,"warning":null},{"id":"5d44bf9e9f3417025f5591f3","modelId":"gpt-oss-120b","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":80.79288563829789,"warning":null},{"id":"3b2ca4005415b9c67d1d2f17","modelId":"gpt-oss-120b","benchmarkId":"aa-mmlu-pro","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":77.47672872340429,"warning":null},{"id":"b859f17d0623493d9640fab8","modelId":"gpt-oss-120b","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":-49.25,"warning":null},{"id":"85cce49d8e81f9bc7ee8e669","modelId":"gpt-oss-120b","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":-53.5,"warning":null},{"id":"c5cacad218d8f218f749612c","modelId":"gpt-oss-120b","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":21.78333333333333,"warning":null},{"id":"6b14b20e7eae2031c4db204c","modelId":"gpt-oss-120b","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":19.8,"warning":null},{"id":"064d9c4a525afbedbc6ba21a","modelId":"gpt-oss-120b","benchmarkId":"aa-scicode","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":34.0277777777778,"warning":null},{"id":"ee3250e4b56c89d51710b045","modelId":"gpt-oss-120b","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":65.78947368421049,"warning":null},{"id":"7ad7f95b7272eedb47c65791","modelId":"gpt-oss-120b","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":45.0292397660819,"warning":null},{"id":"758fe7bc2df23d07bfebbdeb","modelId":"gpt-oss-120b","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":12.7835051546392,"warning":null},{"id":"395696bc85ef5955427217c8","modelId":"gpt-oss-120b","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":2.88659793814433,"warning":null},{"id":"73dfb0b621edcabb4ae37691","modelId":"gpt-oss-120b","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":23.4848484848485,"warning":null},{"id":"f13625dc17e09b4138efe471","modelId":"gpt-oss-120b","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":5.3030303030303,"warning":null},{"id":"3bc4ac5257b66f647e5ae430","modelId":"gpt-oss-120b","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":26.217228464419502,"warning":null},{"id":"758671dfdf22ed273acb98aa","modelId":"gpt-oss-120b","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"gpt-oss-120b (Low)","configuration":"gpt-oss-120b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-120b-low","score":13.8576779026217,"warning":null},{"id":"c9b09902f2f443330adbe092","modelId":"gpt-oss-120b","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"gpt-oss-120b (High)","configuration":"gpt-oss-120b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-120b","score":0,"warning":null},{"id":"5934164164844b160c937a14","modelId":"gpt-oss-20b","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":0,"warning":null},{"id":"0a5d255578b90f7d09ff9212","modelId":"gpt-oss-20b","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":89.3333333333333,"warning":null},{"id":"66bd95532048e68ec1ede2ce","modelId":"gpt-oss-20b","benchmarkId":"aa-aime-2025","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":62.33333333333329,"warning":null},{"id":"c7b26be5a4c56222d288efc6","modelId":"gpt-oss-20b","benchmarkId":"aa-apex-agents-aa","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":0.737463126843658,"warning":null},{"id":"e7a6cc9c71dec4bfe08388c4","modelId":"gpt-oss-20b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":34.6666666666667,"warning":null},{"id":"b17de0b0a31ad76f4638214c","modelId":"gpt-oss-20b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":31,"warning":null},{"id":"02ac4c77aea5f7aa387520c0","modelId":"gpt-oss-20b","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":0.19111706206154658,"warning":null},{"id":"7ced5cf0a1c4e989b1037c6b","modelId":"gpt-oss-20b","benchmarkId":"aa-critpt","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":1.42857142857143,"warning":null},{"id":"3812f70c16f9c303746335f4","modelId":"gpt-oss-20b","benchmarkId":"aa-critpt","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":0,"warning":null},{"id":"8b0841448e02163ad46eae40","modelId":"gpt-oss-20b","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":2,"warning":null},{"id":"b7396c08b518af5d352a13da","modelId":"gpt-oss-20b","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":339.96,"warning":null},{"id":"9bbafc498f6f93d5724d8102","modelId":"gpt-oss-20b","benchmarkId":"aa-global-mmlu-lite","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":75.16145833333333,"warning":null},{"id":"e1438362cb1644cfa2707a35","modelId":"gpt-oss-20b","benchmarkId":"aa-global-mmlu-lite","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":72.84895833333334,"warning":null},{"id":"c8deb519d39812692a6694f2","modelId":"gpt-oss-20b","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":68.7878787878788,"warning":null},{"id":"408cbee3d289cd232faa6ac2","modelId":"gpt-oss-20b","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":61.11111111111111,"warning":null},{"id":"81d5697ed83c92b766bfcd31","modelId":"gpt-oss-20b","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":10.982391102873,"warning":null},{"id":"e539e30aa6409968a9f896f5","modelId":"gpt-oss-20b","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":5.28266913809083,"warning":null},{"id":"fe0021ee9aad24e9e8f07f3a","modelId":"gpt-oss-20b","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":65.10204081632651,"warning":null},{"id":"bc264c47a7aa0aa784268538","modelId":"gpt-oss-20b","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":57.8231292517007,"warning":null},{"id":"1afaea10046864e1e10efd31","modelId":"gpt-oss-20b","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":77.6719576719577,"warning":null},{"id":"013c812ffaa8a292020fab49","modelId":"gpt-oss-20b","benchmarkId":"aa-livecodebench","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":65.1851851851852,"warning":null},{"id":"4e5b624922090d4569657710","modelId":"gpt-oss-20b","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":0.555555555555556,"warning":null},{"id":"9430c43ec50c1dfd7961f3ec","modelId":"gpt-oss-20b","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":74.7672872340426,"warning":null},{"id":"b9e1a0f0a8118e370561f2bc","modelId":"gpt-oss-20b","benchmarkId":"aa-mmlu-pro","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":71.8001994680851,"warning":null},{"id":"4c7158036f4e41eb7f869001","modelId":"gpt-oss-20b","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":-63.05,"warning":null},{"id":"7740302bac02e4bf9f65a113","modelId":"gpt-oss-20b","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":-58.55,"warning":null},{"id":"66942c4e34fd8c153f31e324","modelId":"gpt-oss-20b","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":16,"warning":null},{"id":"799f0ef1838e54e357eb057a","modelId":"gpt-oss-20b","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":15.083333333333334,"warning":null},{"id":"3ef5fecb56a2f9ee3eb881a3","modelId":"gpt-oss-20b","benchmarkId":"aa-scicode","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":38.8888888888889,"warning":null},{"id":"cc0c1d812a9c67106273892a","modelId":"gpt-oss-20b","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":60.23391812865499,"warning":null},{"id":"07be62168e8b26a5fff5793f","modelId":"gpt-oss-20b","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":50.2923976608187,"warning":null},{"id":"6e1c29f079d7bdb3f53c1d1b","modelId":"gpt-oss-20b","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":7.01030927835052,"warning":null},{"id":"6a75156f1b63060dabe563ae","modelId":"gpt-oss-20b","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":10.6060606060606,"warning":null},{"id":"767461fddca474f98603f3d2","modelId":"gpt-oss-20b","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"gpt-oss-20b (Low)","configuration":"gpt-oss-20b (Low); Model source: https://artificialanalysis.ai/models/gpt-oss-20b-low","score":4.545454545454549,"warning":null},{"id":"63de0a7b1b2a5a587da79d3e","modelId":"gpt-oss-20b","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":13.8576779026217,"warning":null},{"id":"12c4a8f22394d87aacdc718f","modelId":"gpt-oss-20b","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"gpt-oss-20b (High)","configuration":"gpt-oss-20b (High); Model source: https://artificialanalysis.ai/models/gpt-oss-20b","score":0,"warning":null},{"id":"fcb20410679cc645a033b3d2","modelId":"grok-4.3","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":8.75,"warning":null},{"id":"0c23e68a2c737d3b15f8d9eb","modelId":"grok-4.3","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":761.88,"warning":null},{"id":"b93ede08c25ca59edbf79a63","modelId":"grok-4.3","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":724.34,"warning":null},{"id":"0c4aa333831711af1795a04b","modelId":"grok-4.3","benchmarkId":"aa-apex-agents-aa","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":17.0353982300885,"warning":null},{"id":"868d29a0a89ca2593435d61e","modelId":"grok-4.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":73,"warning":null},{"id":"5e3d55a805ac62634afe6674","modelId":"grok-4.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":74,"warning":null},{"id":"effcb53f94033bb5b5f966e5","modelId":"grok-4.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":75,"warning":null},{"id":"a8c5adb6e6a04bb44455088a","modelId":"grok-4.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":32.3333333333333,"warning":null},{"id":"3829fa62250ab8e4b12c6784","modelId":"grok-4.3","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":1.375964614032744,"warning":null},{"id":"90dc9f391ef590950102f348","modelId":"grok-4.3","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":0.9304541594200506,"warning":null},{"id":"e0c3283a2818e671b33c3cdc","modelId":"grok-4.3","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":8,"warning":null},{"id":"c065359d1b28a74d95ac6a08","modelId":"grok-4.3","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":0.571428571428571,"warning":null},{"id":"f33b8378a9cc0617ef77c8f9","modelId":"grok-4.3","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":4.857142857142859,"warning":null},{"id":"a58cdd26382eb8561a3891c9","modelId":"grok-4.3","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":0,"warning":null},{"id":"60324e3c9645e7080d999c7b","modelId":"grok-4.3","benchmarkId":"aa-enterprise-ops-gym-aa","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":26.260817666368247,"warning":null},{"id":"1e832199e12ef67612abc6d5","modelId":"grok-4.3","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":5.800000000000001,"warning":null},{"id":"76697b005c207eca60a37292","modelId":"grok-4.3","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":2.8000000000000003,"warning":null},{"id":"0ad5c42503be10e35d595387","modelId":"grok-4.3","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":937.89,"warning":null},{"id":"814a9734e4b2bbfc066ca701","modelId":"grok-4.3","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":942.17,"warning":null},{"id":"91e9772198f05a74363fba57","modelId":"grok-4.3","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":90.1010101010101,"warning":null},{"id":"23d0673b32dfc41559f0e973","modelId":"grok-4.3","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":84.3434343434343,"warning":null},{"id":"3524a178f5b2a716aaa82cb0","modelId":"grok-4.3","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":88.98989898989899,"warning":null},{"id":"87938cd53777003184b7326c","modelId":"grok-4.3","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":65.75757575757581,"warning":null},{"id":"abb38a5513376fec0b76e72b","modelId":"grok-4.3","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":37.2103799814643,"warning":null},{"id":"149738d9f5a4cd1b35b66376","modelId":"grok-4.3","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":18.3503243744208,"warning":null},{"id":"b293d93dd837a3122c1aaeb6","modelId":"grok-4.3","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":29.9814643188137,"warning":null},{"id":"48b6077c8917596c2ef7209d","modelId":"grok-4.3","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":6.76552363299351,"warning":null},{"id":"2e09ee8fbbda2f4baad8ce3a","modelId":"grok-4.3","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":81.2925170068027,"warning":null},{"id":"07f4e5e88aea55a272cde6bf","modelId":"grok-4.3","benchmarkId":"aa-ifbench","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":80.95238095238099,"warning":null},{"id":"6e1f61bfbbd09d45c89a7498","modelId":"grok-4.3","benchmarkId":"aa-ifbench","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":83.3333333333333,"warning":null},{"id":"d6a4bb91832505b0b6e788fd","modelId":"grok-4.3","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":47.6190476190476,"warning":null},{"id":"279bb16a5193ca7753d37635","modelId":"grok-4.3","benchmarkId":"aa-itbench-aa","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":32.7212806026365,"warning":null},{"id":"14a0d3ff6b0f1604c4e00c50","modelId":"grok-4.3","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":10,"warning":null},{"id":"e685db2fccf13af5466ea084","modelId":"grok-4.3","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":78.09248554913299,"warning":null},{"id":"4c9f7b093094673cb9656afd","modelId":"grok-4.3","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":72.7745664739884,"warning":null},{"id":"36e8b5b45a9166179ca20570","modelId":"grok-4.3","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":75.8381502890173,"warning":null},{"id":"f6729a1bc7854706665ff871","modelId":"grok-4.3","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":64.7976878612717,"warning":null},{"id":"09b6428fa759d908ccb5afae","modelId":"grok-4.3","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":17.9833333333333,"warning":null},{"id":"2e2d40dd58af61ba5dcf3f71","modelId":"grok-4.3","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":13.9333333333333,"warning":null},{"id":"4d7a7ea79d14224389100352","modelId":"grok-4.3","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":16.7,"warning":null},{"id":"723887484db7ec3fbb615c5c","modelId":"grok-4.3","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":-33.0833333333333,"warning":null},{"id":"e1b636f4f66564ff9eb90416","modelId":"grok-4.3","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":34.78333333333333,"warning":null},{"id":"899ff9232d3fe76477a4871a","modelId":"grok-4.3","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":26.68333333333333,"warning":null},{"id":"5192d10280a324241a7798e5","modelId":"grok-4.3","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":28.76666666666667,"warning":null},{"id":"ff80f12b5df368a65cf2f2b9","modelId":"grok-4.3","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":23.716666666666665,"warning":null},{"id":"27b5321c003a7d31cf5acac8","modelId":"grok-4.3","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":48.2638888888889,"warning":null},{"id":"0561533dc2911a9c3050e1f3","modelId":"grok-4.3","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":39.3518518518519,"warning":null},{"id":"c088c53febbfe737539c90e3","modelId":"grok-4.3","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":97.6608187134503,"warning":null},{"id":"5455dd2563a8b53a9421655a","modelId":"grok-4.3","benchmarkId":"aa-tau2-bench","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":88.8888888888889,"warning":null},{"id":"9cbfd7e03e0375e2d24f16f5","modelId":"grok-4.3","benchmarkId":"aa-tau2-bench","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":91.2280701754386,"warning":null},{"id":"1676bd960d1324c582d1bb4e","modelId":"grok-4.3","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":65.78947368421049,"warning":null},{"id":"aafdef13117c7ef1ecd64e96","modelId":"grok-4.3","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":12.3711340206186,"warning":null},{"id":"bb11904c98e84a986b9588e1","modelId":"grok-4.3","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":8.04123711340206,"warning":null},{"id":"7e98f92df33d19a1a1db88ed","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":37.878787878787904,"warning":null},{"id":"4088729066d1338b03ef8273","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-hard","level":"low","sourceLabel":"Grok 4.3 (Low)","configuration":"Grok 4.3 (Low); Model source: https://artificialanalysis.ai/models/grok-4-3-low","score":26.5151515151515,"warning":null},{"id":"50efe0ae8d19e57558f6b3e6","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-hard","level":"medium","sourceLabel":"Grok 4.3 (Medium)","configuration":"Grok 4.3 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-3-medium","score":30.303030303030297,"warning":null},{"id":"a316a3c001d815220879252c","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":18.939393939393902,"warning":null},{"id":"99fcba55991fe79d06dfb1d3","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":39.700374531835195,"warning":null},{"id":"4f472a30aeb8eaf66deeba1b","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":34.0823970037453,"warning":null},{"id":"ca38047ed324858c3e495dff","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Grok 4.3 (High)","configuration":"Grok 4.3 (High); Model source: https://artificialanalysis.ai/models/grok-4-3","score":0,"warning":null},{"id":"35daaac2931883507e622ad1","modelId":"grok-4.3","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Grok 4.3 (Non-reasoning)","configuration":"Grok 4.3 (Non-reasoning); Model source: https://artificialanalysis.ai/models/grok-4-3-non-reasoning","score":0,"warning":null},{"id":"3dabd7de22313e1ac7daf67f","modelId":"grok-4.5","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":35,"warning":null},{"id":"71d11b4c4ffe6fcbf4157f60","modelId":"grok-4.5","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":1280.57,"warning":null},{"id":"63f735654eaeb7e5ecc2e494","modelId":"grok-4.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":79.3333333333333,"warning":null},{"id":"c3e471cc21bf38fd994ca247","modelId":"grok-4.5","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":57.93486035084273,"warning":null},{"id":"5cf7c08965507cd0d4d2c0ac","modelId":"grok-4.5","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":15.4285714285714,"warning":null},{"id":"64a2f492b0082eab536fda7f","modelId":"grok-4.5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":40.82363473589973,"warning":null},{"id":"fb530f6406ed45847a6ddf51","modelId":"grok-4.5","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":18.8,"warning":null},{"id":"681c3f9afb34ecdcaf3196d1","modelId":"grok-4.5","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":1384.26,"warning":null},{"id":"8c2590ca713df819e8c04be3","modelId":"grok-4.5","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":93.13131313131309,"warning":null},{"id":"a5cf14fe15d8d387a93d2501","modelId":"grok-4.5","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":42.678405931418,"warning":null},{"id":"fe902bc3437926b36a2718b3","modelId":"grok-4.5","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":15,"warning":null},{"id":"4e0ce99a30705489f0356e12","modelId":"grok-4.5","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":80.4046242774566,"warning":null},{"id":"c682e32d9ea064efabe1dde6","modelId":"grok-4.5","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":25.3166666666667,"warning":null},{"id":"d40cab64d204a29f18a02f30","modelId":"grok-4.5","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":51.55,"warning":null},{"id":"16057f70ea8e1d33f54b0acc","modelId":"grok-4.5","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":54.976851851851805,"warning":null},{"id":"5d28cc8fec74fe882717bfe8","modelId":"grok-4.5","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":42.0618556701031,"warning":null},{"id":"70fda2ccc5e5d33aa3ea7bad","modelId":"grok-4.5","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":81.6479400749064,"warning":null},{"id":"fa92272b6b9aa5d1ab34f5fe","modelId":"grok-4.5","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Model source: https://artificialanalysis.ai/models/grok-4-5","score":10.6060606060606,"warning":null},{"id":"3db181a4b2c903d1c6f5ca32","modelId":"grok-4.5","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Grok 4.5 (High)","configuration":"Grok 4.5 (High); Source model ID: xai-grok-4-5-high; 120 tasks","score":52.64,"warning":null},{"id":"6b3763ee781610c75ebd4561","modelId":"grok-4.5","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Grok 4.5 (Low)","configuration":"Grok 4.5 (Low); Source model ID: xai-grok-4-5-low; 120 tasks","score":33.06,"warning":null},{"id":"f3dd688412c9fb5675915351","modelId":"grok-4.5","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Grok 4.5 (Medium)","configuration":"Grok 4.5 (Medium); Source model ID: xai-grok-4-5-medium; 120 tasks","score":52.64,"warning":null},{"id":"29895fdf00c54de0c01daedb","modelId":"grok-4.5","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"grok-4-5","configuration":"grok-4-5; Reported effort: high; 113 unique tasks; 452 attempts; 4 runs","score":53.8,"warning":null},{"id":"43896251f67c4560c987eefb","modelId":"grok-4.5","benchmarkId":"terminal-cursorcli","level":"high","sourceLabel":"Grok 4.5","configuration":"Grok 4.5; Reported effort: high; 445 trials; recorded 2026-07-08","score":79.33,"warning":null},{"id":"f097cd0ecb24da6e29e3b25e","modelId":"grok-4.6","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":41.25,"warning":null},{"id":"32dad52a859e3688e67953f4","modelId":"grok-4.6","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":1534.25,"warning":null},{"id":"fed456248fce47a1dfad8571","modelId":"grok-4.6","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":1287.61,"warning":null},{"id":"aa13b44d83668e44594a40fc","modelId":"grok-4.6","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":1483.19,"warning":null},{"id":"a05b27a6d78bf2989daa942b","modelId":"grok-4.6","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":1544.92,"warning":null},{"id":"caac1d9fab456ec673932b5f","modelId":"grok-4.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":80.3333333333333,"warning":null},{"id":"a3756d310ea1a7f832e597bc","modelId":"grok-4.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":80.6666666666667,"warning":null},{"id":"aa49e45426197f51738e5ddc","modelId":"grok-4.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":81,"warning":null},{"id":"9a83ca80e442b816d313353b","modelId":"grok-4.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":81,"warning":null},{"id":"c89e8c3a3b6387ddb30e1f52","modelId":"grok-4.6","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":66.67849638121398,"warning":null},{"id":"5f885ebd961ac88f984a7704","modelId":"grok-4.6","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":53.44540059031701,"warning":null},{"id":"e10dc4ce6438d32bac1db209","modelId":"grok-4.6","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":63.151269753678996,"warning":null},{"id":"a46db1e98a53a9b337571cf6","modelId":"grok-4.6","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":66.99736836183247,"warning":null},{"id":"b6ff61e203ecccc9e6c2ee6a","modelId":"grok-4.6","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":17.1428571428571,"warning":null},{"id":"319d744c56b01ed6312991b2","modelId":"grok-4.6","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":5.71428571428571,"warning":null},{"id":"097d77ab7b94acd9ac1fdc59","modelId":"grok-4.6","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":17.714285714285698,"warning":null},{"id":"31d13c37ff4e5cf0553752d3","modelId":"grok-4.6","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":19.7142857142857,"warning":null},{"id":"26b2805c615a5f784ff6efce","modelId":"grok-4.6","benchmarkId":"aa-enterprise-ops-gym-aa","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":48.343777976723366,"warning":null},{"id":"b1f42e7a56eda147f999bfb4","modelId":"grok-4.6","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":17,"warning":null},{"id":"9a8bf5de95e570a3eab39900","modelId":"grok-4.6","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":15.8,"warning":null},{"id":"36e6bda4002c33d4a26d2338","modelId":"grok-4.6","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":17.8,"warning":null},{"id":"8d23777f76973ac28d2c047a","modelId":"grok-4.6","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":17.2,"warning":null},{"id":"8c122c70c951c4afb9d366cb","modelId":"grok-4.6","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":1618.11,"warning":null},{"id":"ee3e645bcb99d1cb0d07e251","modelId":"grok-4.6","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":1410.2,"warning":null},{"id":"ce430630902908f616d092b8","modelId":"grok-4.6","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":1612.18,"warning":null},{"id":"5bddb9605d2c7ea502bea2d8","modelId":"grok-4.6","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":1645.38,"warning":null},{"id":"3a4f7b3fd7d65f88e2ebb12a","modelId":"grok-4.6","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":94.94949494949499,"warning":null},{"id":"455042cc2a2352e34ec6dc96","modelId":"grok-4.6","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":87.8787878787879,"warning":null},{"id":"3d33ea8f9ff471640bd35ed8","modelId":"grok-4.6","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":93.53535353535351,"warning":null},{"id":"0e43ef1722511f9eafa9fe90","modelId":"grok-4.6","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":93.53535353535351,"warning":null},{"id":"7496ffb374d886017a611cb6","modelId":"grok-4.6","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":42.9101019462465,"warning":null},{"id":"4c7b243e05b68d3c1adfc1ba","modelId":"grok-4.6","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":27.6181649675626,"warning":null},{"id":"937d3156bb8d0d9f4f627073","modelId":"grok-4.6","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":42.122335495829496,"warning":null},{"id":"c374d0cb6645fba1f3e93070","modelId":"grok-4.6","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":44.0685820203893,"warning":null},{"id":"692eafdc8c386d4794b260e4","modelId":"grok-4.6","benchmarkId":"aa-mlcr-aa","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":12.2222222222222,"warning":null},{"id":"6f49a8c226bc88f3c493b4ba","modelId":"grok-4.6","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":30.4833333333333,"warning":null},{"id":"505a2829d55dcb2770e948af","modelId":"grok-4.6","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":25.9,"warning":null},{"id":"e347da11309bbb7ed5c362fc","modelId":"grok-4.6","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":28,"warning":null},{"id":"4d8998f59334f3d8011e1014","modelId":"grok-4.6","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":29.3166666666667,"warning":null},{"id":"9042a25a07406f73250d8bba","modelId":"grok-4.6","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":48.233333333333334,"warning":null},{"id":"5fbac91d6ca09c762b0ca355","modelId":"grok-4.6","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":43.28333333333333,"warning":null},{"id":"6c42342ff86119e143676f0c","modelId":"grok-4.6","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":41.93333333333334,"warning":null},{"id":"76a9250c25e23962a7071972","modelId":"grok-4.6","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":43,"warning":null},{"id":"8bbdaf9bb9135654d3bcdac6","modelId":"grok-4.6","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":56.4814814814815,"warning":null},{"id":"1a152b4c4e47ce624c9d5c0b","modelId":"grok-4.6","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":49.421296296296305,"warning":null},{"id":"05d718901ba5f820a250f83c","modelId":"grok-4.6","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":55.9027777777778,"warning":null},{"id":"456f8aba84ca2b03c8b14c6d","modelId":"grok-4.6","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":53.0092592592593,"warning":null},{"id":"2866a9f1d6bddec801f86d03","modelId":"grok-4.6","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":50.7216494845361,"warning":null},{"id":"cbd676d3e23e7861d13eaee8","modelId":"grok-4.6","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":38.144329896907195,"warning":null},{"id":"ed0625bb1734d51f8e228486","modelId":"grok-4.6","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":44.3298969072165,"warning":null},{"id":"3b9378efd7cabab48ea16c28","modelId":"grok-4.6","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":43.298969072165,"warning":null},{"id":"40537ca23e510726cdd85553","modelId":"grok-4.6","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":6.19047619047619,"warning":null},{"id":"cb27d5e2462f1dcbcca8ce0f","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":88.38951310861421,"warning":null},{"id":"ab78e0ea2f5903fc7ae244fd","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":75.2808988764045,"warning":null},{"id":"7235f070956c1bc0fd07c00c","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":84.2696629213483,"warning":null},{"id":"04b824c5fa68cf8f4799f751","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":88.0149812734082,"warning":null},{"id":"b769630d6b04321a1703df93","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Model source: https://artificialanalysis.ai/models/grok-4-6","score":21.2121212121212,"warning":null},{"id":"8dccc22f93c6a00e5ac799c3","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Model source: https://artificialanalysis.ai/models/grok-4-6-low","score":3.03030303030303,"warning":null},{"id":"378296da344991b54973fec3","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Model source: https://artificialanalysis.ai/models/grok-4-6-medium","score":13.1313131313131,"warning":null},{"id":"acac33735d2218bbe612835e","modelId":"grok-4.6","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Grok 4.6 (Xhigh)","configuration":"Grok 4.6 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-6-xhigh","score":17.171717171717198,"warning":null},{"id":"f494121ba1b32eb13d788d7e","modelId":"grok-4.6","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Grok 4.6 (High)","configuration":"Grok 4.6 (High); Source model ID: xai-grok-4-6-high; 120 tasks","score":65.13888888888889,"warning":null},{"id":"c80d116c611cdfdbf0794a0e","modelId":"grok-4.6","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Grok 4.6 (Low)","configuration":"Grok 4.6 (Low); Source model ID: xai-grok-4-6-low; 120 tasks","score":27.63888888888889,"warning":null},{"id":"9cc5971c94263e7ed3b6e5aa","modelId":"grok-4.6","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Grok 4.6 (Medium)","configuration":"Grok 4.6 (Medium); Source model ID: xai-grok-4-6-medium; 120 tasks","score":61.25000000000001,"warning":null},{"id":"055e2840581f27e7013445b1","modelId":"grok-4.6","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"Grok 4.6 (XHigh)","configuration":"Grok 4.6 (XHigh); Source model ID: xai-grok-4-6-xhigh; 120 tasks","score":67.08333333333333,"warning":null},{"id":"3d4862ebd4d7563b0b815232","modelId":"grok-4.6","benchmarkId":"deepswe-minisweagent","level":"high","sourceLabel":"grok-4-6","configuration":"grok-4-6; Reported effort: high; 113 unique tasks; 451 attempts; 4 runs","score":65.2,"warning":null},{"id":"83e82d34b74f64e32fe5e562","modelId":"grok-4.6","benchmarkId":"deepswe-minisweagent","level":"low","sourceLabel":"grok-4-6","configuration":"grok-4-6; Reported effort: low; 113 unique tasks; 449 attempts; 4 runs","score":41.6,"warning":null},{"id":"dee4caacc80397f92e821c17","modelId":"grok-4.6","benchmarkId":"deepswe-minisweagent","level":"medium","sourceLabel":"grok-4-6","configuration":"grok-4-6; Reported effort: medium; 113 unique tasks; 452 attempts; 4 runs","score":67.5,"warning":null},{"id":"bc046d614155011b4f32fd77","modelId":"grok-4.6","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"grok-4-6","configuration":"grok-4-6; Reported effort: xhigh; 113 unique tasks; 451 attempts; 4 runs","score":66.7,"warning":null},{"id":"88d4986edf0441fbb86c3f5e","modelId":"grok-4.7","benchmarkId":"aa-aa-analyst-agent","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":55.00000000000001,"warning":null},{"id":"ec2837169e1f8b98ab031419","modelId":"grok-4.7","benchmarkId":"aa-aa-analyst-agent","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":52.5,"warning":null},{"id":"fa2ec94a46eeece397794f96","modelId":"grok-4.7","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":1632.79,"warning":null},{"id":"38395bbc146d85b41773b2b7","modelId":"grok-4.7","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":1498.6,"warning":null},{"id":"2b69ccdf719cd04281d497ac","modelId":"grok-4.7","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":1641.75,"warning":null},{"id":"5bfb552e224b5e073e2c756c","modelId":"grok-4.7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":77,"warning":null},{"id":"596fd29699291e64305c85ef","modelId":"grok-4.7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":80,"warning":null},{"id":"5caa8fba618fda31e55ccf90","modelId":"grok-4.7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":76.6666666666667,"warning":null},{"id":"bbb1b04a3710bfaeaa939af4","modelId":"grok-4.7","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":63.51937297314658,"warning":null},{"id":"0a4556c4b68bc148bb12e48f","modelId":"grok-4.7","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":56.738374074662836,"warning":null},{"id":"0e287adfb45fde42247168a1","modelId":"grok-4.7","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":65.55835148969457,"warning":null},{"id":"6c16be14a64ddd62a866e476","modelId":"grok-4.7","benchmarkId":"aa-critpt","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":18,"warning":null},{"id":"5ddf4b0e44b7c7f374a21f1d","modelId":"grok-4.7","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":13.4285714285714,"warning":null},{"id":"cf097dc6ba885e21c75400fb","modelId":"grok-4.7","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":17.714285714285698,"warning":null},{"id":"45164a7267b736207948213c","modelId":"grok-4.7","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":23.200000000000003,"warning":null},{"id":"d29b25e8a7b9b6ed2e3ca21e","modelId":"grok-4.7","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":21.4,"warning":null},{"id":"2f9310887cd5aa39bffe8ffa","modelId":"grok-4.7","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":20,"warning":null},{"id":"4d98c7522ce9d75b766dbccd","modelId":"grok-4.7","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":1706.28,"warning":null},{"id":"51ba44b4b10892a5cb41b810","modelId":"grok-4.7","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":1588.11,"warning":null},{"id":"6afe377b2022d0d1d9389fc8","modelId":"grok-4.7","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":1711.07,"warning":null},{"id":"cb87a679236e62548aa0d2c7","modelId":"grok-4.7","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":9.444444444444446,"warning":null},{"id":"0dba0cbed0f8438693fe4aa6","modelId":"grok-4.7","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":42.3076923076923,"warning":null},{"id":"7b3d4a8e4009bd7ee7a18409","modelId":"grok-4.7","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":39.156626506024104,"warning":null},{"id":"b45444502e93ec4564a9b667","modelId":"grok-4.7","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":43.1417979610751,"warning":null},{"id":"65198d32df73ba44f18660a1","modelId":"grok-4.7","benchmarkId":"aa-itbench-aa","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":42.0809792843691,"warning":null},{"id":"7c51934fb450c7aeeb2c2c06","modelId":"grok-4.7","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":15,"warning":null},{"id":"15923d8b71d72b81ad75f32d","modelId":"grok-4.7","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":78.150289017341,"warning":null},{"id":"631e579e32b7dd3021d1344f","modelId":"grok-4.7","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":30.9,"warning":null},{"id":"454e9455a2788947b229c1ce","modelId":"grok-4.7","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":29.6166666666667,"warning":null},{"id":"83ea092e007eb2478396ea2c","modelId":"grok-4.7","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":32.0333333333333,"warning":null},{"id":"74285e625f87512c92b53a27","modelId":"grok-4.7","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":47.81666666666667,"warning":null},{"id":"8a73d97d246ac5b73331dc73","modelId":"grok-4.7","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":49.7,"warning":null},{"id":"f70475a7806986c208e4b475","modelId":"grok-4.7","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":47.449999999999996,"warning":null},{"id":"c266f6af3648f79914736afb","modelId":"grok-4.7","benchmarkId":"aa-scicode","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":57.7546296296296,"warning":null},{"id":"ff1791f34ffe41084fd03a43","modelId":"grok-4.7","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":54.86111111111111,"warning":null},{"id":"6b52ef79e2062a332869f28b","modelId":"grok-4.7","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":57.4074074074074,"warning":null},{"id":"880035a2c237113ecee35a95","modelId":"grok-4.7","benchmarkId":"aa-terminal-bench-science","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":9.52380952380952,"warning":null},{"id":"f07a038bae7cae4fd1379c49","modelId":"grok-4.7","benchmarkId":"aa-terminal-bench-science","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":4.76190476190476,"warning":null},{"id":"d5a917ea2f3c76f70e469243","modelId":"grok-4.7","benchmarkId":"aa-terminal-bench-science","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":4.76190476190476,"warning":null},{"id":"29c4a2f678a2a40bec4900af","modelId":"grok-4.7","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Model source: https://artificialanalysis.ai/models/grok-4-7-high","score":24.747474747474698,"warning":null},{"id":"bc5bf61803cb8e9a3318074c","modelId":"grok-4.7","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Model source: https://artificialanalysis.ai/models/grok-4-7-low","score":16.161616161616198,"warning":null},{"id":"cfde71800b76a410683bd9bd","modelId":"grok-4.7","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Grok 4.7 (Xhigh); Model source: https://artificialanalysis.ai/models/grok-4-7","score":25.7575757575758,"warning":null},{"id":"adf1625727ca01c07bbd8332","modelId":"grok-4.7","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Grok 4.7 (High)","configuration":"Grok 4.7 (High); Source model ID: xai-grok-4-7-high; 120 tasks","score":58.333333333333336,"warning":null},{"id":"ed9081277774b6168520018d","modelId":"grok-4.7","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Grok 4.7 (Low)","configuration":"Grok 4.7 (Low); Source model ID: xai-grok-4-7-low; 120 tasks","score":16.666666666666664,"warning":null},{"id":"cecc3541717ce505ea7525f6","modelId":"grok-4.7","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Grok 4.7 (Medium)","configuration":"Grok 4.7 (Medium); Source model ID: xai-grok-4-7-medium; 120 tasks","score":58.75,"warning":null},{"id":"726551d5156c0a252f55a4d4","modelId":"grok-4.7","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"Grok 4.7 (XHigh)","configuration":"Grok 4.7 (XHigh); Source model ID: xai-grok-4-7-xhigh; 120 tasks","score":61.38888888888888,"warning":null},{"id":"7b1d520fbbabe63fbfb6be51","modelId":"hy3","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":946.66,"warning":null},{"id":"dbe2f8f3df27510117453b6d","modelId":"hy3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":79,"warning":null},{"id":"725cf8be083eeea3761a0746","modelId":"hy3","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.409128371172397,"warning":null},{"id":"ed230f269f79598168400161","modelId":"hy3","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.857142857142859,"warning":null},{"id":"875ff8bc105e8938ba3ab493","modelId":"hy3","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.200000000000001,"warning":null},{"id":"bab2cb3b9298c7559a38c70c","modelId":"hy3","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1066.59,"warning":null},{"id":"c62d7142b2955f3cccde331b","modelId":"hy3","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":89.6969696969697,"warning":null},{"id":"a7a35fd1ba65fa084a53bfff","modelId":"hy3","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.4569045412419,"warning":null},{"id":"4e3737c70846342ba1f20656","modelId":"hy3","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":-18.4666666666667,"warning":null},{"id":"02867ce545a79d28fdca774c","modelId":"hy3","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.966666666666665,"warning":null},{"id":"05c2f7fcf387f4cfcc374ac1","modelId":"hy3","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.6111111111111,"warning":null},{"id":"42aa341468210b376e09693e","modelId":"hy3","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.886597938144302,"warning":null},{"id":"c802eac5e2f5b0cc14a4b3d2","modelId":"hy3","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.4194756554307,"warning":null},{"id":"e9c96fd0e4307f0a3d58c1b2","modelId":"hy3","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Hy3","configuration":"Hy3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/hy3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.505050505050505,"warning":null},{"id":"ef559fa4c43c26694bf787aa","modelId":"hy3-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":42,"warning":null},{"id":"c68b83e1d7b894a93142c739","modelId":"hy3-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":64.6666666666667,"warning":null},{"id":"0b9583430639861c86d85e2f","modelId":"hy3-preview","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"81a106f485bd6b878caf732e","modelId":"hy3-preview","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":4.57142857142857,"warning":null},{"id":"8105b1407e966b7ace150043","modelId":"hy3-preview","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":73.23232323232321,"warning":null},{"id":"c5b5e8eb7d2c919e6919413b","modelId":"hy3-preview","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":86.6666666666667,"warning":null},{"id":"ede59fb652037041ce1a1dd1","modelId":"hy3-preview","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":6.95088044485635,"warning":null},{"id":"fa0d7ae753fb74157d138e17","modelId":"hy3-preview","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":27.757182576459698,"warning":null},{"id":"5a9e1184bb651c83cf370ab6","modelId":"hy3-preview","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":47.959183673469404,"warning":null},{"id":"96069ed2be775a9053422ed9","modelId":"hy3-preview","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":63.1292517006803,"warning":null},{"id":"5c695ae8e66e691e9c8f92ad","modelId":"hy3-preview","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":-34.95,"warning":null},{"id":"79511ae819bfc612c35a390d","modelId":"hy3-preview","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":-35.05,"warning":null},{"id":"e6e605672e21f6d859de5904","modelId":"hy3-preview","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":23.183333333333334,"warning":null},{"id":"3b0fcbd11b64f27c16e87a8d","modelId":"hy3-preview","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":27.933333333333334,"warning":null},{"id":"0695f8b0b910244d05c7a925","modelId":"hy3-preview","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":67.5438596491228,"warning":null},{"id":"9773ef1b60a79df50df32edc","modelId":"hy3-preview","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":92.6900584795322,"warning":null},{"id":"a3c4359e7b3ebe3d66db5807","modelId":"hy3-preview","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Hy3-preview (Non-reasoning)","configuration":"Hy3-preview (Non-reasoning); Model source: https://artificialanalysis.ai/models/hy3-non-reasoning","score":31.818181818181802,"warning":null},{"id":"04ca7fd5e9a8ca87409e10ef","modelId":"hy3-preview","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Hy3-preview (Reasoning)","configuration":"Hy3-preview (Reasoning); Model source: https://artificialanalysis.ai/models/hy3-preview","score":34.0909090909091,"warning":null},{"id":"29ceb6814f660e208c866e8e","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":60,"warning":null},{"id":"210dc08ae104b6881922170a","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.857142857142857,"warning":null},{"id":"64d70b8f34969d4a53e95f09","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.9292929292929,"warning":null},{"id":"099ee1a93cd06c7ab2f69a34","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.5820203892493,"warning":null},{"id":"7e86cbb35e9d7b3c2e6378d0","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":-6.6,"warning":null},{"id":"2780f79aa06f0fd4124fb082","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.05,"warning":null},{"id":"a01cd222d9ce7bcafe04151d","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.0138888888889,"warning":null},{"id":"37a0953e32360929a2941fbd","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.5463917525773,"warning":null},{"id":"b97f8d9cadb3480480d4fc6d","modelId":"k-exaone-2.0-750b-a37b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"K-EXAONE 2.0 0803","configuration":"K-EXAONE 2.0 0803; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/k-exaone-2-0-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.4494382022472,"warning":null},{"id":"940370aaedaeafccbaa815dd","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":44,"warning":null},{"id":"248628f25209685f2e813405","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":90.3333333333334,"warning":null},{"id":"4f07384d9d52a7fe91a6f574","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":53.3333333333333,"warning":null},{"id":"3d7800bef3152ab150d04958","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":61.33333333333329,"warning":null},{"id":"368e01788a286f45ff312089","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":0,"warning":null},{"id":"f6d4cdc6f942e0cd40356df3","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":1.11836734693878,"warning":null},{"id":"fd11ac1482800a35b1dc9043","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":367.49,"warning":null},{"id":"e9b96f3441cf2f7569161137","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":71.03125,"warning":null},{"id":"bcb923d99217ff28ee7af856","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":78.86458333333334,"warning":null},{"id":"2c65d6e1561964eea7a2e608","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":69.4949494949495,"warning":null},{"id":"0437fefd7ee49ace79820857","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":78.2828282828283,"warning":null},{"id":"02cc8dd32ee8e6e63cc2728b","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":5.69972196478221,"warning":null},{"id":"d32f54031e59460e2a5ab5da","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":13.9481000926784,"warning":null},{"id":"5835712a4022a343643485e3","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":39.5918367346939,"warning":null},{"id":"be6b8730183dc85bad1c7223","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":64.69387755102039,"warning":null},{"id":"856a3db006f6d9daa745ceac","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":76.8253968253968,"warning":null},{"id":"27271f132873252b7a8add85","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":80.9757313829787,"warning":null},{"id":"0c644501691c892c4d6e6628","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":83.7516622340426,"warning":null},{"id":"8e4b9809a8173b24a6492b4c","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":-65.4,"warning":null},{"id":"577a7a49fdb90fe94610bce7","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":-57.9666666666667,"warning":null},{"id":"b5681c2b592c28e60853de1c","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":13.65,"warning":null},{"id":"2812d3d5ae36f9f251dd27e7","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":16.35,"warning":null},{"id":"8cc243a5909dbf9cbbc8d10a","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":59.0643274853801,"warning":null},{"id":"1a47670629cd21ee68944b7f","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":74.2690058479532,"warning":null},{"id":"30dab2414c4fc919a207dfd0","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"K-EXAONE (Non-reasoning)","configuration":"K-EXAONE (Non-reasoning); Model source: https://artificialanalysis.ai/models/k-exaone-non-reasoning","score":6.81818181818182,"warning":null},{"id":"7a942a894babe14e3d6e64da","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":22.727272727272698,"warning":null},{"id":"5c4b5d8b71c63b017f4596ad","modelId":"k-exaone-236b-a23b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"K-EXAONE (Reasoning)","configuration":"K-EXAONE (Reasoning); Model source: https://artificialanalysis.ai/models/k-exaone","score":30.337078651685403,"warning":null},{"id":"c0c65d1d51afa78a01a1af9c","modelId":"kimi-k2.6","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":819.82,"warning":null},{"id":"e6b47bd6f606a86747f75627","modelId":"kimi-k2.6","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.4660766961652,"warning":null},{"id":"9a44dc9f59be8bdd5015a312","modelId":"kimi-k2.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":69.6666666666667,"warning":null},{"id":"d4224d6cba3e271e54e68e1b","modelId":"kimi-k2.6","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":81,"warning":null},{"id":"2563f5afc793ef0c6d07cd8f","modelId":"kimi-k2.6","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.020759856386269,"warning":null},{"id":"8b5dee6f02a84bcfc1521def","modelId":"kimi-k2.6","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":1.42857142857143,"warning":null},{"id":"502c49969a4ecdde66e2ab61","modelId":"kimi-k2.6","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":8,"warning":null},{"id":"3d14a2dd2de76000901d9dab","modelId":"kimi-k2.6","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.525813190092514,"warning":null},{"id":"1a5c35e31d7b55148a084a1a","modelId":"kimi-k2.6","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":13,"warning":null},{"id":"6a36893ae047b8ddc312f1bc","modelId":"kimi-k2.6","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":1038.29,"warning":null},{"id":"8dc00c97fd0d8d4c30d4d3ae","modelId":"kimi-k2.6","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":78.7878787878788,"warning":null},{"id":"af97b3aace4b8c10ade79cb2","modelId":"kimi-k2.6","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":91.1111111111111,"warning":null},{"id":"f8fef2fbfef8abe8215008b5","modelId":"kimi-k2.6","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":19.5551436515292,"warning":null},{"id":"9876caaf0789b4d8f2c3993b","modelId":"kimi-k2.6","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.488415199258604,"warning":null},{"id":"072e7d1bc0df4bb01a626b04","modelId":"kimi-k2.6","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":44.2857142857143,"warning":null},{"id":"c1d14aa9f2ac19b07fe9253a","modelId":"kimi-k2.6","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.9863945578231,"warning":null},{"id":"1e536ab84977b3d1dcda2e75","modelId":"kimi-k2.6","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.186440677966097,"warning":null},{"id":"2eff8678da2c3b5a4bf91ee9","modelId":"kimi-k2.6","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.364161849711,"warning":null},{"id":"ca4100db60d1e0bc35d98c00","modelId":"kimi-k2.6","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":-9.18333333333333,"warning":null},{"id":"b666fddeadc496e9d235901c","modelId":"kimi-k2.6","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.3,"warning":null},{"id":"e12c88668886777dfadf2dc1","modelId":"kimi-k2.6","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":24.3,"warning":null},{"id":"1e43b3d72b48539b7330c01d","modelId":"kimi-k2.6","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.6,"warning":null},{"id":"fd0de66f868ec0aa650e1dc1","modelId":"kimi-k2.6","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.5046296296296,"warning":null},{"id":"828fe25b5f302f66f6fc6566","modelId":"kimi-k2.6","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":93.859649122807,"warning":null},{"id":"ceeb0dff0cb147a55964677d","modelId":"kimi-k2.6","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":95.906432748538,"warning":null},{"id":"b0024c6ace8a6e7c4f5b8df5","modelId":"kimi-k2.6","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.298969072164898,"warning":null},{"id":"7e2e74eb92490b361bfd99bf","modelId":"kimi-k2.6","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Kimi K2.6 (Non-reasoning)","configuration":"Kimi K2.6 (Non-reasoning); Model source: https://artificialanalysis.ai/models/kimi-k2-6-non-reasoning","score":37.878787878787904,"warning":null},{"id":"4779b4b11f612232b2ca442d","modelId":"kimi-k2.6","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.9393939393939,"warning":null},{"id":"03e0596ab441c909bef972ee","modelId":"kimi-k2.6","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.9176029962547,"warning":null},{"id":"a9557923995c69135bd637fb","modelId":"kimi-k2.6","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Kimi K2.6 (Reasoning)","configuration":"Kimi K2.6 (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-6; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.505050505050505,"warning":null},{"id":"63bb392b6b3abf71770167e7","modelId":"kimi-k2.7-code","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":858.75,"warning":null},{"id":"88387229e8cce0c4b453061b","modelId":"kimi-k2.7-code","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.3333333333333,"warning":null},{"id":"ab60dcdf1b0dfcfeb47110b6","modelId":"kimi-k2.7-code","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.454304615216714,"warning":null},{"id":"4f813371d1d4f13c92b30639","modelId":"kimi-k2.7-code","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":10,"warning":null},{"id":"83fe42da8f302edfdb023f08","modelId":"kimi-k2.7-code","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.226797970754994,"warning":null},{"id":"5f4cc4e8c65e4c951ef26ba1","modelId":"kimi-k2.7-code","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.200000000000001,"warning":null},{"id":"37ded04223768e3023ce7517","modelId":"kimi-k2.7-code","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":1037.51,"warning":null},{"id":"3e669e225a585809c8dd4fc9","modelId":"kimi-k2.7-code","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":89.5959595959596,"warning":null},{"id":"5a1a5b9db2d6520bd2afa2e7","modelId":"kimi-k2.7-code","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.032437442075995,"warning":null},{"id":"057764867b81c8cbddb8e80c","modelId":"kimi-k2.7-code","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.1292517006803,"warning":null},{"id":"88eae31082764b6f63a7a0fd","modelId":"kimi-k2.7-code","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.1111111111111,"warning":null},{"id":"b0a4a7295832157df8e2a492","modelId":"kimi-k2.7-code","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":-10.2166666666667,"warning":null},{"id":"1f6e336dd0093408d3178e8c","modelId":"kimi-k2.7-code","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.56666666666667,"warning":null},{"id":"fda338029ff27052839f5dbc","modelId":"kimi-k2.7-code","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.8009259259259,"warning":null},{"id":"27d83ba961ef5fc71e6a27ca","modelId":"kimi-k2.7-code","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":90.0584795321637,"warning":null},{"id":"fe217e24de0bc7151b14e323","modelId":"kimi-k2.7-code","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.2061855670103,"warning":null},{"id":"c29648182b4f7975f2c9f0aa","modelId":"kimi-k2.7-code","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.6969696969697,"warning":null},{"id":"472f377c83d4e3dbb7ca25a7","modelId":"kimi-k2.7-code","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.4157303370786,"warning":null},{"id":"40ef2cb1ea3f3e136a7e3b3d","modelId":"kimi-k2.7-code","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/kimi-k2-7-code; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.01010101010101,"warning":null},{"id":"f56f16daf68bd4553392a870","modelId":"kimi-k3","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":38.75,"warning":null},{"id":"86381f7db3e6d1a28ce24624","modelId":"kimi-k3","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":985.76,"warning":null},{"id":"85b0e544c5dc4ef03bfcd15e","modelId":"kimi-k3","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":1499.8,"warning":null},{"id":"b845ca7f716cab02a8a3c794","modelId":"kimi-k3","benchmarkId":"aa-apex-agents-aa","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":41.2979351032448,"warning":null},{"id":"f1ea4b0d7743f9448563ba87","modelId":"kimi-k3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":79.3333333333333,"warning":null},{"id":"b2c605c1c811e4babb220053","modelId":"kimi-k3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":88.6666666666667,"warning":null},{"id":"39b71eadf5b656f0c8895160","modelId":"kimi-k3","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":49.2568024921068,"warning":null},{"id":"fba1004fd477e7bf03c01dbc","modelId":"kimi-k3","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":58.272977800152034,"warning":null},{"id":"edc66784be2fd9021d9d36b2","modelId":"kimi-k3","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":3.14285714285714,"warning":null},{"id":"3bd2eb52b362bec0e2106fdc","modelId":"kimi-k3","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":23.4285714285714,"warning":null},{"id":"043f2f34c6177e1c408aadc2","modelId":"kimi-k3","benchmarkId":"aa-enterprise-ops-gym-aa","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":45.329752312742464,"warning":null},{"id":"7d6657b2fa3ce606caedef83","modelId":"kimi-k3","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":18.6,"warning":null},{"id":"b81f8db18bc55b2f8f818c77","modelId":"kimi-k3","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":22,"warning":null},{"id":"dc3f37475048deaddb53b23f","modelId":"kimi-k3","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":1125.75,"warning":null},{"id":"a0a71ed92b3f1fed456d416b","modelId":"kimi-k3","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":1533.13,"warning":null},{"id":"a428750cf404da911fd1200c","modelId":"kimi-k3","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":84.24242424242419,"warning":null},{"id":"46b1f3758795f88ce1bce474","modelId":"kimi-k3","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":93.53535353535351,"warning":null},{"id":"8a498215ff5d732cdca8d41a","modelId":"kimi-k3","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":5.277777777777778,"warning":null},{"id":"6934be34afb6617ec8138359","modelId":"kimi-k3","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":24.976830398517098,"warning":null},{"id":"14c8da0b1788cd57c8bdfd0d","modelId":"kimi-k3","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":46.8952734012975,"warning":null},{"id":"f569aab802c5a421aa37653c","modelId":"kimi-k3","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":47.6930320150659,"warning":null},{"id":"ac5f0b18f22c569bae306df0","modelId":"kimi-k3","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":38.3333333333333,"warning":null},{"id":"ad5c16749d056c1bf80c4256","modelId":"kimi-k3","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":78.4971098265896,"warning":null},{"id":"78bcca2c63603190742c4b20","modelId":"kimi-k3","benchmarkId":"aa-mmmu-pro","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":80.52023121387279,"warning":null},{"id":"f791f4ab4c535b3db3488fc3","modelId":"kimi-k3","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":3.91666666666667,"warning":null},{"id":"78944636973ac08414ad4917","modelId":"kimi-k3","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":19.7,"warning":null},{"id":"639926b044b3f53d576de1f9","modelId":"kimi-k3","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":45.75,"warning":null},{"id":"86cdec56d6c7e255c7cdf06c","modelId":"kimi-k3","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":47.583333333333336,"warning":null},{"id":"9b559a9494066b39281413ff","modelId":"kimi-k3","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":52.662037037037,"warning":null},{"id":"a527b4e4b99a93b06905f9d1","modelId":"kimi-k3","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":59.4907407407407,"warning":null},{"id":"e0c8248e09af81341b67fa65","modelId":"kimi-k3","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":41.6494845360825,"warning":null},{"id":"137868d280739ea5d207e36a","modelId":"kimi-k3","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":45.979381443299,"warning":null},{"id":"755feb1cb8470edd0b42b167","modelId":"kimi-k3","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":82.3970037453184,"warning":null},{"id":"0f099a2396fb96b0113f9c58","modelId":"kimi-k3","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":85.0187265917603,"warning":null},{"id":"dda2e494f1460938614b91da","modelId":"kimi-k3","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Model source: https://artificialanalysis.ai/models/kimi-k3-low","score":12.6262626262626,"warning":null},{"id":"f50da41bcd7b9a9ec73038d3","modelId":"kimi-k3","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Model source: https://artificialanalysis.ai/models/kimi-k3","score":12.6262626262626,"warning":null},{"id":"98a1e7f8c4f539a73b1ff25c","modelId":"kimi-k3","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"Kimi K3 (High)","configuration":"Kimi K3 (High); Source model ID: moonshot-kimi-k3-high; 120 tasks","score":55.00000000000001,"warning":null},{"id":"b4f3231c817b7f55bb013735","modelId":"kimi-k3","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Kimi K3 (Low)","configuration":"Kimi K3 (Low); Source model ID: moonshot-kimi-k3-low; 120 tasks","score":12.36111111111111,"warning":null},{"id":"e82af3e6f3c2533aff0ab0d7","modelId":"kimi-k3","benchmarkId":"arc-v2semiprivate","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Source model ID: moonshot-kimi-k3-max; 120 tasks","score":60.416666666666664,"warning":null},{"id":"50e28ce7104ff5613d31fa3a","modelId":"kimi-k3","benchmarkId":"deepswe-minisweagent","level":"max","sourceLabel":"kimi-k3","configuration":"kimi-k3; Reported effort: max; 113 unique tasks; 451 attempts; 4 runs","score":68.5,"warning":null},{"id":"1187d83dfc7e0cf1c3cb0f1f","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.3333333333333,"warning":null},{"id":"a28e1bf360f6451fb459a28c","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.000000000000004,"warning":null},{"id":"fce6dfd0e81f07cb9544ac4d","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.2121212121212,"warning":null},{"id":"6f77d8d40ab545d0bf411f78","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.45597775718258,"warning":null},{"id":"dd3256ca33a1679f4f21fe4e","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.095238095238102,"warning":null},{"id":"c35ede4aac100ec236173460","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.7777777777778,"warning":null},{"id":"73b3ca32c6c854d2f898cfc4","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.502327127659605,"warning":null},{"id":"c774e01e9fd4a377f1ad3499","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"4e7fa6459cf119043d1b3938","modelId":"kimi-linear-48b-a3b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Kimi Linear 48B A3B Instruct","configuration":"Kimi Linear 48B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/kimi-linear-48b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.363636363636399,"warning":null},{"id":"f0f2adc43ec65d85b8f45fd5","modelId":"labs-devstral-small-2512","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":495.06,"warning":null},{"id":"4f0df62b80b8c9fb83b0a81e","modelId":"labs-devstral-small-2512","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.3333333333333,"warning":null},{"id":"277826d88691ec9eb915047f","modelId":"labs-devstral-small-2512","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.000000000000004,"warning":null},{"id":"f649238deb9d8f1ed9c39912","modelId":"labs-devstral-small-2512","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.0901842411547693,"warning":null},{"id":"9ccc3445c3156458123a9871","modelId":"labs-devstral-small-2512","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"0e7d98632bdb7b424068a7d8","modelId":"labs-devstral-small-2512","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":1,"warning":null},{"id":"cbf0e8e4ffcacef38790eba5","modelId":"labs-devstral-small-2512","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":528.46,"warning":null},{"id":"5dd89b6fd04dc8216937a0a5","modelId":"labs-devstral-small-2512","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.21875,"warning":null},{"id":"fdb350e52c16cc800f8fdb1a","modelId":"labs-devstral-small-2512","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.232323232323196,"warning":null},{"id":"0eb124052c0750dd77638fa8","modelId":"labs-devstral-small-2512","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.47544022242817,"warning":null},{"id":"5f3a8b308a46b3d8691e4eb2","modelId":"labs-devstral-small-2512","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.156462585034,"warning":null},{"id":"f58575e3393fbefc53600c5f","modelId":"labs-devstral-small-2512","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.814814814814795,"warning":null},{"id":"81f793d6f2f88bd3529fa166","modelId":"labs-devstral-small-2512","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.7775930851064,"warning":null},{"id":"390f00a5572b8173cd2b52e2","modelId":"labs-devstral-small-2512","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.6242774566474,"warning":null},{"id":"2c975ec825cce37abe4db74e","modelId":"labs-devstral-small-2512","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":-57.7,"warning":null},{"id":"589703a977c33adb0239d69f","modelId":"labs-devstral-small-2512","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.8,"warning":null},{"id":"3d79254e334d06a2ca47d4ed","modelId":"labs-devstral-small-2512","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.407407407407405,"warning":null},{"id":"f52fe8389f1bb8f3ded3aa92","modelId":"labs-devstral-small-2512","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.3918128654971,"warning":null},{"id":"6c90b87546d259a25837a5c2","modelId":"labs-devstral-small-2512","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.7216494845361,"warning":null},{"id":"d988ff2eb76e6caf21e5e2a4","modelId":"labs-devstral-small-2512","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.6666666666667,"warning":null},{"id":"f111753b1bee273c77e6e9e0","modelId":"labs-devstral-small-2512","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.588014981273396,"warning":null},{"id":"0dac65a9f62774845dbe0cc7","modelId":"labs-devstral-small-2512","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Devstral Small 2","configuration":"Devstral Small 2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/devstral-small-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"77dc3e409687fdc4123e7382","modelId":"llama-4-maverick","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"faefecf4982a3c56261456c3","modelId":"llama-4-maverick","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.3333333333333,"warning":null},{"id":"31f2f561fab689277fa32daf","modelId":"llama-4-maverick","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":50,"warning":null},{"id":"cdf57f3e51751d8b1b206f3e","modelId":"llama-4-maverick","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.2121918776569101,"warning":null},{"id":"45aff2671367df91ec82a00d","modelId":"llama-4-maverick","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"a1c45db0f194569fedcfab7e","modelId":"llama-4-maverick","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.8,"warning":null},{"id":"68c9b4b591a3e88b0c3b271e","modelId":"llama-4-maverick","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":-274.88,"warning":null},{"id":"48fc8cf05f6cf5be1916a16f","modelId":"llama-4-maverick","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.46354166666667,"warning":null},{"id":"7b3e6d11f2c08cecf4bbdf7b","modelId":"llama-4-maverick","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.0707070707071,"warning":null},{"id":"0faf4ea29ec7471f2ddc2575","modelId":"llama-4-maverick","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.91195551436515,"warning":null},{"id":"0496ddc71cb780c146897871","modelId":"llama-4-maverick","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.9931972789116,"warning":null},{"id":"cc29361336c4897fbe3e986d","modelId":"llama-4-maverick","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.682539682539705,"warning":null},{"id":"39a00cb8aa14d4a8f1630392","modelId":"llama-4-maverick","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.8666666666667,"warning":null},{"id":"83ab2bc25f0bf5b24ca914e2","modelId":"llama-4-maverick","benchmarkId":"aa-mlcr-aa","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.2222222222222197,"warning":null},{"id":"29bc8ec0528066576e43067c","modelId":"llama-4-maverick","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.8510638297872,"warning":null},{"id":"9aa0ba713cbccb8900465da5","modelId":"llama-4-maverick","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":62.1387283236994,"warning":null},{"id":"7eec6f4e726f2ac2843e4d31","modelId":"llama-4-maverick","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":-41.85,"warning":null},{"id":"f6678979dffdce85c6baa1a4","modelId":"llama-4-maverick","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.9,"warning":null},{"id":"6d6139083fe99c0d47d89705","modelId":"llama-4-maverick","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.712962962962997,"warning":null},{"id":"d78fc2575a1485830285d7e8","modelId":"llama-4-maverick","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.8362573099415,"warning":null},{"id":"dac821c6aa9136a83d875bda","modelId":"llama-4-maverick","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.71134020618557,"warning":null},{"id":"604481ef9f14e63577ee5fd8","modelId":"llama-4-maverick","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.81818181818182,"warning":null},{"id":"354eb159671e8738012a81fa","modelId":"llama-4-maverick","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.86516853932584,"warning":null},{"id":"edfcc29c0deb58a191bb162d","modelId":"llama-4-maverick","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Llama 4 Maverick","configuration":"Llama 4 Maverick; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-maverick; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"555d23d461c05d28e5affaad","modelId":"llama-4-scout","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"4885145006b9e6c516d4941f","modelId":"llama-4-scout","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.000000000000002,"warning":null},{"id":"7581f3a2422117f43b197e4b","modelId":"llama-4-scout","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.700000000000003,"warning":null},{"id":"9353269ebd55ac03ed176358","modelId":"llama-4-scout","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.19111706206154658,"warning":null},{"id":"afc01f597c0079772acb9e3f","modelId":"llama-4-scout","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"d524b4da994ccf7a3136f46d","modelId":"llama-4-scout","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"14409d0eb333a7e9cca62704","modelId":"llama-4-scout","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":-162.86,"warning":null},{"id":"66c9c26050db4c0d8d41b4a8","modelId":"llama-4-scout","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.140625,"warning":null},{"id":"37634d64a31c01e3c55ac3eb","modelId":"llama-4-scout","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.6868686868687,"warning":null},{"id":"ab3e6dc6aee996774804be32","modelId":"llama-4-scout","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.7800000000000002,"warning":null},{"id":"9b5c1e36e87698c0a87d0094","modelId":"llama-4-scout","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.5238095238095,"warning":null},{"id":"e7c19c636a0f224e332b9c1e","modelId":"llama-4-scout","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.9470899470899,"warning":null},{"id":"ebf48949184410fed6f20bb0","modelId":"llama-4-scout","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.39999999999999,"warning":null},{"id":"f2341b5314bd44225e9253b3","modelId":"llama-4-scout","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.1745345744681,"warning":null},{"id":"342b5de48406cc12a0d640cc","modelId":"llama-4-scout","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.9479768786127,"warning":null},{"id":"825b3abf9640fc95590a58b5","modelId":"llama-4-scout","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":-52.15,"warning":null},{"id":"29b51ff314d8015f16d8fd73","modelId":"llama-4-scout","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.166666666666668,"warning":null},{"id":"b3e565fe11b60b81a60483f2","modelId":"llama-4-scout","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.296296296296298,"warning":null},{"id":"6ceeb86bffcbd0f7f1c1cb44","modelId":"llama-4-scout","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.4970760233918,"warning":null},{"id":"b7075f4c91bd36c49bbb5930","modelId":"llama-4-scout","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.29896907216495,"warning":null},{"id":"fe0a313c1be6171f1e933267","modelId":"llama-4-scout","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.51515151515152,"warning":null},{"id":"c7b21db2c6c9de33ab88deb3","modelId":"llama-4-scout","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.7453183520599302,"warning":null},{"id":"12f09ed4b289733513c29184","modelId":"llama-4-scout","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Llama 4 Scout","configuration":"Llama 4 Scout; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/llama-4-scout; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f863116be3067c64dc4ded17","modelId":"magistral-medium-2509","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":82,"warning":null},{"id":"96fb815ee59686aa6570879b","modelId":"magistral-medium-2509","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":53,"warning":null},{"id":"27e70b3fb728057a74df17b6","modelId":"magistral-medium-2509","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"c5b956a4fc44bb9688d00821","modelId":"magistral-medium-2509","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":168.69,"warning":null},{"id":"dbfb37115a2ce93f137abe06","modelId":"magistral-medium-2509","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.56770833333333,"warning":null},{"id":"0ed549eecac4ede999a12daa","modelId":"magistral-medium-2509","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.93939393939391,"warning":null},{"id":"37b4571eaeb976c6fe763a9d","modelId":"magistral-medium-2509","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.2873030583874,"warning":null},{"id":"482398fa7f264cd6c99cd143","modelId":"magistral-medium-2509","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.9931972789116,"warning":null},{"id":"ad7ae7e1f65c748e13353d8d","modelId":"magistral-medium-2509","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.026455026455,"warning":null},{"id":"00d28563b1af109ad8f1498d","modelId":"magistral-medium-2509","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.50764627659571,"warning":null},{"id":"f2ea0a245a0ae60d6211be17","modelId":"magistral-medium-2509","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":59.6531791907514,"warning":null},{"id":"b059b84e2446c00ab8d4580f","modelId":"magistral-medium-2509","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":-26.7333333333333,"warning":null},{"id":"07fa28163f5b1f3edc355eed","modelId":"magistral-medium-2509","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.783333333333335,"warning":null},{"id":"40f63a5a594c4ddbf8bb6e60","modelId":"magistral-medium-2509","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.046783625730995,"warning":null},{"id":"86a26110045ab8144e18668e","modelId":"magistral-medium-2509","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.56701030927835,"warning":null},{"id":"f2831b29b79d4466446b9ca7","modelId":"magistral-medium-2509","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.8787878787879,"warning":null},{"id":"f9230a89b29d00bd920b1d0e","modelId":"magistral-medium-2509","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Magistral Medium 1.2","configuration":"Magistral Medium 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-medium-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.3595505617978,"warning":null},{"id":"9557a6f4edfe4134aff16297","modelId":"magistral-small-2509","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.3333333333333,"warning":null},{"id":"754e92a07c653f668b9fd678","modelId":"magistral-small-2509","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.3333333333333,"warning":null},{"id":"b75f74d8527b381f95e1cc00","modelId":"magistral-small-2509","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.19111706206154658,"warning":null},{"id":"7f110124805bb3a2e19625a9","modelId":"magistral-small-2509","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"4256531e4ff2f7ab993d1cba","modelId":"magistral-small-2509","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.49,"warning":null},{"id":"7332d47df21579cbd347434c","modelId":"magistral-small-2509","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.3125,"warning":null},{"id":"d912f20d1c509a3f0d54f0b2","modelId":"magistral-small-2509","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.2626262626263,"warning":null},{"id":"21bd6e25c91e8aaf59a9c346","modelId":"magistral-small-2509","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.44114921223355,"warning":null},{"id":"fac1628ad38a1ae32b1a569b","modelId":"magistral-small-2509","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.353741496598595,"warning":null},{"id":"42ee1aa5bdd7a63aa6435d27","modelId":"magistral-small-2509","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.27513227513231,"warning":null},{"id":"da019b5f437710abfa392ded","modelId":"magistral-small-2509","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.845079787234,"warning":null},{"id":"6e4257fe0cfd1b51a80412dd","modelId":"magistral-small-2509","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.4913294797688,"warning":null},{"id":"f6035266fb17e74a06d13c9a","modelId":"magistral-small-2509","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":-65.1,"warning":null},{"id":"f107c4759c0944a0c47cc731","modelId":"magistral-small-2509","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.450000000000001,"warning":null},{"id":"d21b8957d999afa9c21c52a4","modelId":"magistral-small-2509","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.7777777777778,"warning":null},{"id":"c7a6909b11ed59de7be2f06c","modelId":"magistral-small-2509","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.5360824742267996,"warning":null},{"id":"582e8eaacbf3fc4faf540335","modelId":"magistral-small-2509","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.545454545454549,"warning":null},{"id":"0f4ae3bfcadfaec9cecbf582","modelId":"magistral-small-2509","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Magistral Small 1.2","configuration":"Magistral Small 1.2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/magistral-small-2509; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.49438202247191,"warning":null},{"id":"0548c4d5205cb671c0f67853","modelId":"mimo-v2-flash","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":67.6666666666667,"warning":null},{"id":"18f2737a5c5e1f7f70b602b1","modelId":"mimo-v2-flash","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":96.3333333333333,"warning":null},{"id":"d0737a6dbd590e665e154bd5","modelId":"mimo-v2-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":36.3333333333333,"warning":null},{"id":"18761c742f092bb12f9d5589","modelId":"mimo-v2-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":70.6666666666667,"warning":null},{"id":"2c6d86385f0499ac90db9b10","modelId":"mimo-v2-flash","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":0,"warning":null},{"id":"b30d2768f6110084619ea391","modelId":"mimo-v2-flash","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":4.28571428571429,"warning":null},{"id":"d52c58fc44d27cc31a818289","modelId":"mimo-v2-flash","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":628.18,"warning":null},{"id":"0429ec3293c63221653ac24b","modelId":"mimo-v2-flash","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":62.635416666666664,"warning":null},{"id":"bf092d52c6f3a83f57432b09","modelId":"mimo-v2-flash","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":84.43750000000001,"warning":null},{"id":"d675a347ba1cc300cdd50958","modelId":"mimo-v2-flash","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":65.5555555555556,"warning":null},{"id":"0773db899cdffa4e1ded8ba6","modelId":"mimo-v2-flash","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":84.64646464646471,"warning":null},{"id":"3ad9dff856d49a94cb99de1f","modelId":"mimo-v2-flash","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":8.57275254865616,"warning":null},{"id":"3920da93b773299bcb0edfce","modelId":"mimo-v2-flash","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":22.8452270620945,"warning":null},{"id":"5baca63841da31c57e5ae037","modelId":"mimo-v2-flash","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":39.9319727891157,"warning":null},{"id":"f366555ca241ca7c7571729f","modelId":"mimo-v2-flash","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":64.2176870748299,"warning":null},{"id":"136673bead80a3b356afdee4","modelId":"mimo-v2-flash","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":40.2116402116402,"warning":null},{"id":"7cb51c7248fc435c2871aebc","modelId":"mimo-v2-flash","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":86.7724867724868,"warning":null},{"id":"a1e63baef98defadff1101a6","modelId":"mimo-v2-flash","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":74.4348404255319,"warning":null},{"id":"87b8554680365a774f346ece","modelId":"mimo-v2-flash","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":84.3417553191489,"warning":null},{"id":"25c20a5aaee132cb4c3e9285","modelId":"mimo-v2-flash","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":-48.45,"warning":null},{"id":"f5425b9754a5e445a5236ede","modelId":"mimo-v2-flash","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":-45.35,"warning":null},{"id":"644bce3e0d96665ff5259654","modelId":"mimo-v2-flash","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":15.633333333333333,"warning":null},{"id":"1b6fad8193ad71cc3e2e3c6b","modelId":"mimo-v2-flash","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":24.833333333333332,"warning":null},{"id":"3cc13fc1b52d73266be2282d","modelId":"mimo-v2-flash","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":83.9181286549708,"warning":null},{"id":"9dd928416fa90717a399e327","modelId":"mimo-v2-flash","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":95.0292397660819,"warning":null},{"id":"f6a30fe9e813cbabe7dcdc98","modelId":"mimo-v2-flash","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":25.7575757575758,"warning":null},{"id":"bae7463f2643e0ed3fc2532b","modelId":"mimo-v2-flash","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiMo-V2-Flash (Reasoning)","configuration":"MiMo-V2-Flash (Reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash-reasoning","score":28.030303030303,"warning":null},{"id":"be9cfbf93f75099ff67dd311","modelId":"mimo-v2-flash","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"MiMo-V2-Flash (Non-reasoning)","configuration":"MiMo-V2-Flash (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-flash","score":61.797752808988804,"warning":null},{"id":"e3d095cc53522b1dd3374852","modelId":"mimo-v2.5","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":753.75,"warning":null},{"id":"6e83c247df799a06891d5c23","modelId":"mimo-v2.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":73,"warning":null},{"id":"2d2e86088b5f8c7b172e4017","modelId":"mimo-v2.5","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.42951233004315,"warning":null},{"id":"064950ae79afc2e1a558cfd4","modelId":"mimo-v2.5","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.7142857142857104,"warning":null},{"id":"7a6c3c90bb654755ab0227ad","modelId":"mimo-v2.5","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":4,"warning":null},{"id":"01a4feea17ae775024862426","modelId":"mimo-v2.5","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":997.19,"warning":null},{"id":"dec96f1d4a54fa1a862e15bf","modelId":"mimo-v2.5","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.94949494949499,"warning":null},{"id":"3d0236e6e49d8d8d28e01c73","modelId":"mimo-v2.5","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.201112140871196,"warning":null},{"id":"832157a7f55c7b6cd246f105","modelId":"mimo-v2.5","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.14285714285711,"warning":null},{"id":"1d587cd054a074e2950f881b","modelId":"mimo-v2.5","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.4335260115607,"warning":null},{"id":"dce9246667e76d78df4c7435","modelId":"mimo-v2.5","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":-9.83333333333333,"warning":null},{"id":"adb5cf425682f21a39d89ebd","modelId":"mimo-v2.5","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.75,"warning":null},{"id":"3ad50ec8ea74deb9634c82d7","modelId":"mimo-v2.5","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.8657407407407,"warning":null},{"id":"d41a88e61beaa1a787af3bfb","modelId":"mimo-v2.5","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":90.6432748538012,"warning":null},{"id":"75e18d8f2290eb1f93e4b341","modelId":"mimo-v2.5","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.65979381443299,"warning":null},{"id":"9e3e743e86bb15ebf4c91aa8","modelId":"mimo-v2.5","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.6666666666667,"warning":null},{"id":"8b19ccc651ba74f8d9c9cfc8","modelId":"mimo-v2.5","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.6704119850187,"warning":null},{"id":"5a2f61a9349f73f419db3844","modelId":"mimo-v2.5","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"MiMo-V2.5","configuration":"MiMo-V2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-0424; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"1a46c8ba6f28b6d8dd483d71","modelId":"mimo-v2.5-pro","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":20,"warning":null},{"id":"51ea14a8ce56bc0a99ab2afe","modelId":"mimo-v2.5-pro","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":882.79,"warning":null},{"id":"5a15382fab26c34263418de9","modelId":"mimo-v2.5-pro","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.43362831858407,"warning":null},{"id":"fc1a2332e16a8d2dccb00939","modelId":"mimo-v2.5-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":41.6666666666667,"warning":null},{"id":"3b8448d3648bbeb7263db8cb","modelId":"mimo-v2.5-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.6666666666667,"warning":null},{"id":"72594856b4e1cb93f7132e97","modelId":"mimo-v2.5-pro","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.703616508004313,"warning":null},{"id":"b0f93b30c77f797af5ca0bf9","modelId":"mimo-v2.5-pro","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":1.1428571428571401,"warning":null},{"id":"21e37732c4ff88cb6ded8c9c","modelId":"mimo-v2.5-pro","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":4,"warning":null},{"id":"d9c98f36972c329291520ffc","modelId":"mimo-v2.5-pro","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":4,"warning":null},{"id":"0ff7816ff55c62ec3f07c78f","modelId":"mimo-v2.5-pro","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":1120.77,"warning":null},{"id":"e62c663d0e43ba6605976071","modelId":"mimo-v2.5-pro","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":76.1616161616162,"warning":null},{"id":"5660ea795d814c65f26e6e63","modelId":"mimo-v2.5-pro","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.5656565656566,"warning":null},{"id":"91caea072d8597c8ac5fb15b","modelId":"mimo-v2.5-pro","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":14.7822057460612,"warning":null},{"id":"9d09c59fe08151f3f31ee5b8","modelId":"mimo-v2.5-pro","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.6811862835959,"warning":null},{"id":"cb3f9696c12ea026ad23ee03","modelId":"mimo-v2.5-pro","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":42.7210884353741,"warning":null},{"id":"ae9dba4bb979effa6758161b","modelId":"mimo-v2.5-pro","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.8639455782313,"warning":null},{"id":"abedda29820bdf8da7124506","modelId":"mimo-v2.5-pro","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.229755178907695,"warning":null},{"id":"0dfd69ba4b44c38faaa26cfe","modelId":"mimo-v2.5-pro","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.44444444444444,"warning":null},{"id":"f20dd3ebb423e6ee03f6caee","modelId":"mimo-v2.5-pro","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":-37.8,"warning":null},{"id":"3b6ed8afa3616df97c1fc388","modelId":"mimo-v2.5-pro","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.25,"warning":null},{"id":"21677ace23284fc1d6430259","modelId":"mimo-v2.5-pro","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":27.18333333333333,"warning":null},{"id":"3981340313913d06e51f5448","modelId":"mimo-v2.5-pro","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.416666666666664,"warning":null},{"id":"36e58dcca98e609ebe513916","modelId":"mimo-v2.5-pro","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.578703703703695,"warning":null},{"id":"e722cf4319818646ca6e10b5","modelId":"mimo-v2.5-pro","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":72.5146198830409,"warning":null},{"id":"4b8edb52fb345ddecfdf6923","modelId":"mimo-v2.5-pro","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":94.15204678362569,"warning":null},{"id":"ff4646abf2020b07149b5be5","modelId":"mimo-v2.5-pro","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.896907216494839,"warning":null},{"id":"9c637329ea6a800195133a96","modelId":"mimo-v2.5-pro","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"MiMo-V2.5-Pro (Non-reasoning)","configuration":"MiMo-V2.5-Pro (Non-reasoning); Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro-non-reasoning","score":35.6060606060606,"warning":null},{"id":"40087eccc17d9ac466894a0c","modelId":"mimo-v2.5-pro","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.1818181818182,"warning":null},{"id":"c6934581a308f5ee1bd361df","modelId":"mimo-v2.5-pro","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.1685393258427,"warning":null},{"id":"1ecc4ba26dcb434fcdfd844a","modelId":"mimo-v2.5-pro","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"MiMo-V2.5-Pro (Reasoning)","configuration":"MiMo-V2.5-Pro (Reasoning); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-5-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"02ec04b133a03777c28066d2","modelId":"mimo-v2.6-flash","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":1490.91,"warning":null},{"id":"8c22416805f3c4e8c3c3d8b2","modelId":"mimo-v2.6-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.3333333333333,"warning":null},{"id":"65de3de7df63204b45773586","modelId":"mimo-v2.6-flash","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.05168250285656,"warning":null},{"id":"98129755ccf2312a90a197eb","modelId":"mimo-v2.6-flash","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":12,"warning":null},{"id":"fe77cce788f23ed9c06684e0","modelId":"mimo-v2.6-flash","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":9,"warning":null},{"id":"8f542771d68c38f890ccca1a","modelId":"mimo-v2.6-flash","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":1607.51,"warning":null},{"id":"e7c18dad171d498790d9ad9c","modelId":"mimo-v2.6-flash","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.0787766450417,"warning":null},{"id":"2bdb8afd42f009de6afbda6c","modelId":"mimo-v2.6-flash","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.06358381502889,"warning":null},{"id":"94ee2962270ea8b2bb18e906","modelId":"mimo-v2.6-flash","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":-12.6833333333333,"warning":null},{"id":"d41204ec960b032b5b9392b5","modelId":"mimo-v2.6-flash","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":27,"warning":null},{"id":"534da180c485e3050ccb192c","modelId":"mimo-v2.6-flash","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.273148148148195,"warning":null},{"id":"d2b92e6a23c7b5ce87a91cbb","modelId":"mimo-v2.6-flash","benchmarkId":"aa-terminal-bench-science","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.28571428571429,"warning":null},{"id":"a607fe45bb63aa87aade863b","modelId":"mimo-v2.6-flash","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"MiMo-V2.6-Flash","configuration":"MiMo-V2.6-Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.727272727272698,"warning":null},{"id":"81a965259ac5e93c8b7fe447","modelId":"mimo-v2.6-pro","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.5,"warning":null},{"id":"b906f2e0cda7893217ad854f","modelId":"mimo-v2.6-pro","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":1514.66,"warning":null},{"id":"75f0c69a78ca1dab0de299b1","modelId":"mimo-v2.6-pro","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.3333333333333,"warning":null},{"id":"8d84a8c92adfd0afb9129449","modelId":"mimo-v2.6-pro","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.623409434190854,"warning":null},{"id":"15be65e0318e450acf345369","modelId":"mimo-v2.6-pro","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.5714285714286,"warning":null},{"id":"7775a970da4b9c26b20aad4e","modelId":"mimo-v2.6-pro","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.2,"warning":null},{"id":"e0d5039eede1621c59f4ffe3","modelId":"mimo-v2.6-pro","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":1684.34,"warning":null},{"id":"3b7b525551509be42a38bb9c","modelId":"mimo-v2.6-pro","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.351251158480096,"warning":null},{"id":"ce288b2688c645bd032f2322","modelId":"mimo-v2.6-pro","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.3333333333333,"warning":null},{"id":"5a0b4274183464617d2486cc","modelId":"mimo-v2.6-pro","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.38333333333333,"warning":null},{"id":"21ef70dcbf535f4d68255e63","modelId":"mimo-v2.6-pro","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.849999999999994,"warning":null},{"id":"d78e8317840a98a149311435","modelId":"mimo-v2.6-pro","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":60.8796296296296,"warning":null},{"id":"aafee3cc04823e20b4d3b57d","modelId":"mimo-v2.6-pro","benchmarkId":"aa-terminal-bench-science","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.71428571428571,"warning":null},{"id":"c5a5427b2cb4bfccfbeee48b","modelId":"mimo-v2.6-pro","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"MiMo-V2.6-Pro","configuration":"MiMo-V2.6-Pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mimo-v2-6-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.8484848484849,"warning":null},{"id":"87779cb61f121999771e5f46","modelId":"minicpm5-2b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":422.99,"warning":null},{"id":"b2a906d4e7d0bcb771c59190","modelId":"minicpm5-2b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":59.33333333333331,"warning":null},{"id":"eab3ca6e668837d32dbf0f84","modelId":"minicpm5-2b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.8006007862431708,"warning":null},{"id":"d6f50cbeea24cc50749a6588","modelId":"minicpm5-2b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"89fa7cef514475fe6cb85d96","modelId":"minicpm5-2b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.4000000000000001,"warning":null},{"id":"7c568bc15f8ccb1c667d6db2","modelId":"minicpm5-2b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":707.2,"warning":null},{"id":"7b22165b4cb228728b966f75","modelId":"minicpm5-2b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.2020202020202,"warning":null},{"id":"58c0e4b12f0557e3f0c56eb2","modelId":"minicpm5-2b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.943466172381841,"warning":null},{"id":"97a83445db5e3087c342c4f8","modelId":"minicpm5-2b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-11.65,"warning":null},{"id":"9da5239ef85d9ee0a317f40f","modelId":"minicpm5-2b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.416666666666666,"warning":null},{"id":"b37840fcc4887b193a7da1b9","modelId":"minicpm5-2b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.273148148148103,"warning":null},{"id":"dff041001a0b378a9dc8c1ff","modelId":"minicpm5-2b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.8247422680412,"warning":null},{"id":"0c83b7be0704c58f9f95b8b9","modelId":"minicpm5-2b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.61423220973783,"warning":null},{"id":"30eccba22a3fc77326f6a727","modelId":"minicpm5-2b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"MiniCPM5-2B","configuration":"MiniCPM5-2B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minicpm5-2b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"6a9234ac9a3f32d0ebc2abf0","modelId":"minimax-m1-40k","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.6666666666667,"warning":null},{"id":"788f2e240a00d67b9967bfe3","modelId":"minimax-m1-40k","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.1818181818182,"warning":null},{"id":"1c98715ed4f5b7aa5cacd5fb","modelId":"minimax-m1-40k","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.83132530120482,"warning":null},{"id":"c47fefff4069984b5210ed42","modelId":"minimax-m1-40k","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.2244897959184,"warning":null},{"id":"7a456ca77982a503622aef14","modelId":"minimax-m1-40k","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.71428571428571,"warning":null},{"id":"eec2171903ac55545a6ef3bb","modelId":"minimax-m1-40k","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":97.2,"warning":null},{"id":"c03e16d248d044b91a74c874","modelId":"minimax-m1-40k","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.8261303191489,"warning":null},{"id":"500f48b1cccab1db90df44b5","modelId":"minimax-m1-40k","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.5789473684211,"warning":null},{"id":"763cf25103d998af4d1f28fc","modelId":"minimax-m1-40k","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiniMax M1 40k","configuration":"MiniMax M1 40k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-40k; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.27272727272727,"warning":null},{"id":"baa673fd7f683726d8a5dda4","modelId":"minimax-m1-80k","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":61,"warning":null},{"id":"7d6db873825a7618efbd9abb","modelId":"minimax-m1-80k","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":57.6666666666667,"warning":null},{"id":"e82a3c3e3750d84732cf515e","modelId":"minimax-m1-80k","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"569994eadbf73e950041b1f9","modelId":"minimax-m1-80k","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.6969696969697,"warning":null},{"id":"1551ef2a33f01b47020ef547","modelId":"minimax-m1-80k","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.89712696941613,"warning":null},{"id":"35517fb9c2159a955b38a320","modelId":"minimax-m1-80k","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.7687074829932,"warning":null},{"id":"76e122aa7a3557f5540aa180","modelId":"minimax-m1-80k","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.1111111111111,"warning":null},{"id":"bf49f69842e743085d3b29d4","modelId":"minimax-m1-80k","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":98,"warning":null},{"id":"707035a3df53472cb7e55178","modelId":"minimax-m1-80k","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.6240026595745,"warning":null},{"id":"32652eb449f7c6cb3047e850","modelId":"minimax-m1-80k","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":-48.5,"warning":null},{"id":"7bd753242e76b19ea002e45b","modelId":"minimax-m1-80k","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.466666666666665,"warning":null},{"id":"3b570ef41663865b99bb5f67","modelId":"minimax-m1-80k","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.2105263157895,"warning":null},{"id":"70b65a26ecb30426a99bbd11","modelId":"minimax-m1-80k","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiniMax M1 80k","configuration":"MiniMax M1 80k; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m1-80k; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.03030303030303,"warning":null},{"id":"15e173c41fc9f2886eb66b86","modelId":"minimax-m2","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.3333333333333,"warning":null},{"id":"4b2a76acf855c3e0dc2d8d5b","modelId":"minimax-m2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.3333333333333,"warning":null},{"id":"6b058f286079d54d349739c0","modelId":"minimax-m2","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.857142857142857,"warning":null},{"id":"c8ceb595a568197b84bb8dcf","modelId":"minimax-m2","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.91666666666666,"warning":null},{"id":"d5a6567b3645829fc2a3454f","modelId":"minimax-m2","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.6767676767677,"warning":null},{"id":"e93f40e73566ac13cdce81f1","modelId":"minimax-m2","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.670064874884199,"warning":null},{"id":"46c99f54bc78b97ddd3fca48","modelId":"minimax-m2","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.312925170068,"warning":null},{"id":"cb9c2d9b8a2a2cbceac239dc","modelId":"minimax-m2","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.6455026455027,"warning":null},{"id":"8dffa85148103a6fbf5866f2","modelId":"minimax-m2","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.0229388297872,"warning":null},{"id":"511c96d55ac43f5883ed839a","modelId":"minimax-m2","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":-46.55,"warning":null},{"id":"a8c566f58d9598384af26fad","modelId":"minimax-m2","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.233333333333334,"warning":null},{"id":"294e3daadab33e631e294791","modelId":"minimax-m2","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.8421052631579,"warning":null},{"id":"20afbe1ed17985e55a8f213e","modelId":"minimax-m2","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiniMax-M2","configuration":"MiniMax-M2; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.7575757575758,"warning":null},{"id":"012ec764d84111d300f13dac","modelId":"minimax-m2.1","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.6666666666667,"warning":null},{"id":"415c1a5b1c6ee36b839c2b75","modelId":"minimax-m2.1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.6666666666667,"warning":null},{"id":"706205dd32d2865bdad3c837","modelId":"minimax-m2.1","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"f89ac3dfe86a385c12009ee9","modelId":"minimax-m2.1","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.00000000000001,"warning":null},{"id":"06bba500a66ed5571b158392","modelId":"minimax-m2.1","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.030303030303,"warning":null},{"id":"bac805b323fcd165819d80b9","modelId":"minimax-m2.1","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.1696014828545,"warning":null},{"id":"63fcf0b40ce80d3935457cc3","modelId":"minimax-m2.1","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.8639455782313,"warning":null},{"id":"42d6364f9e1e5af2e60567ba","modelId":"minimax-m2.1","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.95238095238099,"warning":null},{"id":"607ba4f5a3bc386b9243147d","modelId":"minimax-m2.1","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":87.4916888297872,"warning":null},{"id":"c5300b173b3b63c3bb16dbd2","modelId":"minimax-m2.1","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":-32.8666666666667,"warning":null},{"id":"63491bdbc3d101722d1e08b6","modelId":"minimax-m2.1","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.166666666666668,"warning":null},{"id":"ca2ec2abdaaeae0bda0f8bf1","modelId":"minimax-m2.1","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":85.3801169590643,"warning":null},{"id":"49baf6fa212628aa4d1c49eb","modelId":"minimax-m2.1","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiniMax-M2.1","configuration":"MiniMax-M2.1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.7878787878788,"warning":null},{"id":"fdd30d2c9b426a753f20b9be","modelId":"minimax-m2.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.3333333333333,"warning":null},{"id":"a1617509d3cbb2fd7d00a142","modelId":"minimax-m2.5","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.1428571428571401,"warning":null},{"id":"9ce6c4532cb91f00ecbfccda","modelId":"minimax-m2.5","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.23958333333333,"warning":null},{"id":"32c51132daff11e2ed08e2f7","modelId":"minimax-m2.5","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.8484848484848,"warning":null},{"id":"e5ce79cceaa2a277376cead9","modelId":"minimax-m2.5","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.5282669138091,"warning":null},{"id":"7e865e71141b2c84968e7539","modelId":"minimax-m2.5","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.6326530612245,"warning":null},{"id":"aa77f8427f177890a4cbbdd7","modelId":"minimax-m2.5","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":-38.8666666666667,"warning":null},{"id":"c0846c958c8a9391b1607a28","modelId":"minimax-m2.5","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.18333333333333,"warning":null},{"id":"d891e4daf4fb77148b30e881","modelId":"minimax-m2.5","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":95.3216374269006,"warning":null},{"id":"79f3006b495cd426a1b73144","modelId":"minimax-m2.5","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiniMax-M2.5","configuration":"MiniMax-M2.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.8484848484849,"warning":null},{"id":"321b203cb1bcd8963cb0d3f1","modelId":"minimax-m2.7","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.25,"warning":null},{"id":"31bb6816c84a9177336f1c57","modelId":"minimax-m2.7","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":718.55,"warning":null},{"id":"0a2ca8def46f136850189c8d","modelId":"minimax-m2.7","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.6194690265487,"warning":null},{"id":"fb3d618157a3f1a25a01a961","modelId":"minimax-m2.7","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.3333333333333,"warning":null},{"id":"26a788cce0f3865839b1a6a3","modelId":"minimax-m2.7","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.435290250372898,"warning":null},{"id":"794fb014d2eb798e8b2998f4","modelId":"minimax-m2.7","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.571428571428571,"warning":null},{"id":"567794a1ed9be3483b007022","modelId":"minimax-m2.7","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":6,"warning":null},{"id":"e254b9330a2ddeee1dfc96f0","modelId":"minimax-m2.7","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":1010.1,"warning":null},{"id":"f0c7870f4e72d890a6e64bb4","modelId":"minimax-m2.7","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":87.3737373737374,"warning":null},{"id":"d5472132d9679cc02891ff10","modelId":"minimax-m2.7","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.610750695088,"warning":null},{"id":"ee98703c061223da691631e2","modelId":"minimax-m2.7","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.7142857142857,"warning":null},{"id":"b5993fc570c20cd787f9e13a","modelId":"minimax-m2.7","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.459510357815404,"warning":null},{"id":"b989c5a8f886f6da83a747a2","modelId":"minimax-m2.7","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.33333333333333,"warning":null},{"id":"e4e76f6bd4e0359ded238ed2","modelId":"minimax-m2.7","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.766666666666667,"warning":null},{"id":"c4cc58c4b043d4096f9904e6","modelId":"minimax-m2.7","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.8,"warning":null},{"id":"ffed6553b2632c7198bffaf9","modelId":"minimax-m2.7","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.1157407407407,"warning":null},{"id":"11881577f9e60c3849db70ae","modelId":"minimax-m2.7","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.7953216374269,"warning":null},{"id":"30e77a7107692ba54da4230b","modelId":"minimax-m2.7","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.896907216494839,"warning":null},{"id":"787cc9d04ef43d1609bcd3bd","modelId":"minimax-m2.7","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.3939393939394,"warning":null},{"id":"f8ae42068ed742b6900a2f63","modelId":"minimax-m2.7","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.4307116104869,"warning":null},{"id":"eee673418f1fbd3067f5fc74","modelId":"minimax-m2.7","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"MiniMax-M2.7","configuration":"MiniMax-M2.7; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m2-7; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"1b190b8dbfc16399d02b9178","modelId":"minimax-m3","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":10,"warning":null},{"id":"111682105ee966bede64fbaa","modelId":"minimax-m3","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1090.37,"warning":null},{"id":"c2e74f20e737aa48dad4c020","modelId":"minimax-m3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":83,"warning":null},{"id":"248b861b8fed37090c1f8c0b","modelId":"minimax-m3","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.25125760074941,"warning":null},{"id":"1f198b5e25c25ed5607f80d6","modelId":"minimax-m3","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.7142857142857104,"warning":null},{"id":"951a4a9311deecbd1ebfeba8","modelId":"minimax-m3","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.10981796478663,"warning":null},{"id":"658dddbeb79c99f8170b45c6","modelId":"minimax-m3","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.8,"warning":null},{"id":"357efeb942fca6d8756ee9fc","modelId":"minimax-m3","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1242.47,"warning":null},{"id":"256f9459d7bae8c1220d17aa","modelId":"minimax-m3","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":92.9292929292929,"warning":null},{"id":"6fda615c4d86b98b9bcce13a","modelId":"minimax-m3","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"e81c6a6abc1d601d7c3cc6e5","modelId":"minimax-m3","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.9712696941613,"warning":null},{"id":"1215dcb06b2956ad29b293d1","modelId":"minimax-m3","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.85714285714289,"warning":null},{"id":"b71623bbe996f6dabd9c34c8","modelId":"minimax-m3","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.2222222222222,"warning":null},{"id":"309e3a0e336f395747e154e9","modelId":"minimax-m3","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.5549132947977,"warning":null},{"id":"4fe2dbb70cb706ac86051368","modelId":"minimax-m3","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.35,"warning":null},{"id":"c8a6c5af1be278d8485103a7","modelId":"minimax-m3","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.7,"warning":null},{"id":"ca3f36ad153ea45bc8abefbe","modelId":"minimax-m3","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.1064814814815,"warning":null},{"id":"dfa5db0c68d02042c0e75111","modelId":"minimax-m3","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.8888888888889,"warning":null},{"id":"5ec3544112e3145637229862","modelId":"minimax-m3","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.2577319587629,"warning":null},{"id":"220c4cbb8ac4b35ba8a8b0d1","modelId":"minimax-m3","benchmarkId":"aa-terminal-bench-science","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.476190476190476,"warning":null},{"id":"28e428e681b3d40c5d5bd4b7","modelId":"minimax-m3","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.4242424242424,"warning":null},{"id":"2daf109e7eda8d8d1b90de72","modelId":"minimax-m3","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.1685393258427,"warning":null},{"id":"ceeed623352a3b0afbc75255","modelId":"minimax-m3","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"MiniMax-M3","configuration":"MiniMax-M3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/minimax-m3; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.02020202020202,"warning":null},{"id":"35c7a7125cb6c529790a7d4f","modelId":"ministral-3-14b","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":131.09,"warning":null},{"id":"42349f526e6d0332bf405daf","modelId":"ministral-3-14b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":30,"warning":null},{"id":"a0c8321b9301b9f1798b9cfc","modelId":"ministral-3-14b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.333333333333297,"warning":null},{"id":"0070a33608cc4dff46bd33ae","modelId":"ministral-3-14b","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.5512132673736124,"warning":null},{"id":"846b8d7e1064ca9b051085c7","modelId":"ministral-3-14b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"cda2910501e76db2d2736f56","modelId":"ministral-3-14b","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.2,"warning":null},{"id":"11b41c650ac68e8872edf26a","modelId":"ministral-3-14b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":247.72,"warning":null},{"id":"6a3eca23d8d091c16f3cfd1c","modelId":"ministral-3-14b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.32291666666667,"warning":null},{"id":"6b2250ddf133129bcae06578","modelId":"ministral-3-14b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":57.1717171717172,"warning":null},{"id":"587f9758a421063bc87f1bd3","modelId":"ministral-3-14b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.6339202965709,"warning":null},{"id":"8e84b419d65d216d6cbfa5af","modelId":"ministral-3-14b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.0408163265306,"warning":null},{"id":"5fb5342b1e2722933a4b16a0","modelId":"ministral-3-14b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.1322751322751,"warning":null},{"id":"f87c3cd084923ae9b2c7f4ce","modelId":"ministral-3-14b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.3234707446808,"warning":null},{"id":"0e5192b6eba15650e9b5f30b","modelId":"ministral-3-14b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.8265895953757,"warning":null},{"id":"7d6441354088c9ffdd6ef7a3","modelId":"ministral-3-14b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-66.3666666666667,"warning":null},{"id":"8fea1cc0e155ab32e6483df1","modelId":"ministral-3-14b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.583333333333334,"warning":null},{"id":"41e2bdc01aaf88a7862f9af3","modelId":"ministral-3-14b","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.842592592592602,"warning":null},{"id":"97bfb615237aae3a3c921741","modelId":"ministral-3-14b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.192982456140403,"warning":null},{"id":"a7b6b01c635f177b41f7ae72","modelId":"ministral-3-14b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.5979381443299,"warning":null},{"id":"a19663d7c4239500d8bf6258","modelId":"ministral-3-14b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.545454545454549,"warning":null},{"id":"c19f710109a638e2159d0164","modelId":"ministral-3-14b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.73782771535581,"warning":null},{"id":"36a15daaabc73dc6d5977ead","modelId":"ministral-3-14b","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Ministral 3 14B","configuration":"Ministral 3 14B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-14b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"5be0abee5a900841a5329430","modelId":"ministral-3-3b","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"9a944bda300cc996ed6055a5","modelId":"ministral-3-3b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":22,"warning":null},{"id":"377e02abab9d9b855350cd51","modelId":"ministral-3-3b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":17,"warning":null},{"id":"843d9648b3e5e036b0acb2be","modelId":"ministral-3-3b","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.6658657772309807,"warning":null},{"id":"8ed0c4631ac65f1f25e3f7ca","modelId":"ministral-3-3b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"cb5b018f13832c2363fbacae","modelId":"ministral-3-3b","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"630c9aaa4743c002a57b43d6","modelId":"ministral-3-3b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.62,"warning":null},{"id":"b537260129d057deead47b8d","modelId":"ministral-3-3b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":56.72916666666667,"warning":null},{"id":"58e5a9f2cd204be72403efef","modelId":"ministral-3-3b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.7575757575758,"warning":null},{"id":"d1a4aca9ced7ddd73ed98e8c","modelId":"ministral-3-3b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.37534754402224,"warning":null},{"id":"0bf69fe4d67896af5c1f65c8","modelId":"ministral-3-3b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.802721088435398,"warning":null},{"id":"5bcb8f2f21bc3ebaf7da173c","modelId":"ministral-3-3b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.6560846560847,"warning":null},{"id":"283f4b4a8c76f509c6d422e7","modelId":"ministral-3-3b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.35206117021281,"warning":null},{"id":"3b5a9373617c66fda5b546f0","modelId":"ministral-3-3b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.092485549132896,"warning":null},{"id":"928da08d1319107c56189f07","modelId":"ministral-3-3b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-64,"warning":null},{"id":"7a796ce9f7d23b28986ff5e8","modelId":"ministral-3-3b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.983333333333334,"warning":null},{"id":"719ae7fb0aa3a8f58828d8ad","modelId":"ministral-3-3b","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.277777777777802,"warning":null},{"id":"83db4f32b2b311f3f83c14d5","modelId":"ministral-3-3b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.8538011695906,"warning":null},{"id":"c9c4e289cdf58ffe5ffa14b4","modelId":"ministral-3-3b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.74226804123711,"warning":null},{"id":"4f010862ca6747cc6d8401ce","modelId":"ministral-3-3b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"77514ba2a733e772619ec66b","modelId":"ministral-3-3b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"7009edb8b996cc961ae08907","modelId":"ministral-3-3b","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Ministral 3 3B","configuration":"Ministral 3 3B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f31efa2ad660815a88149738","modelId":"ministral-3-8b","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.44,"warning":null},{"id":"3dd4f3c338791cdbb4285fe1","modelId":"ministral-3-8b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.6666666666667,"warning":null},{"id":"6007b52d64bc356c474923d7","modelId":"ministral-3-8b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.6666666666667,"warning":null},{"id":"605bc0d7c72327162c223151","modelId":"ministral-3-8b","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.516749466666553,"warning":null},{"id":"a8ad5cdcd0e1929baa7929f2","modelId":"ministral-3-8b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"dea8abc42d6581d240a14318","modelId":"ministral-3-8b","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.8,"warning":null},{"id":"3af1b66cf28b459b2943e79d","modelId":"ministral-3-8b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":214.05,"warning":null},{"id":"7183c2052a20887069af7aa7","modelId":"ministral-3-8b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.32291666666667,"warning":null},{"id":"41188e61e8f63338df1815bf","modelId":"ministral-3-8b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.0707070707071,"warning":null},{"id":"b6bf00c2a753272d88d86510","modelId":"ministral-3-8b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.2632066728452305,"warning":null},{"id":"05a6e30b5856ecc160d17ae0","modelId":"ministral-3-8b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.115646258503403,"warning":null},{"id":"e7c5d463ec1b8d046eeeffef","modelId":"ministral-3-8b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.264550264550298,"warning":null},{"id":"e640a3de368a7ff58a8b9410","modelId":"ministral-3-8b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.1539228723404,"warning":null},{"id":"faad5ca0d804e52efc4f202b","modelId":"ministral-3-8b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":45.9537572254335,"warning":null},{"id":"75e22a9892b66c0e65dfd55d","modelId":"ministral-3-8b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-68.95,"warning":null},{"id":"edcff04c2243d437ad690a4d","modelId":"ministral-3-8b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.983333333333333,"warning":null},{"id":"60d1bb31feedab5df58faad3","modelId":"ministral-3-8b","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.717592592592602,"warning":null},{"id":"43a3d3fe9aa6ddda6930bd0e","modelId":"ministral-3-8b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.6081871345029,"warning":null},{"id":"4d95912317d4bb5d72f9b21d","modelId":"ministral-3-8b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.71134020618557,"warning":null},{"id":"d5d3b3be768714fdf63001cf","modelId":"ministral-3-8b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.545454545454549,"warning":null},{"id":"0c7d5dd66f6c30384d34f3fd","modelId":"ministral-3-8b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.11985018726592,"warning":null},{"id":"345253b8f88be2712721eb6a","modelId":"ministral-3-8b","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Ministral 3 8B","configuration":"Ministral 3 8B; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/ministral-3-8b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"dbe2eeeed1d775de800dcb31","modelId":"mistral-large-2407","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"348a6de2555440e02838cb61","modelId":"mistral-large-2407","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":2,"warning":null},{"id":"b7a16d7e8dee63f9d55e9d63","modelId":"mistral-large-2407","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.1717171717172,"warning":null},{"id":"d63bb6149699d19daa535c35","modelId":"mistral-large-2407","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.9499999999999997,"warning":null},{"id":"28e7e03881e741b0ee543153","modelId":"mistral-large-2407","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.632653061224502,"warning":null},{"id":"b0ca3209dcb5204b311ef0d3","modelId":"mistral-large-2407","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.6666666666667,"warning":null},{"id":"381cfdc71bbd8167f239652c","modelId":"mistral-large-2407","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.39999999999999,"warning":null},{"id":"13edbb09a0d6f48740227bf4","modelId":"mistral-large-2407","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.2596409574468,"warning":null},{"id":"05576a4ab91490da5b5f7e76","modelId":"mistral-large-2407","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Mistral Large 2 (Jul '24)","configuration":"Mistral Large 2 (Jul '24); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-2407; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.0409356725146,"warning":null},{"id":"b174853ca58c57fa9ba2f879","modelId":"mistral-large-3","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":228.3,"warning":null},{"id":"2819c29ad8fdb7646a035be2","modelId":"mistral-large-3","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":38,"warning":null},{"id":"bab6a467560df59ccde7382d","modelId":"mistral-large-3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":36,"warning":null},{"id":"42b857e698d4260f5223208e","modelId":"mistral-large-3","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.4974701708193463,"warning":null},{"id":"3437f9bc1e0ac1692f5c5bda","modelId":"mistral-large-3","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"1a823311b9704e57063bf129","modelId":"mistral-large-3","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.2,"warning":null},{"id":"278ea7c2642b862adf3db1ea","modelId":"mistral-large-3","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":420.4,"warning":null},{"id":"07149ebe5716925ff1f2d43b","modelId":"mistral-large-3","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.33333333333333,"warning":null},{"id":"d565292efd326186f35512e8","modelId":"mistral-large-3","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.979797979798,"warning":null},{"id":"e982d1f0c2e679c26443a26c","modelId":"mistral-large-3","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.17052826691381,"warning":null},{"id":"b95857dee1e294d54839285f","modelId":"mistral-large-3","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.190476190476204,"warning":null},{"id":"118e84cb8fab9c3a1fb4fa3b","modelId":"mistral-large-3","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.4550264550264,"warning":null},{"id":"00443aa27838a4d61aa17081","modelId":"mistral-large-3","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.6515957446808,"warning":null},{"id":"04902275b4289db5be2e8909","modelId":"mistral-large-3","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.6647398843931,"warning":null},{"id":"fe1a4682c062f12dbe7de410","modelId":"mistral-large-3","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":-39.5666666666667,"warning":null},{"id":"fac9bcbdc0f81c8e44b0ac55","modelId":"mistral-large-3","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.95,"warning":null},{"id":"4eccbe09a60fef497d086249","modelId":"mistral-large-3","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.5740740740741,"warning":null},{"id":"55e1c75933ec7e8373537acd","modelId":"mistral-large-3","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.5614035087719,"warning":null},{"id":"9a30580079b16a6231c310bb","modelId":"mistral-large-3","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.77319587628866,"warning":null},{"id":"d5906c4f5d64bad2845778a8","modelId":"mistral-large-3","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.909090909090901,"warning":null},{"id":"f223a37edff3ec7188daa1f8","modelId":"mistral-large-3","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.9850187265918,"warning":null},{"id":"5bb842631418e84f512ae48a","modelId":"mistral-large-3","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"db5fc334551285ccedd155f3","modelId":"mistral-large-4","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":1391.97,"warning":null},{"id":"92d9702142f2af68c10b4239","modelId":"mistral-large-4","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.3333333333333,"warning":null},{"id":"4f54cdec0a23286c1583ed6a","modelId":"mistral-large-4","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":59.90136246808686,"warning":null},{"id":"6f7541c0b8b1eedd20d04b1c","modelId":"mistral-large-4","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.5714285714286,"warning":null},{"id":"0bf963a14be34961a024621d","modelId":"mistral-large-4","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.6,"warning":null},{"id":"24287cbe3e82dbd902e9bbb9","modelId":"mistral-large-4","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":1422.17,"warning":null},{"id":"3e85c09f0b5659043ce66c06","modelId":"mistral-large-4","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.0555555555555554,"warning":null},{"id":"a4bf80014cf8127d1186acee","modelId":"mistral-large-4","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.032437442075995,"warning":null},{"id":"0eb64d0a4de068b9aa6ba652","modelId":"mistral-large-4","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.4161849710983,"warning":null},{"id":"530427155dcfa01e7115c354","modelId":"mistral-large-4","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":-5.3,"warning":null},{"id":"b28eb0bf4146edef49d664a1","modelId":"mistral-large-4","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.816666666666666,"warning":null},{"id":"75d0b63baab906a80f8fef86","modelId":"mistral-large-4","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.1666666666667,"warning":null},{"id":"d3cacfd5ef49c303c25c3c00","modelId":"mistral-large-4","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Mistral Large 4 Preview","configuration":"Mistral Large 4 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-large-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.7676767676768,"warning":null},{"id":"8598ad2d6ffa40f68ee3112e","modelId":"mistral-medium-2505","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.3333333333333,"warning":null},{"id":"cd3eea1dc8c7bbb962e269d1","modelId":"mistral-medium-2505","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.3333333333333,"warning":null},{"id":"dda57f94dfcd00bf84f043f2","modelId":"mistral-medium-2505","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"5c500f8d1d9109499173dc22","modelId":"mistral-medium-2505","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":57.77777777777779,"warning":null},{"id":"8a627e4cf810decf4c58ca18","modelId":"mistral-medium-2505","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.05,"warning":null},{"id":"32fb52a7ef7e80c24bf51218","modelId":"mistral-medium-2505","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.2517006802721,"warning":null},{"id":"b7ecce4ab23d7e0b1d66c046","modelId":"mistral-medium-2505","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":40,"warning":null},{"id":"e80fbfd55ae35c4506737fb1","modelId":"mistral-medium-2505","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":90.6666666666667,"warning":null},{"id":"77cc970d5bca332420c6f124","modelId":"mistral-medium-2505","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.9724069148936,"warning":null},{"id":"366f3b0107c0572f5d49886f","modelId":"mistral-medium-2505","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.0057803468208,"warning":null},{"id":"98b8086ecdd729adb943befa","modelId":"mistral-medium-2505","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":-31.4,"warning":null},{"id":"ef1572ebbb7db6d1d5e0ea19","modelId":"mistral-medium-2505","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.333333333333332,"warning":null},{"id":"853fba57bc04b7caeed6606b","modelId":"mistral-medium-2505","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.269005847953203,"warning":null},{"id":"54ae2b40694a6eddf2507953","modelId":"mistral-medium-2505","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Mistral Medium 3","configuration":"Mistral Medium 3; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.78787878787879,"warning":null},{"id":"698b73e2f7193f10d163518e","modelId":"mistral-medium-2508","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":529.91,"warning":null},{"id":"e7008acf3eac0efcb0ed528e","modelId":"mistral-medium-2508","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.3333333333333,"warning":null},{"id":"5ec131118ff7bb227102e336","modelId":"mistral-medium-2508","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.6666666666667,"warning":null},{"id":"0060828c4561c3146a4d2fbe","modelId":"mistral-medium-2508","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.146227762526245,"warning":null},{"id":"8b3ef38d2b5d2fc9f19a3ed9","modelId":"mistral-medium-2508","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"ab90f51570bba0df1c889314","modelId":"mistral-medium-2508","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.2,"warning":null},{"id":"88996cd23c0b19475183f5ab","modelId":"mistral-medium-2508","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":385.53,"warning":null},{"id":"350c2c641e04fcb49688e650","modelId":"mistral-medium-2508","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.7878787878788,"warning":null},{"id":"ba265b790d9a002b971910c1","modelId":"mistral-medium-2508","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.68025949953661,"warning":null},{"id":"3a8b8a4a21798fb14026769d","modelId":"mistral-medium-2508","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.7959183673469,"warning":null},{"id":"53fa70129bf83f83ab6219a3","modelId":"mistral-medium-2508","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.634920634920604,"warning":null},{"id":"3f9dfe82f4892e139fabc3f4","modelId":"mistral-medium-2508","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.3261303191489,"warning":null},{"id":"dae076a0eacaa63ec8f192c9","modelId":"mistral-medium-2508","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.1618497109827,"warning":null},{"id":"98f555a90b9ec68d099b3ccf","modelId":"mistral-medium-2508","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":-46.2833333333333,"warning":null},{"id":"07b15db546d50bc3a15c79c9","modelId":"mistral-medium-2508","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.75,"warning":null},{"id":"0bc62ae3ef77f27ee9feb8c3","modelId":"mistral-medium-2508","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.407407407407405,"warning":null},{"id":"02e0fffe433bd9bbb17718ce","modelId":"mistral-medium-2508","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.643274853801195,"warning":null},{"id":"96d61f35ce89c5e402b475d3","modelId":"mistral-medium-2508","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.24742268041237,"warning":null},{"id":"bcef63b769b65103b4ee5212","modelId":"mistral-medium-2508","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.6060606060606,"warning":null},{"id":"8ff2dfac5f466e44c25543b3","modelId":"mistral-medium-2508","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.8576779026217,"warning":null},{"id":"b13b890114b7d4270d307654","modelId":"mistral-medium-2508","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Mistral Medium 3.1","configuration":"Mistral Medium 3.1; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-1; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"ddec1a46d0c8daa3d68bf673","modelId":"mistral-medium-3.5","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.5,"warning":null},{"id":"3fd61197d4971be5f002d699","modelId":"mistral-medium-3.5","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":522.97,"warning":null},{"id":"567e2e221b95563a85e802ed","modelId":"mistral-medium-3.5","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.3333333333333,"warning":null},{"id":"3104fceb9c278740e37b41b3","modelId":"mistral-medium-3.5","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.299648656993127,"warning":null},{"id":"d258144824b33f3d3324f00a","modelId":"mistral-medium-3.5","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"5df2fd6f09a89049a344da89","modelId":"mistral-medium-3.5","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.72127723067741,"warning":null},{"id":"5bdde6072a786cea090d962b","modelId":"mistral-medium-3.5","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.8000000000000003,"warning":null},{"id":"7a0edad534dfbb30af22a472","modelId":"mistral-medium-3.5","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":760.86,"warning":null},{"id":"c1df964d67d6a114f0ae15fc","modelId":"mistral-medium-3.5","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.8484848484849,"warning":null},{"id":"0e63c2ddb3ce5bb283ced808","modelId":"mistral-medium-3.5","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.7627432808156,"warning":null},{"id":"a46292af76538ed9cd457f74","modelId":"mistral-medium-3.5","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.7755102040816,"warning":null},{"id":"ed8f320f4381500052c687c9","modelId":"mistral-medium-3.5","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.66666666666667,"warning":null},{"id":"103f44c413960d3a1373d498","modelId":"mistral-medium-3.5","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.8554913294798,"warning":null},{"id":"df8c0054af3ea000e16148ed","modelId":"mistral-medium-3.5","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":-36.8,"warning":null},{"id":"fe26eeb3eee7044147e39ad9","modelId":"mistral-medium-3.5","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.666666666666668,"warning":null},{"id":"46f40618c1d6eceefad11b5e","modelId":"mistral-medium-3.5","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.162037037037,"warning":null},{"id":"b8f431b7af5ced6f9c6ffda2","modelId":"mistral-medium-3.5","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":94.15204678362569,"warning":null},{"id":"80753495b3c7881924eaa865","modelId":"mistral-medium-3.5","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.051546391752598,"warning":null},{"id":"54f2e07eae2bee6eeebe5743","modelId":"mistral-medium-3.5","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.3333333333333,"warning":null},{"id":"a1e4976e556451286cd3bdcf","modelId":"mistral-medium-3.5","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.561797752809,"warning":null},{"id":"8696355798a7473fc208ccd1","modelId":"mistral-medium-3.5","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/mistral-medium-3-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"a1ecc289157138b5f8e71437","modelId":"mistral-small-2506","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"6da1b8d1b2167564bf428fb0","modelId":"mistral-small-2506","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":27,"warning":null},{"id":"53bc5a579aa23fead108267c","modelId":"mistral-small-2506","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.3333333333333,"warning":null},{"id":"e63a9f1379393c3cc66d6071","modelId":"mistral-small-2506","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.2854023922677291,"warning":null},{"id":"821ded825d6af49adc589ec6","modelId":"mistral-small-2506","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f90922973ea706908ad9cc65","modelId":"mistral-small-2506","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"783454d1c36d690bc8de2501","modelId":"mistral-small-2506","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":-194.91,"warning":null},{"id":"c46cb1fb5b137f1522f36b9e","modelId":"mistral-small-2506","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.078125,"warning":null},{"id":"39d3bdb6484f58e20f1c50fe","modelId":"mistral-small-2506","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.5050505050505,"warning":null},{"id":"167f1a758c5db842ce1feca6","modelId":"mistral-small-2506","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.30954587581094,"warning":null},{"id":"68d2c1aeb87f2c629ac7672d","modelId":"mistral-small-2506","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.469387755102,"warning":null},{"id":"b84c0ce3c4a5433cf625b30f","modelId":"mistral-small-2506","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.513227513227502,"warning":null},{"id":"891301aed44b6f2989b8b04d","modelId":"mistral-small-2506","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.26666666666671,"warning":null},{"id":"2d02db7530f0e9d255d63470","modelId":"mistral-small-2506","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.05186170212771,"warning":null},{"id":"34bd5ae77357e239d89ce014","modelId":"mistral-small-2506","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.9768786127168,"warning":null},{"id":"be0fd3096da38b78e3aef989","modelId":"mistral-small-2506","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":-50.5833333333333,"warning":null},{"id":"27cf85547226acfd3effbbb7","modelId":"mistral-small-2506","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.833333333333334,"warning":null},{"id":"7a032bb7d27ac4f4a54dc043","modelId":"mistral-small-2506","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.587962962962997,"warning":null},{"id":"93d03cab54cbb7dfc0b435ab","modelId":"mistral-small-2506","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.5321637426901,"warning":null},{"id":"f595e0feb6f1773fe5a6fc61","modelId":"mistral-small-2506","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.18556701030928,"warning":null},{"id":"340512192be278774e77e5da","modelId":"mistral-small-2506","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.81818181818182,"warning":null},{"id":"832c778cd48cdca55f1c79c3","modelId":"mistral-small-2506","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.61797752808989,"warning":null},{"id":"6e9568251c46a4d7f2d38ad5","modelId":"mistral-small-2506","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Mistral Small 3.2","configuration":"Mistral Small 3.2; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/mistral-small-3-2; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"9188eefd377f2e9c23d9a8b6","modelId":"mistral-small-4","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":1.25,"warning":null},{"id":"7c8005ebeaffc703ae8d1204","modelId":"mistral-small-4","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":241.11,"warning":null},{"id":"213e91745b08c415820c541d","modelId":"mistral-small-4","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":28.3333333333333,"warning":null},{"id":"979d64a0eb75ed822e25c686","modelId":"mistral-small-4","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":49.6666666666667,"warning":null},{"id":"4f6658801d33d9c4bbf58519","modelId":"mistral-small-4","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":0.9635410038882158,"warning":null},{"id":"8b914ad32f7c9e5f7d244d1d","modelId":"mistral-small-4","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"bf81d819d89c15dd4fe6b89b","modelId":"mistral-small-4","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":0.28571428571428603,"warning":null},{"id":"8f94ed73fa6de47d55b25077","modelId":"mistral-small-4","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":0,"warning":null},{"id":"c643e395d5da97f1a5f85700","modelId":"mistral-small-4","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":362.6,"warning":null},{"id":"6680a21e2f5f81331071ba82","modelId":"mistral-small-4","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":57.07070707070711,"warning":null},{"id":"2f6bef1401dd3c874f04eef8","modelId":"mistral-small-4","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":76.8686868686869,"warning":null},{"id":"ae1fa7961d49d684f2981dff","modelId":"mistral-small-4","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":3.79981464318814,"warning":null},{"id":"c8e3afa3aef8179b40e448f0","modelId":"mistral-small-4","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":9.870250231696021,"warning":null},{"id":"1175374611f43823a69bb917","modelId":"mistral-small-4","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":32.7891156462585,"warning":null},{"id":"5231791d1c29e104c93e1247","modelId":"mistral-small-4","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":48.1632653061224,"warning":null},{"id":"f06380625c147eb9e14efb36","modelId":"mistral-small-4","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":46.4739884393064,"warning":null},{"id":"81f942ea40a39fe097e24d66","modelId":"mistral-small-4","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":56.8208092485549,"warning":null},{"id":"2af0431ee57da42eb9cd13d5","modelId":"mistral-small-4","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":-48.5166666666667,"warning":null},{"id":"50d33b56ae65630b7fd1e35d","modelId":"mistral-small-4","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":-30.4,"warning":null},{"id":"5bf3b46928f29139d0cfdd0e","modelId":"mistral-small-4","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":16.566666666666666,"warning":null},{"id":"301e793977c259094f7093c7","modelId":"mistral-small-4","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":21.683333333333334,"warning":null},{"id":"f2789b044f471f6a2756d4f1","modelId":"mistral-small-4","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":38.7731481481481,"warning":null},{"id":"211745cfecdb17b01d94607f","modelId":"mistral-small-4","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":18.4210526315789,"warning":null},{"id":"70cdd67007153234faacd560","modelId":"mistral-small-4","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":41.2280701754386,"warning":null},{"id":"206a3af7332c726f2d612cb0","modelId":"mistral-small-4","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":4.948453608247419,"warning":null},{"id":"5e4e6f1b78b4531611aa9a8d","modelId":"mistral-small-4","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Mistral Small 4 (Non-reasoning)","configuration":"Mistral Small 4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4-non-reasoning","score":10.6060606060606,"warning":null},{"id":"e6390ae449afa5ff6fecfd51","modelId":"mistral-small-4","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":17.4242424242424,"warning":null},{"id":"2599f4ee8ffaec96bbf2b7aa","modelId":"mistral-small-4","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":20.9737827715356,"warning":null},{"id":"f7c7c53f6cba7efa0c09fcb8","modelId":"mistral-small-4","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Mistral Small 4 (Reasoning)","configuration":"Mistral Small 4 (Reasoning); Model source: https://artificialanalysis.ai/models/mistral-small-4","score":0,"warning":null},{"id":"3353892598e7c19ca04df2ef","modelId":"muse-spark-1.1","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":850.65,"warning":null},{"id":"0f5c978186aa20919054c624","modelId":"muse-spark-1.1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":77.6666666666667,"warning":null},{"id":"8aab2cd221a6d58473cfe6cb","modelId":"muse-spark-1.1","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":38.84891624542286,"warning":null},{"id":"abcff636e38ccf97afb0ff6e","modelId":"muse-spark-1.1","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":15.1428571428571,"warning":null},{"id":"8b60281a082682104202592c","modelId":"muse-spark-1.1","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":14.399999999999999,"warning":null},{"id":"2c7275e6b1411bd10e1914c5","modelId":"muse-spark-1.1","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":1212.48,"warning":null},{"id":"0e7fa948144019ee7d8e70ad","modelId":"muse-spark-1.1","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":89.79797979797979,"warning":null},{"id":"86de7388fbdefea73e771c1b","modelId":"muse-spark-1.1","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":46.2001853568119,"warning":null},{"id":"181518e22756782a0bcc6529","modelId":"muse-spark-1.1","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":28.0833333333333,"warning":null},{"id":"74473dc50823ae0374d8dfbe","modelId":"muse-spark-1.1","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":52.05,"warning":null},{"id":"5fe09def3846dc24c2d5043a","modelId":"muse-spark-1.1","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":58.796296296296305,"warning":null},{"id":"06875d1d5d119ae76117a900","modelId":"muse-spark-1.1","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":31.752577319587598,"warning":null},{"id":"92c082ce86283ce9b497ce79","modelId":"muse-spark-1.1","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":77.9026217228464,"warning":null},{"id":"d156c443ef4093c113846a83","modelId":"muse-spark-1.1","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Xhigh)","configuration":"Muse Spark 1.1 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-1","score":6.06060606060606,"warning":null},{"id":"8fa6b5f2d5612bc7ff3ca112","modelId":"muse-spark-1.1","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"muse-spark-1-1","configuration":"muse-spark-1-1; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":53.3,"warning":null},{"id":"4db8c7b47a313e0e8d926986","modelId":"muse-spark-1.1","benchmarkId":"terminal-minisweagent","level":"xhigh","sourceLabel":"Muse Spark 1.1","configuration":"Muse Spark 1.1; Reported effort: xhigh; 445 trials; recorded 2026-07-09","score":76.18,"warning":null},{"id":"01f9bb81e93035eae908ffd2","modelId":"muse-spark-1.2","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":1328.29,"warning":null},{"id":"68bb725a7147d37f691cc776","modelId":"muse-spark-1.2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":79,"warning":null},{"id":"232d05db27565366b51b3c03","modelId":"muse-spark-1.2","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":40.61337838423964,"warning":null},{"id":"92e7edda18428b3c391082fa","modelId":"muse-spark-1.2","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":17.714285714285698,"warning":null},{"id":"59c8fb45a1d69006d9ccbbdf","modelId":"muse-spark-1.2","benchmarkId":"aa-enterprise-ops-gym-aa","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":47.26947179946285,"warning":null},{"id":"8facaaeaddc71a9cb970c3e5","modelId":"muse-spark-1.2","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":17.4,"warning":null},{"id":"8d598f26800d9569ce4fe997","modelId":"muse-spark-1.2","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":1472.97,"warning":null},{"id":"bf5e779bd5c1e5819a7725d0","modelId":"muse-spark-1.2","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":90.4040404040404,"warning":null},{"id":"5effad373b0205da782f5fd8","modelId":"muse-spark-1.2","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":45.458758109360495,"warning":null},{"id":"bd09a08eaa6b2525ba1535bd","modelId":"muse-spark-1.2","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":31.1111111111111,"warning":null},{"id":"7fc466dd87333128e87ed96e","modelId":"muse-spark-1.2","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":27.2,"warning":null},{"id":"2be3e290103f2379e2a8ce97","modelId":"muse-spark-1.2","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":45.38333333333333,"warning":null},{"id":"e7a82f88e55a398d5b725fff","modelId":"muse-spark-1.2","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":57.4074074074074,"warning":null},{"id":"a93c4241caec6baf37b38e51","modelId":"muse-spark-1.2","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":34.8453608247423,"warning":null},{"id":"b2c7712ced082f6fa4d42381","modelId":"muse-spark-1.2","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":80.1498127340824,"warning":null},{"id":"1ccd58acbbacd38e573c0695","modelId":"muse-spark-1.2","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Muse Spark 1.2 (Xhigh)","configuration":"Muse Spark 1.2 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-2","score":7.07070707070707,"warning":null},{"id":"1585a41209b88f2db63e82f8","modelId":"muse-spark-1.2","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"muse-spark-1-2","configuration":"muse-spark-1-2; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs","score":54.9,"warning":null},{"id":"b9f6f607919d776f11304a54","modelId":"muse-spark-1.3","benchmarkId":"aa-aa-analyst-agent","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":37.5,"warning":null},{"id":"8a22bc488a644b1e2508fc4e","modelId":"muse-spark-1.3","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":1581.34,"warning":null},{"id":"86532f47a6831d9874503331","modelId":"muse-spark-1.3","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":1485.51,"warning":null},{"id":"f230d5a9f5b9039b7143d623","modelId":"muse-spark-1.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":83,"warning":null},{"id":"2a6a2fba20f75c83ae00c919","modelId":"muse-spark-1.3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":83,"warning":null},{"id":"1c082cbbfe9e67d7bb984ec5","modelId":"muse-spark-1.3","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":57.853437865386134,"warning":null},{"id":"5f63a44e32cdfbe7e4663717","modelId":"muse-spark-1.3","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":56.83609003249621,"warning":null},{"id":"148730301f5957a9fe79d9f6","modelId":"muse-spark-1.3","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":24.8571428571429,"warning":null},{"id":"6f1189e171742bcde8826993","modelId":"muse-spark-1.3","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":26,"warning":null},{"id":"c23b2e2aae97c9e2aec8097b","modelId":"muse-spark-1.3","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":26.6,"warning":null},{"id":"518d72ce50dabb22d8ce510f","modelId":"muse-spark-1.3","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":24.2,"warning":null},{"id":"1171f6b2b6a56815c83aef48","modelId":"muse-spark-1.3","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":1680.99,"warning":null},{"id":"a662897f590e0f63e0111b40","modelId":"muse-spark-1.3","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":1627.25,"warning":null},{"id":"49ff66a3ac04722cbfb9f96e","modelId":"muse-spark-1.3","benchmarkId":"aa-gpqa-diamond","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":93.53535353535351,"warning":null},{"id":"ca59d7ad9a49b5deebe091fb","modelId":"muse-spark-1.3","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":94.1414141414141,"warning":null},{"id":"7a8c3d4f8b3a35c6f9de6348","modelId":"muse-spark-1.3","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":8.888888888888888,"warning":null},{"id":"cf85cb3387484c802e3812ee","modelId":"muse-spark-1.3","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":48.7025023169602,"warning":null},{"id":"ec206653481e28f9c528335f","modelId":"muse-spark-1.3","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":47.451343836886004,"warning":null},{"id":"c5c0b548ca3c5060f14f5e70","modelId":"muse-spark-1.3","benchmarkId":"aa-itbench-aa","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":33.239171374764595,"warning":null},{"id":"936456dd1867d1a4b847c4d3","modelId":"muse-spark-1.3","benchmarkId":"aa-mlcr-aa","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":43.3333333333333,"warning":null},{"id":"2162b35173af7a019770ca7a","modelId":"muse-spark-1.3","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":82.0231213872832,"warning":null},{"id":"4f219c79469e44d8cf6604a6","modelId":"muse-spark-1.3","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":25,"warning":null},{"id":"afb2cd939c6a958c7a34bc70","modelId":"muse-spark-1.3","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":23.1166666666667,"warning":null},{"id":"99de52795a9876970ef87d8e","modelId":"muse-spark-1.3","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":43.583333333333336,"warning":null},{"id":"c5701d3f4aa1d335153fb4b3","modelId":"muse-spark-1.3","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":41.516666666666666,"warning":null},{"id":"e2006099a5d0417910b49e29","modelId":"muse-spark-1.3","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":58.796296296296305,"warning":null},{"id":"2580065e526b137c06fb4d34","modelId":"muse-spark-1.3","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":59.7222222222222,"warning":null},{"id":"7a390f1067521bbb145e68dd","modelId":"muse-spark-1.3","benchmarkId":"aa-tau3-banking","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":50.51546391752581,"warning":null},{"id":"2f8296c739b4a1f204c99e8e","modelId":"muse-spark-1.3","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":47.2164948453608,"warning":null},{"id":"d63dffcd50afdb1eb351b494","modelId":"muse-spark-1.3","benchmarkId":"aa-terminal-bench-science","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":10.952380952380999,"warning":null},{"id":"f2cbe7235dce6c1db8869015","modelId":"muse-spark-1.3","benchmarkId":"aa-terminalbench-v2-1","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":84.2696629213483,"warning":null},{"id":"1dc29385569902eaa7f40057","modelId":"muse-spark-1.3","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":85.3932584269663,"warning":null},{"id":"ee74081579b2a3ebb41daa4a","modelId":"muse-spark-1.3","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Model source: https://artificialanalysis.ai/models/muse-spark-1-3","score":33.3333333333333,"warning":null},{"id":"9e7d357f563b5a73cfc2d18d","modelId":"muse-spark-1.3","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Muse Spark 1.3 (Xhigh); Model source: https://artificialanalysis.ai/models/muse-spark-1-3-xhigh","score":16.6666666666667,"warning":null},{"id":"e2f5630b078937ae5c53af35","modelId":"muse-wonder","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":1658.65,"warning":null},{"id":"409f64db18ddf3ebbbdb5c1f","modelId":"muse-wonder","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":80.6666666666667,"warning":null},{"id":"14f57fb9ba5c08e88680727f","modelId":"muse-wonder","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":68.09455131207758,"warning":null},{"id":"95e7a4ba29c6c86b27d71097","modelId":"muse-wonder","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":28.2857142857143,"warning":null},{"id":"7972ff078593589a9b53c45e","modelId":"muse-wonder","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":30.599999999999998,"warning":null},{"id":"5cd9b0ce6dbfe3dd8a4d7eed","modelId":"muse-wonder","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":1770.34,"warning":null},{"id":"62285399003445302807f39d","modelId":"muse-wonder","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":10,"warning":null},{"id":"7f1ee21fbb1b83cea7ef3bd4","modelId":"muse-wonder","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":50.0926784059314,"warning":null},{"id":"a702440f484ecf067ab37730","modelId":"muse-wonder","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":25.4833333333333,"warning":null},{"id":"0f58b6fee38fc8e39f5342e5","modelId":"muse-wonder","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":41.099999999999994,"warning":null},{"id":"2494c3776c4a648aac942118","modelId":"muse-wonder","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":60.300925925925895,"warning":null},{"id":"b2f531240ffd5178f869f695","modelId":"muse-wonder","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Muse Wonder (Max)","configuration":"Muse Wonder (Max); Model source: https://artificialanalysis.ai/models/muse-wonder","score":52.020202020202,"warning":null},{"id":"4447279d9098afb439d91f31","modelId":"muse-wonder-full","benchmarkId":"aa-aa-briefcase","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":1729.32,"warning":null},{"id":"fd1c73523af0a64b020f6d7c","modelId":"muse-wonder-full","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":83.6666666666667,"warning":null},{"id":"44b450aa27027808b9e57eb5","modelId":"muse-wonder-full","benchmarkId":"aa-automationbench-aa","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":71.33074973223975,"warning":null},{"id":"89cf6157d93f8c2efc2ad7f9","modelId":"muse-wonder-full","benchmarkId":"aa-critpt","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":32.857142857142904,"warning":null},{"id":"6121843833838096234efb94","modelId":"muse-wonder-full","benchmarkId":"aa-gdp-pdf","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":32.6,"warning":null},{"id":"b52ede219ffa5fc33ccd26c1","modelId":"muse-wonder-full","benchmarkId":"aa-gdpval-aa","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":1857.09,"warning":null},{"id":"7ce197385cfe06e9ee349cd6","modelId":"muse-wonder-full","benchmarkId":"aa-harvey-lab-aa-v1.1","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":13.055555555555557,"warning":null},{"id":"5ad56bef5143cd852c7e0a23","modelId":"muse-wonder-full","benchmarkId":"aa-humanitys-last-exam","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":53.6144578313253,"warning":null},{"id":"54b82abd2705526e243350f3","modelId":"muse-wonder-full","benchmarkId":"aa-omniscience","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":30.15,"warning":null},{"id":"c95e0955619a34d31cb1413e","modelId":"muse-wonder-full","benchmarkId":"aa-omniscience-accuracy","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":43.21666666666666,"warning":null},{"id":"9841a8670fe760a5c96f4521","modelId":"muse-wonder-full","benchmarkId":"aa-scicode","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":62.6157407407407,"warning":null},{"id":"0b29c27c06286dae67a3281d","modelId":"muse-wonder-full","benchmarkId":"aa-terminalbench-v4-0","level":"max","sourceLabel":"Muse Wonder Full (Max)","configuration":"Muse Wonder Full (Max); Model source: https://artificialanalysis.ai/models/muse-wonder-full","score":57.07070707070711,"warning":null},{"id":"3df372e3699dbb72d0107949","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.6666666666667,"warning":null},{"id":"7794c31b12edebaf055cadae","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"89e8cd465d74d7784c276a56","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":229.86,"warning":null},{"id":"eb0cd457f42770f092325384","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.8686868686869,"warning":null},{"id":"b8891413ad5cd82d8d171e25","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.81927710843374,"warning":null},{"id":"cb4b24a95d446b8f38f48a4f","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.1972789115646,"warning":null},{"id":"be2d9f71ad2fc32c4685a3e8","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.179190751445105,"warning":null},{"id":"ac0a28355ceae30603057943","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-57.4333333333333,"warning":null},{"id":"141eae2f30ceb3d7b2bde165","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.2,"warning":null},{"id":"b46e094398644f85af26779b","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":45.3216374269006,"warning":null},{"id":"c9b7eeb3faf1de8c8d016598","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.33333333333333,"warning":null},{"id":"baf677e18985ef29d7904ae8","modelId":"nemotron-3-nano-omni-30b-a3b-reasoning-bf16","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Nemotron 3 Nano Omni 30B A3B Reasoning","configuration":"Nemotron 3 Nano Omni 30B A3B Reasoning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-nano-omni-30b-a3b; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.741573033707859,"warning":null},{"id":"330c9ea34b512489404a52c8","modelId":"nemotron-3-ultra","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":6.25,"warning":null},{"id":"5a389b5bb1b65bf9c70de16c","modelId":"nemotron-3-ultra","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":876.28,"warning":null},{"id":"5ab4998706916204907bddaf","modelId":"nemotron-3-ultra","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":79.3333333333333,"warning":null},{"id":"8a53396aab5b755f3de4dee6","modelId":"nemotron-3-ultra","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":2.986349541593992,"warning":null},{"id":"33379178b6158f7ad04022e6","modelId":"nemotron-3-ultra","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":3.14285714285714,"warning":null},{"id":"034db885626e3df3ca80c3b4","modelId":"nemotron-3-ultra","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":28.886899433005077,"warning":null},{"id":"6ea2a53159a414f21eb5080f","modelId":"nemotron-3-ultra","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":5,"warning":null},{"id":"f736aa94d6c8dda205b28165","modelId":"nemotron-3-ultra","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":1013.25,"warning":null},{"id":"4cbe7ebbe159a64df52897d4","modelId":"nemotron-3-ultra","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":86.6666666666667,"warning":null},{"id":"51ea81ebfbd4d73fa92af147","modelId":"nemotron-3-ultra","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":28.405931417979602,"warning":null},{"id":"9896e0be3fe5460d4ed0ea51","modelId":"nemotron-3-ultra","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":81.3605442176871,"warning":null},{"id":"14ac5ca3a74e93eeb0b5d4ea","modelId":"nemotron-3-ultra","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":11.1111111111111,"warning":null},{"id":"5d823713f33e043edf26a0f7","modelId":"nemotron-3-ultra","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":-0.4,"warning":null},{"id":"d3fc36c3f92168bc3ee1be8e","modelId":"nemotron-3-ultra","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":22.583333333333332,"warning":null},{"id":"30b7b7ecf698984e25f319d1","modelId":"nemotron-3-ultra","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":40.2777777777778,"warning":null},{"id":"b9c479a24e769b4d00767391","modelId":"nemotron-3-ultra","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":83.3333333333333,"warning":null},{"id":"25ac32575044be41804c364b","modelId":"nemotron-3-ultra","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":14.2268041237113,"warning":null},{"id":"d3a5cd1a45f812308f2df5f7","modelId":"nemotron-3-ultra","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":36.363636363636395,"warning":null},{"id":"857c5f249a729e596ef2a8e4","modelId":"nemotron-3-ultra","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":53.9325842696629,"warning":null},{"id":"d3e9fc29cd1b8b5326223cc8","modelId":"nemotron-3-ultra","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Nemotron 3 Ultra 550B A55B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b","score":0.505050505050505,"warning":null},{"id":"1bead579ad997c55817f50a2","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":77.36,"warning":null},{"id":"4a17e10223ee1411771a3af5","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":13.3333333333333,"warning":null},{"id":"b2dc7594e6af2cc119bbf07f","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":91,"warning":null},{"id":"5a0f57a5757378689b26c0ef","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":10.6666666666667,"warning":null},{"id":"6942b61505042b6b4b415fa4","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":38,"warning":null},{"id":"740ea85e8bb22f60c115f846","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":0.9714694010433006,"warning":null},{"id":"4725e625f7ee2b6f05fc8d81","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":0,"warning":null},{"id":"04c131410c9c4622e4220c54","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":0.857142857142857,"warning":null},{"id":"105df0aa2225f0561f8563cb","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":1,"warning":null},{"id":"b8ea47eb24b0aec118cd61ea","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":-356.49,"warning":null},{"id":"84b55699614a8f191fed696f","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":255.22,"warning":null},{"id":"dc5659065cf79eedaaa45f2b","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":61.21875,"warning":null},{"id":"303aa21ce366cec63aaa88cc","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":39.8989898989899,"warning":null},{"id":"3bcc88ebddfd31d3a6afbaf9","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":75.6565656565657,"warning":null},{"id":"f396723527d1d08e73b2ade8","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":4.5875810936051895,"warning":null},{"id":"faa9167f4773b2090ba9857d","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":11.3994439295644,"warning":null},{"id":"d2d234d9d2b294ff915e8975","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":37.4829931972789,"warning":null},{"id":"920d7dc6fe4b7a6441efce8b","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":71.08843537414971,"warning":null},{"id":"ecfb7d6f27acde66bc3833ec","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":35.978835978836,"warning":null},{"id":"7cfb3235eb6f77d41014f6bc","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":74.0740740740741,"warning":null},{"id":"d5cf104e5d3166bb18ea1ac1","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":57.8956117021277,"warning":null},{"id":"b34aa90172208f886564ae04","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":79.37998670212771,"warning":null},{"id":"055179e91fb700bc3a6f59f6","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":-69.3,"warning":null},{"id":"691104d3d04d79d51b9cc955","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":-51.5666666666667,"warning":null},{"id":"3b39cae17a85fc9da8893154","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":11.366666666666667,"warning":null},{"id":"3e96872c29bb0036fc03afd9","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":17.299999999999997,"warning":null},{"id":"7592763efd7af6d017ddbac4","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":30.555555555555603,"warning":null},{"id":"6d927cefa7d4ed8898286a1c","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":25.4385964912281,"warning":null},{"id":"afc36157fe035615d4c4dcfb","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":40.935672514619895,"warning":null},{"id":"deb3f5b0ae380724472197f6","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":5.97938144329897,"warning":null},{"id":"36f02769540bd50aba5ce319","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b","score":12.1212121212121,"warning":null},{"id":"b8471a3b510bb8f93eb57bbf","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":13.6363636363636,"warning":null},{"id":"cae5947e36d58b255b600621","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":6.741573033707859,"warning":null},{"id":"72227cbdbda7131737c24853","modelId":"nvidia-nemotron-3-nano-30b-a3b-bf16","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning)","configuration":"NVIDIA Nemotron 3 Nano 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-30b-a3b-reasoning","score":0,"warning":null},{"id":"905b3fa1176d6dcfc0fd8f6a","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"b7085be742c2946fbbc31711","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.6666666666667,"warning":null},{"id":"25f9dc9602b3a57a955dd676","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.4483886019625834,"warning":null},{"id":"7fe21d4763c029be3e3150d3","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"414e997cb6b9d7038bad9641","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.4,"warning":null},{"id":"ffe59af75619d8f410f890a6","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-58.37,"warning":null},{"id":"89190aafce97ca2ae081fd7a","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.3131313131313,"warning":null},{"id":"44a97e00847114128f35d524","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.86561631139944,"warning":null},{"id":"8faa35e0975d2edcb8c369dc","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.23129251700681,"warning":null},{"id":"a9e95986995916b5c3500760","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-70.5333333333333,"warning":null},{"id":"74a2cfad0a1a8204637fd1e1","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.566666666666666,"warning":null},{"id":"6547929be39454ebdb26b3f5","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.087962962963,"warning":null},{"id":"13b5a452eb77a8d72c5aee9e","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.0701754385965,"warning":null},{"id":"df2ed9679fa5e588c1a56afc","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.81818181818182,"warning":null},{"id":"fc311965ffa7d6bd33acdb98","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.7453183520599302,"warning":null},{"id":"f8f038a0315adbed947e6c2f","modelId":"nvidia-nemotron-3-nano-4b-bf16","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"NVIDIA Nemotron 3 Nano 4B","configuration":"NVIDIA Nemotron 3 Nano 4B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-nano-4b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"b1f86a0d6ed3aac7cfa5a9f0","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":0,"warning":null},{"id":"e0d8ceb1e0ff885eb209f347","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":1.84365781710914,"warning":null},{"id":"22dcb321de77eebd96aca5cc","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":65.6666666666667,"warning":null},{"id":"5e4aa8b40fd444337c32809b","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":3.7790306947452605,"warning":null},{"id":"73ed6ba64bf2059d55d8be7e","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":3.14285714285714,"warning":null},{"id":"fce7835c797d7abb5d285155","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":23.545210384959713,"warning":null},{"id":"1cc0d1f19c8293a989d67698","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":2.6,"warning":null},{"id":"7b76eea2873079ee17182d99","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":490.4,"warning":null},{"id":"e03108aae6e2496d2303c047","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":80,"warning":null},{"id":"f892e23b0db5ed5a425cd84f","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":20.7599629286376,"warning":null},{"id":"3227135ad004c9ae9505d2ba","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":71.4965986394558,"warning":null},{"id":"fb0bf7feab41fec0f6aa816b","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":1.12994350282486,"warning":null},{"id":"310aba84e1b25c307bc27dbc","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":3.33333333333333,"warning":null},{"id":"c608bf4dd61434f7ad16f374","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":-41.5,"warning":null},{"id":"d8569bb6a64e4b906358c8d9","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":24.316666666666666,"warning":null},{"id":"deb1da1366abb23e09076986","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":36.2268518518519,"warning":null},{"id":"bd4190abebbd6c254565fd40","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":67.83625730994149,"warning":null},{"id":"79e868c3529be4536f58ca91","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":10.3092783505155,"warning":null},{"id":"5803ab6f2e085b119e49725d","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":28.7878787878788,"warning":null},{"id":"9f43288e9868e748c6d12ce8","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":38.5767790262172,"warning":null},{"id":"c8565a66e2745e03531942d4","modelId":"nvidia-nemotron-3-super-120b-a12b-bf16","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Nemotron 3 Super 120B A12B (Reasoning)","configuration":"Nemotron 3 Super 120B A12B (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-3-super-120b-a12b","score":0,"warning":null},{"id":"da634ce07b19c1ca5d16b7f9","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":506.17,"warning":null},{"id":"21959688eb48ffff3fe84528","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":60.33333333333331,"warning":null},{"id":"af2766360679f1e677876ff9","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.8410069634719541,"warning":null},{"id":"8328d0c67d5667e15aea1472","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"cb59004d5b0f5c683a653e25","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.352730528200535,"warning":null},{"id":"9fa52eeea400c87c7dcac0f2","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":3,"warning":null},{"id":"55bebf2bbc35fb34bcfd53af","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":638.57,"warning":null},{"id":"83dba6dbeba046f0f7d51cfa","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.3434343434343,"warning":null},{"id":"f70dadd761ac409f5b15d090","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.5653382761817,"warning":null},{"id":"4ac3635f000bb105bf308376","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.1111111111111098,"warning":null},{"id":"b1606fde6708193b5680178c","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":-17.7166666666667,"warning":null},{"id":"7132c5ebfae4d2f596ff0d23","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.433333333333334,"warning":null},{"id":"685493fcb0926afc1fc64549","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.060185185185205,"warning":null},{"id":"7edad9d658746f2d63c7df37","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.8659793814433,"warning":null},{"id":"b950df4c5b583a803b26d5bb","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.3445692883895,"warning":null},{"id":"328c4ebcaeafd433ae54b333","modelId":"nvidia-nemotron-3.5-lightning-30b-a3b-bf16","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Nemotron 3.5 Lightning","configuration":"Nemotron 3.5 Lightning; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/nemotron-3-5-lightning; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.505050505050505,"warning":null},{"id":"feba9b16c718555279f17565","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":26.6666666666667,"warning":null},{"id":"f85bf97903a3efa738c3c09d","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":75,"warning":null},{"id":"c0c3995b50ac95fdbbe5a33a","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":20.6666666666667,"warning":null},{"id":"fccbf930a0cd235301fd9c30","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":41,"warning":null},{"id":"a002c70aa4a84b956eb1d7a4","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":0,"warning":null},{"id":"947b304c37119f3e69f321e1","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":0,"warning":null},{"id":"91ab8f5ba0ad27187e93d250","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":43.9393939393939,"warning":null},{"id":"a3864c283775f7f8392f8896","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":57.1717171717172,"warning":null},{"id":"7c14b2fff22f134a398b1ba6","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":4.2632066728452305,"warning":null},{"id":"39f85a1aa619cb4553d41337","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":5.514365152919369,"warning":null},{"id":"959c553cf61235f930271918","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":25.8503401360544,"warning":null},{"id":"5a93007cac0b3f2e36f76b61","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":31.9047619047619,"warning":null},{"id":"2bb6cd4194cb711be51eeefe","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":34.4973544973545,"warning":null},{"id":"4a13875440065eb6af7f7070","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":69.4179894179894,"warning":null},{"id":"94684e452e66e092d884cf33","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":64.9185505319149,"warning":null},{"id":"fdf6426314faa35ff1e8e3a9","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":75.93085106382979,"warning":null},{"id":"0a091815e44913c3e3f3f3eb","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":44.508670520231206,"warning":null},{"id":"b4d8805b551c7b4e396cd13e","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":52.89017341040459,"warning":null},{"id":"794273354d8e571966e3c3cd","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":-72.2333333333333,"warning":null},{"id":"0afea5460705df6d4ceff65e","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":-64.2333333333333,"warning":null},{"id":"45875b87e9ce3fe9118e9ba9","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":11.916666666666668,"warning":null},{"id":"4aa3f76c32cc1abeba33a5cc","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":14.616666666666667,"warning":null},{"id":"a136762d0e55a8e533046dd3","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":19.2982456140351,"warning":null},{"id":"21db66dacfcddd9994ebaf85","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":21.3450292397661,"warning":null},{"id":"26ebef6866839033a02ce9af","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl","score":0,"warning":null},{"id":"d03e56381bfc8647adb6de5f","modelId":"nvidia-nemotron-nano-12b-v2-vl-bf16","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning)","configuration":"NVIDIA Nemotron Nano 12B v2 VL (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-12b-v2-vl-reasoning","score":4.545454545454549,"warning":null},{"id":"ce49848989035f224c0a9abe","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":62.33333333333329,"warning":null},{"id":"bf3009a8f08e136dd031ca3e","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":69.6666666666667,"warning":null},{"id":"2a318e72c061446677b7e820","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":24,"warning":null},{"id":"097cde742534d8b869269f2c","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":22.6666666666667,"warning":null},{"id":"33327d7ad7ed35ac4c558fd3","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":0,"warning":null},{"id":"c31a0552226bd1dfa238efb5","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":0,"warning":null},{"id":"8701a0a6fca8a6f69f770e8e","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":55.656565656565704,"warning":null},{"id":"e197387edb4f2e7ff0aa63b0","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":56.969696969697004,"warning":null},{"id":"5e44ddbffbe9cd84adbb0929","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":4.5875810936051895,"warning":null},{"id":"0ebb34a4287e8734ee82c77c","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":4.86561631139944,"warning":null},{"id":"8eda6116a8883034780a1873","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":27.142857142857103,"warning":null},{"id":"9ec18de31c80adfa15c57028","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":27.6190476190476,"warning":null},{"id":"fee2a2c4e302f056e0af9649","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":70.05291005291011,"warning":null},{"id":"14b88287eaf9ef9d92720eb8","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":72.38095238095241,"warning":null},{"id":"091ed4bb386bcf9a5a52226c","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":73.936170212766,"warning":null},{"id":"5afa45ddf4b7f15df4d108f4","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":74.2270611702128,"warning":null},{"id":"0c9b0458dabd028dcd7317d2","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":-57.25,"warning":null},{"id":"dd9503a5b07df03d5dd74bde","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":-41.55,"warning":null},{"id":"efdbda71278dabccbcce9c68","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":9.816666666666666,"warning":null},{"id":"21f2ccc498801a3c11aa79d5","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":11.799999999999999,"warning":null},{"id":"a617ac3d85cbb542b061ea8e","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":23.3918128654971,"warning":null},{"id":"f9acb65ceba2998896351b43","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":21.9298245614035,"warning":null},{"id":"2784590b707e50b89ab8f946","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2","score":0.757575757575758,"warning":null},{"id":"ebdd1011cfb7d6007184a931","modelId":"nvidia-nemotron-nano-9b-v2","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"NVIDIA Nemotron Nano 9B V2 (Reasoning)","configuration":"NVIDIA Nemotron Nano 9B V2 (Reasoning); Model source: https://artificialanalysis.ai/models/nvidia-nemotron-nano-9b-v2-reasoning","score":1.51515151515152,"warning":null},{"id":"d43df74ab94b7ff5b19c51f5","modelId":"o1","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":65,"warning":null},{"id":"d4b61d84c417f9d0cdad5d5d","modelId":"o1","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.28571428571428603,"warning":null},{"id":"dff08073706a671e10ffd04f","modelId":"o1","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.7474747474747,"warning":null},{"id":"c51ba771a259cf9fda6b04e1","modelId":"o1","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.01,"warning":null},{"id":"a99adc9667d02b7495975f92","modelId":"o1","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.3401360544218,"warning":null},{"id":"95eccd5b098daf7dce5b90a0","modelId":"o1","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.9365079365079,"warning":null},{"id":"7edb809ee87346c119299a36","modelId":"o1","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":97,"warning":null},{"id":"d68b4dbd3f5e0035224e7c76","modelId":"o1","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.0591755319149,"warning":null},{"id":"fe3b117476be77296f685c2b","modelId":"o1","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":-11.05,"warning":null},{"id":"b253a77469ef4f301baed94b","modelId":"o1","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.53333333333333,"warning":null},{"id":"ef4ce5d877f52434b3beb1fc","modelId":"o1","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":62.5730994152047,"warning":null},{"id":"9b71a59f4c3b1e0589de4a70","modelId":"o1","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"o1","configuration":"o1; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o1; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.8787878787879,"warning":null},{"id":"370cf31a2407efae21be583a","modelId":"o3","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.3333333333333,"warning":null},{"id":"f466d671d2eef974ea005b2b","modelId":"o3","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.6666666666667,"warning":null},{"id":"75946bb5200d898db3265c34","modelId":"o3","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.1428571428571401,"warning":null},{"id":"4f70a218b1f6a8fc2114d48e","modelId":"o3","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.7272727272727,"warning":null},{"id":"b3e92ab090534be36366d3ec","modelId":"o3","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.05,"warning":null},{"id":"cd4f93e7f4f1a45735079084","modelId":"o3","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.4285714285714,"warning":null},{"id":"30ee4a241c6f54cb0ca85362","modelId":"o3","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.8465608465608,"warning":null},{"id":"019b987204bf0a37c545c13c","modelId":"o3","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":99.2,"warning":null},{"id":"99c65e088b8907331a5c2554","modelId":"o3","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":85.3474069148936,"warning":null},{"id":"e6f4e33c7c2b7b4ca2e20e28","modelId":"o3","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.0578034682081,"warning":null},{"id":"350484fd270e9192ba0525d4","modelId":"o3","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":-15.55,"warning":null},{"id":"7cf10f4304e8d80bbcf66d39","modelId":"o3","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.56666666666666,"warning":null},{"id":"9a1204eb9a9a16442ef47947","modelId":"o3","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.7017543859649,"warning":null},{"id":"001226139bfffd4e21b870af","modelId":"o3","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"o3","configuration":"o3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.121212121212096,"warning":null},{"id":"d7e49807f538aef93f8c591f","modelId":"o3","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"o3 (High)","configuration":"o3 (High); Source model ID: o3-2025-04-16-high; 120 tasks","score":6.529999999999999,"warning":null},{"id":"c4c43e9a8ca07f6c7e48de89","modelId":"o3","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"o3 (Low)","configuration":"o3 (Low); Source model ID: o3-2025-04-16-low; 120 tasks","score":1.9900000000000002,"warning":null},{"id":"44ec3c5d0da4e3f62912c05f","modelId":"o3","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"o3 (Medium)","configuration":"o3 (Medium); Source model ID: o3-2025-04-16-medium; 120 tasks","score":2.98,"warning":null},{"id":"de6fe228603e9238b659ea7d","modelId":"o3","benchmarkId":"livecode-generationpass1","level":"high","sourceLabel":"O3 (High)","configuration":"O3 (High); Exact source model: o3__high; 454 tasks","score":75.8,"warning":null},{"id":"f798621cd8408dccf78c4bde","modelId":"o3-mini","benchmarkId":"aa-aa-briefcase","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":141.71,"warning":null},{"id":"8b5ae0dfc51e3593ff7eea9a","modelId":"o3-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":43,"warning":null},{"id":"ce25bd811a8c9be5cf5fcad6","modelId":"o3-mini","benchmarkId":"aa-automationbench-aa","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":0.5301828617801994,"warning":null},{"id":"3184656bb24fb157a043efea","modelId":"o3-mini","benchmarkId":"aa-critpt","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":0.28571428571428603,"warning":null},{"id":"aaeb7affc681c033e5f22f42","modelId":"o3-mini","benchmarkId":"aa-gdp-pdf","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":2,"warning":null},{"id":"9231fbe27f842af6222ef99e","modelId":"o3-mini","benchmarkId":"aa-gdpval-aa","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":230.33,"warning":null},{"id":"135ae8cbbf53474e7fb1efc9","modelId":"o3-mini","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":77.27272727272731,"warning":null},{"id":"811d9c240ea23b82e185a589","modelId":"o3-mini","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"o3-mini","configuration":"o3-mini; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.8484848484848,"warning":null},{"id":"14a282feb31ea65634e84a45","modelId":"o3-mini","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":12.04,"warning":null},{"id":"e161b5a9b646a096149330a7","modelId":"o3-mini","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"o3-mini","configuration":"o3-mini; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.91,"warning":null},{"id":"c582d7a8519fb434760e4e6b","modelId":"o3-mini","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":67.14285714285711,"warning":null},{"id":"9400c37858523c955de8d6e4","modelId":"o3-mini","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":73.4391534391534,"warning":null},{"id":"318e34735b7b00c4d11a9a09","modelId":"o3-mini","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"o3-mini","configuration":"o3-mini; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.7460317460317,"warning":null},{"id":"0e7815d134f07daf47fea590","modelId":"o3-mini","benchmarkId":"aa-math-500","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":98.4666666666667,"warning":null},{"id":"b03c9376d3f09d3c80e63a54","modelId":"o3-mini","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"o3-mini","configuration":"o3-mini; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":97.3333333333333,"warning":null},{"id":"00d00ac56f6a940bfa64bfd0","modelId":"o3-mini","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":80.1944813829787,"warning":null},{"id":"e266c73c6ec784e58caaf44a","modelId":"o3-mini","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"o3-mini","configuration":"o3-mini; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.05585106382979,"warning":null},{"id":"9d9797f0abd5c758079a5414","modelId":"o3-mini","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":-42.5666666666667,"warning":null},{"id":"e13cbca155f11e1f6ecf4e46","modelId":"o3-mini","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":21.266666666666666,"warning":null},{"id":"c215eed3a150084b6b203302","modelId":"o3-mini","benchmarkId":"aa-scicode","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":42.8240740740741,"warning":null},{"id":"8d2677b32d5f0f4491afc41c","modelId":"o3-mini","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":31.2865497076023,"warning":null},{"id":"19eb94c3b4a5578e45bc7056","modelId":"o3-mini","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"o3-mini","configuration":"o3-mini; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.6549707602339,"warning":null},{"id":"2ebe91127bd5a4a67fd681d4","modelId":"o3-mini","benchmarkId":"aa-tau3-banking","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":5.1546391752577305,"warning":null},{"id":"bbb4811951b5e83028f5f3c5","modelId":"o3-mini","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":6.06060606060606,"warning":null},{"id":"4da7c16ca683432d6e1cd5ba","modelId":"o3-mini","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"o3-mini","configuration":"o3-mini; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.81818181818182,"warning":null},{"id":"10f2bd645bc2a415a416f1fb","modelId":"o3-mini","benchmarkId":"aa-terminalbench-v2-1","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":4.49438202247191,"warning":null},{"id":"00bc53e0be6ceece3af8f02f","modelId":"o3-mini","benchmarkId":"aa-terminalbench-v4-0","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Model source: https://artificialanalysis.ai/models/o3-mini-high","score":0,"warning":null},{"id":"c2c3b8b661de8636e2ceca83","modelId":"o3-mini","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"o3-mini (High)","configuration":"o3-mini (High); Source model ID: o3-mini-2025-01-31-high; 120 tasks","score":2.9899999999999998,"warning":null},{"id":"78ba06f11057b6bec4542c33","modelId":"o3-mini","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"o3-mini (Low)","configuration":"o3-mini (Low); Source model ID: o3-mini-2025-01-31-low; 120 tasks","score":0,"warning":null},{"id":"e76fe3fd7aff227cd9253390","modelId":"o3-mini","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"o3-mini (Medium)","configuration":"o3-mini (Medium); Source model ID: o3-mini-2025-01-31-medium; 120 tasks","score":2.08,"warning":null},{"id":"bd98f3407d228d7b34d15a3b","modelId":"o3-pro","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"o3-pro","configuration":"o3-pro; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/o3-pro; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.5454545454545,"warning":null},{"id":"549730e49987d6154a62d579","modelId":"o3-pro","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"o3-Pro (High)","configuration":"o3-Pro (High); Source model ID: o3-pro-2025-06-10-high; 120 tasks","score":4.859999999999999,"warning":null},{"id":"342bab39cacc26dd8a82dd79","modelId":"o3-pro","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"o3-Pro (Low)","configuration":"o3-Pro (Low); Source model ID: o3-pro-2025-06-10-low; 120 tasks","score":2.0500000000000003,"warning":null},{"id":"f5bae5777f5b932bda30b31f","modelId":"o3-pro","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"o3-Pro (Medium)","configuration":"o3-Pro (Medium); Source model ID: o3-pro-2025-06-10-medium; 120 tasks","score":1.94,"warning":null},{"id":"89cc7d75dceb87b2426e3c5b","modelId":"o4-mini","benchmarkId":"aa-aime-2025","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":90.6666666666667,"warning":null},{"id":"b8d061eeef87eecd7597aeca","modelId":"o4-mini","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":61,"warning":null},{"id":"6e17cd87254e16c222ad4bc3","modelId":"o4-mini","benchmarkId":"aa-critpt","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":0.571428571428571,"warning":null},{"id":"e7191e9fd0004f5a99fc6482","modelId":"o4-mini","benchmarkId":"aa-gpqa-diamond","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":78.3838383838384,"warning":null},{"id":"570a0a08c9367786911ce39e","modelId":"o4-mini","benchmarkId":"aa-humanitys-last-exam","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":16.5430954587581,"warning":null},{"id":"bc1a2bb785bdc2ede2ef377b","modelId":"o4-mini","benchmarkId":"aa-ifbench","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":68.7074829931973,"warning":null},{"id":"d96f6107d7d9bc4904585219","modelId":"o4-mini","benchmarkId":"aa-livecodebench","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":85.9259259259259,"warning":null},{"id":"08f77210e749e1a2b05bb72e","modelId":"o4-mini","benchmarkId":"aa-math-500","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":98.8666666666667,"warning":null},{"id":"58b38221383ebed4eb393aed","modelId":"o4-mini","benchmarkId":"aa-mmlu-pro","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":83.1948138297872,"warning":null},{"id":"5b0a929ebb9c554267ebe86a","modelId":"o4-mini","benchmarkId":"aa-mmmu-pro","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":69.2485549132948,"warning":null},{"id":"254ca466b5e8ceb4828ba2d6","modelId":"o4-mini","benchmarkId":"aa-omniscience","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":-35.7,"warning":null},{"id":"08c9b038b1daaba61f3a3d12","modelId":"o4-mini","benchmarkId":"aa-omniscience-accuracy","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":24.8,"warning":null},{"id":"cb71123eba414880ed924c38","modelId":"o4-mini","benchmarkId":"aa-tau2-bench","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":55.5555555555556,"warning":null},{"id":"6f3287925c9aa8200e326720","modelId":"o4-mini","benchmarkId":"aa-terminalbench-hard","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Model source: https://artificialanalysis.ai/models/o4-mini","score":15.151515151515198,"warning":null},{"id":"e74df9fbfd59b581eb76ffa2","modelId":"o4-mini","benchmarkId":"arc-v2semiprivate","level":"high","sourceLabel":"o4-mini (High)","configuration":"o4-mini (High); Source model ID: o4-mini-2025-04-16-high; 120 tasks","score":6.11,"warning":null},{"id":"a03ef69f5f0bf9f19a64b6c1","modelId":"o4-mini","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"o4-mini (Low)","configuration":"o4-mini (Low); Source model ID: o4-mini-2025-04-16-low; 120 tasks","score":1.67,"warning":null},{"id":"904517858e9c5b0473e556b3","modelId":"o4-mini","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"o4-mini (Medium)","configuration":"o4-mini (Medium); Source model ID: o4-mini-2025-04-16-medium; 120 tasks","score":2.36,"warning":null},{"id":"ee9d1ca8d2407e02d504ac6b","modelId":"o4-mini","benchmarkId":"livecode-generationpass1","level":"high","sourceLabel":"O4-Mini (High)","configuration":"O4-Mini (High); Exact source model: o4-mini__high; 454 tasks","score":80.2,"warning":null},{"id":"c16742f35a5ca0f1da2cd5d9","modelId":"o4-mini","benchmarkId":"livecode-generationpass1","level":"low","sourceLabel":"O4-Mini (Low)","configuration":"O4-Mini (Low); Exact source model: o4-mini__low; 454 tasks","score":65.9,"warning":null},{"id":"dbd3c0c14e4d3b745599aed7","modelId":"o4-mini","benchmarkId":"livecode-generationpass1","level":"medium","sourceLabel":"O4-Mini (Medium)","configuration":"O4-Mini (Medium); Exact source model: o4-mini__medium; 454 tasks","score":74.2,"warning":null},{"id":"77b72d8bb8f6229c4935bd1d","modelId":"phi-4","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":18,"warning":null},{"id":"7913c6e2413a96102bbbf1fe","modelId":"phi-4","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"834342d113eaeacdec6c68c2","modelId":"phi-4","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"c45844e7e930f528caf96218","modelId":"phi-4","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":57.4747474747475,"warning":null},{"id":"92a3672a2fcdd9eae2cfe4c8","modelId":"phi-4","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.7600000000000002,"warning":null},{"id":"70dd1ef196043632b7c37a03","modelId":"phi-4","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.5374149659864,"warning":null},{"id":"f2fa14f24036f5ed63850347","modelId":"phi-4","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.068783068783098,"warning":null},{"id":"b1300da3198b950627b64868","modelId":"phi-4","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":81,"warning":null},{"id":"e87a2d3e2af83fbba103ecc7","modelId":"phi-4","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.4428191489362,"warning":null},{"id":"d4d5188cfd91b9b6ddb85ec5","modelId":"phi-4","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":-55.7,"warning":null},{"id":"aff564f7946147cb44a15141","modelId":"phi-4","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.066666666666666,"warning":null},{"id":"68eef27c9393c82acd930591","modelId":"phi-4","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"a0d8027ad98e252733a703ca","modelId":"phi-4","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Phi-4","configuration":"Phi-4; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.78787878787879,"warning":null},{"id":"516823481eb1495f3fb3c99c","modelId":"phi-4-mini-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.66666666666667,"warning":null},{"id":"601f36f2404c745b7bc714e7","modelId":"phi-4-mini-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.333333333333298,"warning":null},{"id":"ab4cc6c2aa36c216f8568f92","modelId":"phi-4-mini-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"ea4ce21671423f998900aef4","modelId":"phi-4-mini-instruct","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":-408.3,"warning":null},{"id":"5438058f26b61c57fa0f9afc","modelId":"phi-4-mini-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.1313131313131,"warning":null},{"id":"4d6bf305ef56bc9d7d14e527","modelId":"phi-4-mini-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.45,"warning":null},{"id":"f1959c33d56b49b87d365a30","modelId":"phi-4-mini-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.0884353741497,"warning":null},{"id":"cad43079054208b6d9bfa144","modelId":"phi-4-mini-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.592592592592599,"warning":null},{"id":"b9dfce1492ec029021a05b9e","modelId":"phi-4-mini-instruct","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.6,"warning":null},{"id":"8aa58d9564291e79f8c7d5aa","modelId":"phi-4-mini-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.484375,"warning":null},{"id":"ec4c88a12f8e0d0998cc68b9","modelId":"phi-4-mini-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":-61.0833333333333,"warning":null},{"id":"70e69f2115f632139a54d1c0","modelId":"phi-4-mini-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.5,"warning":null},{"id":"c13bc87ea7fbb9edee07b054","modelId":"phi-4-mini-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.18713450292398,"warning":null},{"id":"64dbb96bbf3ee55255c6a592","modelId":"phi-4-mini-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"53b7d3d06cae0753bfaa177b","modelId":"phi-4-mini-instruct","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Phi-4 Mini Instruct","configuration":"Phi-4 Mini Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-mini; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.374531835205993,"warning":null},{"id":"be0d629d53023aac889d0050","modelId":"phi-4-multimodal-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Phi-4 Multimodal Instruct","configuration":"Phi-4 Multimodal Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-multimodal; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.5151515151515,"warning":null},{"id":"5e56b10f487f5015dd377634","modelId":"phi-4-multimodal-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Phi-4 Multimodal Instruct","configuration":"Phi-4 Multimodal Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-multimodal; Mode from evaluator isReasoning metadata; effort budget not specified","score":5,"warning":null},{"id":"1c00cc4d194123d3e4240049","modelId":"phi-4-multimodal-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Phi-4 Multimodal Instruct","configuration":"Phi-4 Multimodal Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-multimodal; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.1216931216931,"warning":null},{"id":"9b113a9a9d3b6af3846f819a","modelId":"phi-4-multimodal-instruct","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Phi-4 Multimodal Instruct","configuration":"Phi-4 Multimodal Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-multimodal; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.2666666666667,"warning":null},{"id":"88f07e63051ff123b88335b7","modelId":"phi-4-multimodal-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Phi-4 Multimodal Instruct","configuration":"Phi-4 Multimodal Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-multimodal; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.545545212766,"warning":null},{"id":"10aaef8ddcae5e8c1c02ab70","modelId":"phi-4-multimodal-instruct","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Phi-4 Multimodal Instruct","configuration":"Phi-4 Multimodal Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/phi-4-multimodal; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.5086705202312,"warning":null},{"id":"95791dc88f64b23b16d4019b","modelId":"pixtral-large-2411","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.33333333333333,"warning":null},{"id":"528da2e0681c24dea179b207","modelId":"pixtral-large-2411","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.5050505050505,"warning":null},{"id":"0f3e5bc4908fb6b8e1788d00","modelId":"pixtral-large-2411","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.8000000000000003,"warning":null},{"id":"52b14f5c6812c2e2fb14b98f","modelId":"pixtral-large-2411","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.4897959183673,"warning":null},{"id":"cd0a839399405ae15ba9139d","modelId":"pixtral-large-2411","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.137566137566097,"warning":null},{"id":"2a7f996917775f67ba770d2f","modelId":"pixtral-large-2411","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.39999999999999,"warning":null},{"id":"d563ff92ab0b6ba9cff379b5","modelId":"pixtral-large-2411","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.1130319148936,"warning":null},{"id":"eb614b4b073a151ea844b8a5","modelId":"pixtral-large-2411","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.5780346820809,"warning":null},{"id":"f3c6fe94a03132a6f2f3db5f","modelId":"pixtral-large-2411","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Pixtral Large","configuration":"Pixtral Large; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/pixtral-large-2411; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.5497076023392,"warning":null},{"id":"4b4864c5698ea2d2631a1c9b","modelId":"qwen-3.8-max","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":1384.6,"warning":null},{"id":"d0cc822040713a4a3efe730d","modelId":"qwen-3.8-max","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.3333333333333,"warning":null},{"id":"817cf9c2f5eef9bc5445ff87","modelId":"qwen-3.8-max","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.23210407553669,"warning":null},{"id":"efcf7cbb7c05f575e871db63","modelId":"qwen-3.8-max","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":20,"warning":null},{"id":"656bbefe676aa7366bf925d2","modelId":"qwen-3.8-max","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.200000000000003,"warning":null},{"id":"4a9f8192937c5c89a78b9cf9","modelId":"qwen-3.8-max","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":1604.43,"warning":null},{"id":"636a97545dd50cd6369243eb","modelId":"qwen-3.8-max","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":92.7272727272727,"warning":null},{"id":"db4ce26e91d3f5e8dd8b84fa","modelId":"qwen-3.8-max","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.0491195551437,"warning":null},{"id":"5239aacc45eb8d8b7ddd3048","modelId":"qwen-3.8-max","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.4444444444444,"warning":null},{"id":"77aaa1bbb6b116ec021af6ec","modelId":"qwen-3.8-max","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.3121387283237,"warning":null},{"id":"3d42354ed91a3ab11dea3013","modelId":"qwen-3.8-max","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.4,"warning":null},{"id":"122f3dc0a4526b7d40b88c00","modelId":"qwen-3.8-max","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.85,"warning":null},{"id":"2132db3ad39a78473d68e747","modelId":"qwen-3.8-max","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":53.2407407407407,"warning":null},{"id":"4623a80d9f8f5d6145f2f78a","modelId":"qwen-3.8-max","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.340206185567006,"warning":null},{"id":"67ab23b646557e62640853e6","modelId":"qwen-3.8-max","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.2734082397004,"warning":null},{"id":"2cc3e4c4ce6d2eae59b4380d","modelId":"qwen-3.8-max","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max-0803; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.6868686868687,"warning":null},{"id":"fdf7a8eca4003c72a2a46e18","modelId":"qwen-3.8-max","benchmarkId":"deepswe-minisweagent","level":"xhigh","sourceLabel":"qwen3-8-max","configuration":"qwen3-8-max; Reported effort: xhigh; 113 unique tasks; 449 attempts; 4 runs","score":57.5,"warning":null},{"id":"da52383d0adbea5bac566881","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":45,"warning":null},{"id":"206fd8d279c674cdec835285","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":1612.8,"warning":null},{"id":"fa1a1bdf910ef317b8401dea","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.4041297935103,"warning":null},{"id":"c19a4187a390134377264605","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.3333333333333,"warning":null},{"id":"b0859f1b81ba43c4145f5564","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":56.1905381087678,"warning":null},{"id":"e75523cf36922d42a24da9fb","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.714285714285698,"warning":null},{"id":"60804c4839d8668fdf2ea433","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.62757385854969,"warning":null},{"id":"0b88e41bb58b9c73ec1d1365","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.8,"warning":null},{"id":"d175d2876882d79f1d19033a","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":1665.58,"warning":null},{"id":"cd3d162756c1542303a70143","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":92.82828282828281,"warning":null},{"id":"568a1843c4cc6821607175c1","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.0954587581094,"warning":null},{"id":"cf2c0f94f35b14e6893dc878","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.254237288135606,"warning":null},{"id":"e64bf939cef3770af807eec9","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":20,"warning":null},{"id":"ffd1cced0f7fffa9fbef8d49","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.7745664739884,"warning":null},{"id":"c5748c028be706993862b536","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.9833333333333,"warning":null},{"id":"c52b4fcacf00d9875e5df164","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.683333333333337,"warning":null},{"id":"ea6fb5c196f9227bddf85436","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.0833333333333,"warning":null},{"id":"97cfbdd8759773dcb23ca28c","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.835051546391796,"warning":null},{"id":"d206ce1d616b3fb7831293d2","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-terminal-bench-science","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.9047619047619,"warning":null},{"id":"b23b336fbb3ed47d2adb3456","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.7640449438202,"warning":null},{"id":"4511d031aae80264db54af32","modelId":"qwen-3.8-max-0902","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.8 Max (0902)","configuration":"Qwen3.8 Max (0902); Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.8888888888889,"warning":null},{"id":"e07d9dd5005dfa0df02827b8","modelId":"qwen3-0.6b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":10.3333333333333,"warning":null},{"id":"a6f7a5f79bbca4f2a67412d9","modelId":"qwen3-0.6b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":18,"warning":null},{"id":"c8d0f96fa682deeb06e51740","modelId":"qwen3-0.6b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":0,"warning":null},{"id":"ecde6783ec0d47e1bf98a46a","modelId":"qwen3-0.6b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":0,"warning":null},{"id":"b338841b63f737ffb17ed7e0","modelId":"qwen3-0.6b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":0,"warning":null},{"id":"902940590320d8648dca2883","modelId":"qwen3-0.6b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":0,"warning":null},{"id":"f04fd1b531b951f250617193","modelId":"qwen3-0.6b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":23.1313131313131,"warning":null},{"id":"3d1a9de8bbd963f3c5f6bf0d","modelId":"qwen3-0.6b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":23.9393939393939,"warning":null},{"id":"90e7d8ad72bd6576a81f42eb","modelId":"qwen3-0.6b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":4.93,"warning":null},{"id":"6518f5de94545adfe1985361","modelId":"qwen3-0.6b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":5.64,"warning":null},{"id":"230731584c91a733879819a5","modelId":"qwen3-0.6b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":21.904761904761898,"warning":null},{"id":"a6bbd08d320ee6b9dfd559fd","modelId":"qwen3-0.6b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":23.3333333333333,"warning":null},{"id":"55a17c98e627b355cd8b17da","modelId":"qwen3-0.6b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":7.3015873015873005,"warning":null},{"id":"0c6f9d11fcb4e03bd1b8a9a7","modelId":"qwen3-0.6b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":12.0634920634921,"warning":null},{"id":"20398efbf3c87c56cf33356f","modelId":"qwen3-0.6b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":52.066666666666706,"warning":null},{"id":"8e582fa9a22661b9a0279e7a","modelId":"qwen3-0.6b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":75,"warning":null},{"id":"a370fde9d0ef2c43e306e7f1","modelId":"qwen3-0.6b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":23.0718085106383,"warning":null},{"id":"9fa5e0ad520b1ed9244765c2","modelId":"qwen3-0.6b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":34.682513297872305,"warning":null},{"id":"607485fd1a1b0b822e375d3e","modelId":"qwen3-0.6b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":-84.95,"warning":null},{"id":"7427f5729e938d7772017aa2","modelId":"qwen3-0.6b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":-78.85,"warning":null},{"id":"b50bf9a788cc1febbbb1372a","modelId":"qwen3-0.6b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":4.583333333333333,"warning":null},{"id":"692d1a2005d56e466b6dc93b","modelId":"qwen3-0.6b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":6.333333333333334,"warning":null},{"id":"aaf8eb92ad1848d2c667404a","modelId":"qwen3-0.6b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":14.619883040935699,"warning":null},{"id":"62476ba134bc76ca975eaf3c","modelId":"qwen3-0.6b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":21.052631578947402,"warning":null},{"id":"5cc385d16cc1b35f2bcc4b90","modelId":"qwen3-0.6b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 0.6B (Non-reasoning)","configuration":"Qwen3 0.6B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct","score":0,"warning":null},{"id":"2d2875437a15ced44a23ebcc","modelId":"qwen3-0.6b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 0.6B (Reasoning)","configuration":"Qwen3 0.6B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-0.6b-instruct-reasoning","score":0,"warning":null},{"id":"f0903b0785448402e3ee7b14","modelId":"qwen3-1.7b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":7.33333333333333,"warning":null},{"id":"1708d3fe7d21b8dc9194f8c4","modelId":"qwen3-1.7b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":38.6666666666667,"warning":null},{"id":"8b2f14005445f5c569f1b1ce","modelId":"qwen3-1.7b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":0,"warning":null},{"id":"6081c63272fd15a08492c8db","modelId":"qwen3-1.7b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":0,"warning":null},{"id":"fd4f239d936984cd38e77d8c","modelId":"qwen3-1.7b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":0,"warning":null},{"id":"ee141c0067f8a08b74460ce2","modelId":"qwen3-1.7b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":0,"warning":null},{"id":"0d1130f0b8f0226c88c62a1e","modelId":"qwen3-1.7b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":48.839285714285715,"warning":null},{"id":"5f5f56834853b82731dea301","modelId":"qwen3-1.7b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":28.282828282828298,"warning":null},{"id":"5605706863518eb467d8a1b5","modelId":"qwen3-1.7b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":35.5555555555556,"warning":null},{"id":"8661b817f3af2404661a7e11","modelId":"qwen3-1.7b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":5.319999999999999,"warning":null},{"id":"5392dc4a9cc4b181f5b920c6","modelId":"qwen3-1.7b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":4.64,"warning":null},{"id":"a1f40f1a67da3871ed644d2c","modelId":"qwen3-1.7b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":21.0884353741497,"warning":null},{"id":"88058002752a6680711097c0","modelId":"qwen3-1.7b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":26.8707482993197,"warning":null},{"id":"a6d45937d0d15bb4efa4365c","modelId":"qwen3-1.7b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":12.592592592592599,"warning":null},{"id":"a87339e0e9504b36ddb6bded","modelId":"qwen3-1.7b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":30.7936507936508,"warning":null},{"id":"50e7de8cdf0b7ec58dc21ada","modelId":"qwen3-1.7b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":71.7333333333333,"warning":null},{"id":"f2145165c64ee568b19fe04a","modelId":"qwen3-1.7b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":89.4,"warning":null},{"id":"30e97a4e868b3d4ca68973be","modelId":"qwen3-1.7b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":41.057180851063805,"warning":null},{"id":"c412281a2b4367841d9fc1eb","modelId":"qwen3-1.7b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":56.9730718085106,"warning":null},{"id":"460ee126fa722ac88fa60ce9","modelId":"qwen3-1.7b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":-84.0833333333333,"warning":null},{"id":"33f988c4c7a2096720a36eb5","modelId":"qwen3-1.7b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":-77.5333333333333,"warning":null},{"id":"82004cae536a8489b3f16bc2","modelId":"qwen3-1.7b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":6.916666666666667,"warning":null},{"id":"b545a2bd1e66b5dcad42acc3","modelId":"qwen3-1.7b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":8.866666666666667,"warning":null},{"id":"5782bd6d71560bdef6e6bbf0","modelId":"qwen3-1.7b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":21.6374269005848,"warning":null},{"id":"f7bac8d61f92b0d58d694350","modelId":"qwen3-1.7b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":26.023391812865498,"warning":null},{"id":"7a6bdbbd5bb164a640d8cff1","modelId":"qwen3-1.7b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 1.7B (Non-reasoning)","configuration":"Qwen3 1.7B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct","score":0,"warning":null},{"id":"8631ae15df88ea157c982d73","modelId":"qwen3-1.7b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 1.7B (Reasoning)","configuration":"Qwen3 1.7B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-1.7b-instruct-reasoning","score":0,"warning":null},{"id":"c200f8f5974cced91b94c3a7","modelId":"qwen3-14b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":0,"warning":null},{"id":"7823070fce74572c14440227","modelId":"qwen3-14b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":57.99999999999999,"warning":null},{"id":"3309e13df52fd1fabb19ab86","modelId":"qwen3-14b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":55.6666666666667,"warning":null},{"id":"360c4369d2880bd3f33bf48d","modelId":"qwen3-14b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":0,"warning":null},{"id":"93da3ff3c5d7b1bff4c94ecb","modelId":"qwen3-14b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":0,"warning":null},{"id":"3b2a114e7d0f60358dc89582","modelId":"qwen3-14b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":0.19111706206154658,"warning":null},{"id":"5bafcc8253fd4f179cd575bd","modelId":"qwen3-14b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":0,"warning":null},{"id":"22963f5bec954903994c9380","modelId":"qwen3-14b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":0,"warning":null},{"id":"28e79ab1e630d682cf1391f6","modelId":"qwen3-14b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":1.2,"warning":null},{"id":"db976faf2da60e8e24944da4","modelId":"qwen3-14b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":-24.33,"warning":null},{"id":"76b7eda2e03b9fb81d741935","modelId":"qwen3-14b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":46.969696969697,"warning":null},{"id":"e62d0d2d9690c57703bf9f13","modelId":"qwen3-14b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":60.4040404040404,"warning":null},{"id":"fff6cdef363e6e229f008d13","modelId":"qwen3-14b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":4.09,"warning":null},{"id":"29501eeeebecd801dc346bab","modelId":"qwen3-14b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":4.53,"warning":null},{"id":"c6cac11306b7bc7f25813687","modelId":"qwen3-14b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":23.9455782312925,"warning":null},{"id":"c9662b1cbc42cb80f1003cb1","modelId":"qwen3-14b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":40.5442176870748,"warning":null},{"id":"4a6b0f3481ccc7e6d3e485b4","modelId":"qwen3-14b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":28.042328042328002,"warning":null},{"id":"d182d825eeccc745161b2067","modelId":"qwen3-14b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":52.27513227513229,"warning":null},{"id":"f92bf192d6248fe1b4e85427","modelId":"qwen3-14b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":87.1333333333333,"warning":null},{"id":"41d90092001ad63411e08f39","modelId":"qwen3-14b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":96.1333333333333,"warning":null},{"id":"30b278ac1c8c5464b1a1def2","modelId":"qwen3-14b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":67.4617686170213,"warning":null},{"id":"f1ac22cc2380f41a90cbfc68","modelId":"qwen3-14b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":77.3769946808511,"warning":null},{"id":"f9876ccbd46960970f8661c0","modelId":"qwen3-14b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":-66.6666666666667,"warning":null},{"id":"fffa4d922bb3d13549db3532","modelId":"qwen3-14b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":-48.95,"warning":null},{"id":"6c21071980d804a1d67a0c8c","modelId":"qwen3-14b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":13.333333333333334,"warning":null},{"id":"ea5ca53936049bd3f26c2477","modelId":"qwen3-14b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":15.283333333333331,"warning":null},{"id":"a32d713ac22ff457128d16eb","modelId":"qwen3-14b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":30.6712962962963,"warning":null},{"id":"05dcd8e868462f666322c651","modelId":"qwen3-14b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":32.1637426900585,"warning":null},{"id":"9cf6758bbc1a1642912303d1","modelId":"qwen3-14b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":34.5029239766082,"warning":null},{"id":"7f1d3dd2bf9d082e2778499a","modelId":"qwen3-14b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":5.56701030927835,"warning":null},{"id":"57f6dbe626d153af997eff0b","modelId":"qwen3-14b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 14B (Non-reasoning)","configuration":"Qwen3 14B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct","score":5.3030303030303,"warning":null},{"id":"2ce50946464c3e30979c50bf","modelId":"qwen3-14b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":3.78787878787879,"warning":null},{"id":"7eee7a8e12e744a002d4d6b4","modelId":"qwen3-14b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":4.8689138576779,"warning":null},{"id":"1b5bf342de1a14490bce8c1a","modelId":"qwen3-14b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3 14B (Reasoning)","configuration":"Qwen3 14B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-14b-instruct-reasoning","score":0,"warning":null},{"id":"38375a3f39d92c5ad2706da2","modelId":"qwen3-235b-a22b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":23.6666666666667,"warning":null},{"id":"89305b27a86a0ce63a7e4eb8","modelId":"qwen3-235b-a22b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":82,"warning":null},{"id":"56d76f45737dce7a63a667ff","modelId":"qwen3-235b-a22b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":0,"warning":null},{"id":"14f9f716a2a328d472ca21e0","modelId":"qwen3-235b-a22b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":0,"warning":null},{"id":"ca93a3563807fd6a9ff50462","modelId":"qwen3-235b-a22b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":0,"warning":null},{"id":"e658e6c4bd06b5c59d0cdf8b","modelId":"qwen3-235b-a22b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":0,"warning":null},{"id":"ca162e62aabc277e7d0ea215","modelId":"qwen3-235b-a22b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":61.313131313131294,"warning":null},{"id":"c361e8d72495bb5786b6cf90","modelId":"qwen3-235b-a22b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":70,"warning":null},{"id":"8fb4c5a3c327a127ebc0ebc9","modelId":"qwen3-235b-a22b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":4.19,"warning":null},{"id":"cf462f7e796944884abc14f1","modelId":"qwen3-235b-a22b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":10.979999999999999,"warning":null},{"id":"839df9c48331bc134df3997d","modelId":"qwen3-235b-a22b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":36.5986394557823,"warning":null},{"id":"b0696a5cc76427177558d015","modelId":"qwen3-235b-a22b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":38.7074829931973,"warning":null},{"id":"ab7dc27978537cbe461952da","modelId":"qwen3-235b-a22b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":34.285714285714306,"warning":null},{"id":"87125f56c97261ce790af1db","modelId":"qwen3-235b-a22b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":62.2222222222222,"warning":null},{"id":"a5ec4ea7bfb7ef8568384a61","modelId":"qwen3-235b-a22b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":90.2,"warning":null},{"id":"43cf687682a8eb0a638a3c45","modelId":"qwen3-235b-a22b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":93,"warning":null},{"id":"fe9af6c837da0b3d9cf4a593","modelId":"qwen3-235b-a22b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":76.2134308510638,"warning":null},{"id":"227ee0acf1c6c720d1350ecf","modelId":"qwen3-235b-a22b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":82.7958776595745,"warning":null},{"id":"c8b44554d63181fae5555e52","modelId":"qwen3-235b-a22b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":-52.3833333333333,"warning":null},{"id":"c34a248fd9868cde08c5ce20","modelId":"qwen3-235b-a22b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":-44.7,"warning":null},{"id":"69f9f3d7ea24015d72835b85","modelId":"qwen3-235b-a22b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":18.566666666666666,"warning":null},{"id":"860f80781817eb047ddb5ff5","modelId":"qwen3-235b-a22b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":18.53333333333333,"warning":null},{"id":"0326ce95ca6eb85d64c1291b","modelId":"qwen3-235b-a22b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":27.192982456140403,"warning":null},{"id":"5ce73290decba9279d8cafa3","modelId":"qwen3-235b-a22b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":23.9766081871345,"warning":null},{"id":"b923b18effc8abc789063841","modelId":"qwen3-235b-a22b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B (Non-reasoning)","configuration":"Qwen3 235B A22B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct","score":6.06060606060606,"warning":null},{"id":"1a1ddf3972ddca61d88d99d3","modelId":"qwen3-235b-a22b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 235B A22B (Reasoning)","configuration":"Qwen3 235B A22B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-reasoning","score":6.06060606060606,"warning":null},{"id":"d77b602a8a35d0a4a648dd95","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.6666666666667,"warning":null},{"id":"67ad2e1e4f3abbfc96fd72d0","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.900000000000006,"warning":null},{"id":"1d7d03c896b07879689bc033","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"1622a07edd5343e87803e57a","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.2525252525253,"warning":null},{"id":"d64fc0c5fd25c6f3f2839386","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.1214087117702,"warning":null},{"id":"02dd73192a322d1d0f8c6c1d","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.0544217687075,"warning":null},{"id":"f1461f99f24065e4929bc6f5","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.380952380952394,"warning":null},{"id":"d34742eb5d738ba73157d238","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":98,"warning":null},{"id":"e8399cdab1190c86e4e1550d","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.7875664893617,"warning":null},{"id":"5f17f8f143ab442b52c0ca93","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":-43.9833333333333,"warning":null},{"id":"c7456329253fb027797f510f","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.73333333333333,"warning":null},{"id":"70d0cd044393c24c57a06146","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.3333333333333,"warning":null},{"id":"9b42443b66c734d0f2d7c944","modelId":"qwen3-235b-a22b-instruct-2507","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 235B A22B 2507 Instruct","configuration":"Qwen3 235B A22B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.151515151515198,"warning":null},{"id":"cb58398f025c3edde1f70568","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":169.28,"warning":null},{"id":"5ecc084f491a729e9b7a74c9","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":91,"warning":null},{"id":"3a5ed8eea3db9089c7dbc5ff","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":72,"warning":null},{"id":"bb8fa7dd28b53b28a7ead28e","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":2.1540366122608225,"warning":null},{"id":"c1990682da03247d86412c31","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":0,"warning":null},{"id":"1a399e93d52acd969298026c","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":4.8,"warning":null},{"id":"acbe4339e44a4d24ab40f47c","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":313.6,"warning":null},{"id":"0558174eaf2fcd898108553e","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":85.29166666666667,"warning":null},{"id":"72ec400df622abf0993c0b92","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":78.989898989899,"warning":null},{"id":"07bdaa927df321564f2c1d84","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":15.8943466172382,"warning":null},{"id":"44d7a6991cc91d58c55ff9a6","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":51.2244897959184,"warning":null},{"id":"cabc23b6d1570087cdf40e87","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":78.8359788359788,"warning":null},{"id":"67ad8c31109848b751082bc1","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":98.4,"warning":null},{"id":"1f706dbdb38fdc0c25ef1282","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":84.2586436170213,"warning":null},{"id":"963e940e7972f41eb041f884","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":-46.5166666666667,"warning":null},{"id":"e126146d0f6fd5e675ea5077","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":22.816666666666666,"warning":null},{"id":"257a65544bb4192d4e6972b6","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":41.435185185185205,"warning":null},{"id":"e51a19f3141204938b7e3a3c","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":53.2163742690059,"warning":null},{"id":"2acb2f76d344a5a0435308de","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":7.83505154639175,"warning":null},{"id":"0c1071699ff2079d8fdebc1b","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":13.6363636363636,"warning":null},{"id":"2f2ba29d0427aaad4c86e0c6","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":11.9850187265918,"warning":null},{"id":"2209fa80f3ef69cff5c6933e","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3 235B A22B 2507 (Reasoning)","configuration":"Qwen3 235B A22B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-235b-a22b-instruct-2507-reasoning","score":0,"warning":null},{"id":"5e6dd137fbbb57f48edbb48f","modelId":"qwen3-30b-a3b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":21.6666666666667,"warning":null},{"id":"192afae643d6adf6c2b4c358","modelId":"qwen3-30b-a3b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":72.3333333333333,"warning":null},{"id":"70efb52fb8f76af5b6b52a6e","modelId":"qwen3-30b-a3b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":0,"warning":null},{"id":"fdf5b9f8d714afb06ad143b4","modelId":"qwen3-30b-a3b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":0,"warning":null},{"id":"ae488d8b4035636896c7c04b","modelId":"qwen3-30b-a3b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":0,"warning":null},{"id":"4381440088e84230d7416ca0","modelId":"qwen3-30b-a3b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":0,"warning":null},{"id":"9eba4f44d37ec442eae1c0b6","modelId":"qwen3-30b-a3b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":51.5151515151515,"warning":null},{"id":"3fef3ec80ad899c411e920db","modelId":"qwen3-30b-a3b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":61.616161616161605,"warning":null},{"id":"eeeec44dbdcbcd1fec5088ff","modelId":"qwen3-30b-a3b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":4.6,"warning":null},{"id":"0d7d919efd90d5c6c7d2d155","modelId":"qwen3-30b-a3b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":6.16,"warning":null},{"id":"ebd843a2940c41c39943f6fc","modelId":"qwen3-30b-a3b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":31.9047619047619,"warning":null},{"id":"dca37a8019d4b22d0d4f863a","modelId":"qwen3-30b-a3b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":41.4965986394558,"warning":null},{"id":"660d0d9d49749f2371fee225","modelId":"qwen3-30b-a3b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":32.1693121693122,"warning":null},{"id":"2c9b11fa4972aba196c050ec","modelId":"qwen3-30b-a3b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":50.5820105820106,"warning":null},{"id":"85fa93b4a127efea07f4ee97","modelId":"qwen3-30b-a3b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":86.26666666666671,"warning":null},{"id":"faac05e386302fdce6abc902","modelId":"qwen3-30b-a3b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":95.93333333333331,"warning":null},{"id":"a744d10897af345ffbf1637e","modelId":"qwen3-30b-a3b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":71.0438829787234,"warning":null},{"id":"bafe260a148c38a7ada467f2","modelId":"qwen3-30b-a3b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":77.6761968085106,"warning":null},{"id":"be9c4f9166e8a88c6c27dc71","modelId":"qwen3-30b-a3b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":-65.9833333333333,"warning":null},{"id":"48c5c21ee84158570d29b986","modelId":"qwen3-30b-a3b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":-51.4833333333333,"warning":null},{"id":"811a45a5a0d28bd953acd4a1","modelId":"qwen3-30b-a3b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":11.966666666666667,"warning":null},{"id":"f5dec5d5bb04db837fefe24f","modelId":"qwen3-30b-a3b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":16.183333333333334,"warning":null},{"id":"a9de0ffeb600f9f6945e64b7","modelId":"qwen3-30b-a3b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":22.2222222222222,"warning":null},{"id":"ad5665745ea3639e88c0f813","modelId":"qwen3-30b-a3b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":26.023391812865498,"warning":null},{"id":"af5c211c7fbd820d3a9078a1","modelId":"qwen3-30b-a3b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B (Non-reasoning)","configuration":"Qwen3 30B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct","score":6.81818181818182,"warning":null},{"id":"8a719f06b81ca947a876e5a9","modelId":"qwen3-30b-a3b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 30B A3B (Reasoning)","configuration":"Qwen3 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-instruct-reasoning","score":2.27272727272727,"warning":null},{"id":"e6194a0353b1168809b0b254","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.3333333333333,"warning":null},{"id":"77ce916db696117abf426908","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.333333333333297,"warning":null},{"id":"727953477da1c7bfd3dddf55","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"bbff8e891ca3ce9fb273e4e0","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.01041666666667,"warning":null},{"id":"2d5e8811e66cf17e31e93e44","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.85858585858591,"warning":null},{"id":"fdacb847dded341069a86e17","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.90454124189064,"warning":null},{"id":"9119e88a171b62aff4d1d30f","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.0612244897959,"warning":null},{"id":"04f2969c4b63ad25ed889181","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.5343915343915,"warning":null},{"id":"d2d8294eaf80d456436b1a75","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":97.5333333333333,"warning":null},{"id":"4251c0820afca56a08be5811","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.7177526595745,"warning":null},{"id":"62854b9cc7a0052762824704","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":-66.15,"warning":null},{"id":"de8c589cdba00a5c5afca9c9","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.966666666666667,"warning":null},{"id":"587a06eec8c56d9a9b6aedb0","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.233918128654999,"warning":null},{"id":"b8383df5670c6809c2eef77c","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 30B A3B 2507 Instruct","configuration":"Qwen3 30B A3B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.06060606060606,"warning":null},{"id":"df25541f05879d48387d20a6","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":5.32,"warning":null},{"id":"d6db70276e35943ca61c814e","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":56.3333333333333,"warning":null},{"id":"2321ae88075ead248809d8e1","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":61.33333333333329,"warning":null},{"id":"1f1f0c80edae6f3a634beda2","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":0.795143729768833,"warning":null},{"id":"370b5b9bebce490bc7802be4","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":0.28571428571428603,"warning":null},{"id":"b4d7a3feb92d122e78ebcd62","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":0.4,"warning":null},{"id":"810636e49e15f8ab21754d20","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":67.96,"warning":null},{"id":"82ee76c486480d1331cac8a7","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":70.7070707070707,"warning":null},{"id":"694a6ae93a990d591913241b","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":10.2873030583874,"warning":null},{"id":"7ece9316f357641df5610882","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":50.6802721088435,"warning":null},{"id":"f4e0fece0be006e5564b5eea","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":70.6878306878307,"warning":null},{"id":"48efe4f78a59711b33996eb4","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":97.6,"warning":null},{"id":"1bfc9cea4d180f9ec285a7ad","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":80.5019946808511,"warning":null},{"id":"d309ffb9a64b671bdd150c8c","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":-56.1333333333333,"warning":null},{"id":"a0c4514c75c30f131fadd06c","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":16.3,"warning":null},{"id":"97fe37c7f9321a9afbce9419","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":32.9861111111111,"warning":null},{"id":"2143491abf4fa705ee0ff478","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":28.0701754385965,"warning":null},{"id":"eb94bdb3a73cbd6888dd9b81","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":5.36082474226804,"warning":null},{"id":"7da6e30fc453a9160b7fc090","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":5.3030303030303,"warning":null},{"id":"da91fae6c59400b775705b5c","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":1.4981273408239701,"warning":null},{"id":"52eff8771abcc56b8e2ca851","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3 30B A3B 2507 (Reasoning)","configuration":"Qwen3 30B A3B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-30b-a3b-2507-reasoning","score":0,"warning":null},{"id":"74a1fb34b8502ef69002e5e6","modelId":"qwen3-32b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":0,"warning":null},{"id":"50bc16029e3f89646a8db5d5","modelId":"qwen3-32b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":19.6666666666667,"warning":null},{"id":"236503dc2ece70ff04cb2662","modelId":"qwen3-32b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":73,"warning":null},{"id":"467c0d1a8d9fbd1da7d58c18","modelId":"qwen3-32b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":0,"warning":null},{"id":"dc2bbac9236f9e228cd498ce","modelId":"qwen3-32b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":0,"warning":null},{"id":"f14f0ee78f85450317855ae0","modelId":"qwen3-32b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":0.9711289785341921,"warning":null},{"id":"32e698f91999aa66dbf0d33b","modelId":"qwen3-32b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":0.28571428571428603,"warning":null},{"id":"2881dc70ef3c09ffe8b7aaf6","modelId":"qwen3-32b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":1.6,"warning":null},{"id":"edcb1e7969f07b76a885c9b5","modelId":"qwen3-32b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":33.77,"warning":null},{"id":"ef0a55af7e9bca42a6969898","modelId":"qwen3-32b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":53.5353535353535,"warning":null},{"id":"09c94da7f63cc4f35fcaa8ef","modelId":"qwen3-32b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":66.7676767676768,"warning":null},{"id":"48c85cef75aca509424f96d8","modelId":"qwen3-32b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":4.130000000000001,"warning":null},{"id":"532667768a038808e1c06a66","modelId":"qwen3-32b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":7.41,"warning":null},{"id":"cc30d25c84e9b4477b649acc","modelId":"qwen3-32b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":31.496598639455797,"warning":null},{"id":"fe95c5c72ff58ffbfe5f6823","modelId":"qwen3-32b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":36.3265306122449,"warning":null},{"id":"c013a49962eaeffd601049e4","modelId":"qwen3-32b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":28.7830687830688,"warning":null},{"id":"366e308a05ab543e85742e62","modelId":"qwen3-32b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":54.6031746031746,"warning":null},{"id":"ba8a236b0d865aa499356fe1","modelId":"qwen3-32b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":86.8666666666667,"warning":null},{"id":"0b986f53a1a49d882b80d396","modelId":"qwen3-32b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":96.0666666666667,"warning":null},{"id":"29346831ae44b3d310de7180","modelId":"qwen3-32b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 32B (Non-reasoning)","configuration":"Qwen3 32B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct","score":72.7476728723404,"warning":null},{"id":"239624201d84931ac65ea002","modelId":"qwen3-32b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":79.8454122340426,"warning":null},{"id":"864945a3e7a9e02b314d1f06","modelId":"qwen3-32b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":-50.3666666666667,"warning":null},{"id":"7986bc6ac6e65fd08a7872d5","modelId":"qwen3-32b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":17.433333333333334,"warning":null},{"id":"4f069f41ce299bb5371886aa","modelId":"qwen3-32b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":35.9953703703704,"warning":null},{"id":"49a11e10c53668cb78a23aff","modelId":"qwen3-32b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":29.8245614035088,"warning":null},{"id":"66b3f17222441498b28217a6","modelId":"qwen3-32b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":5.36082474226804,"warning":null},{"id":"0473d12b293f64f0a0d6891e","modelId":"qwen3-32b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":3.03030303030303,"warning":null},{"id":"600f05c1a0f31fa75134db9e","modelId":"qwen3-32b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":5.2434456928839,"warning":null},{"id":"051c088ca8dff8e607638eba","modelId":"qwen3-32b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3 32B (Reasoning)","configuration":"Qwen3 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-32b-instruct-reasoning","score":0,"warning":null},{"id":"3fb8f25022ef3577afaf7e10","modelId":"qwen3-4b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":22.3333333333333,"warning":null},{"id":"37ddd191ba8a7706c22ede89","modelId":"qwen3-4b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":0,"warning":null},{"id":"65f0ddfbfd712968af3079af","modelId":"qwen3-4b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 4B (Non-reasoning)","configuration":"Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct","score":39.7979797979798,"warning":null},{"id":"91155b95de942a7455e43323","modelId":"qwen3-4b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":52.2222222222222,"warning":null},{"id":"6a3477ae7039d94212e11fec","modelId":"qwen3-4b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 4B (Non-reasoning)","configuration":"Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct","score":3.32,"warning":null},{"id":"603076858d0e2948a85c123d","modelId":"qwen3-4b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":4.42,"warning":null},{"id":"87904f931dc2ffb73802afcf","modelId":"qwen3-4b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":32.5170068027211,"warning":null},{"id":"7eda1f86bc4fd88090113e80","modelId":"qwen3-4b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 4B (Non-reasoning)","configuration":"Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct","score":23.2804232804233,"warning":null},{"id":"e0bb53c1cf9d782756209f3d","modelId":"qwen3-4b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":46.4550264550264,"warning":null},{"id":"f7989a07f2d2c06d46c14671","modelId":"qwen3-4b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 4B (Non-reasoning)","configuration":"Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct","score":84.2666666666667,"warning":null},{"id":"b4374f7d2c16022ba21b76a9","modelId":"qwen3-4b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":93.3333333333333,"warning":null},{"id":"629a36632de288da7f71fb5d","modelId":"qwen3-4b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 4B (Non-reasoning)","configuration":"Qwen3 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct","score":58.59375,"warning":null},{"id":"f46c6d830769a6c4e5c99df7","modelId":"qwen3-4b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":69.5644946808511,"warning":null},{"id":"32c6def6785bf325dc4f87ed","modelId":"qwen3-4b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 4B (Reasoning)","configuration":"Qwen3 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-instruct-reasoning","score":19.0058479532164,"warning":null},{"id":"b8304749e7352f154e474794","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.3333333333333,"warning":null},{"id":"15a1c599f7692208475bf7ec","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.3333333333333,"warning":null},{"id":"56d2377da4fb0704eb1ac2d3","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"fa65e4aae5f09070d5102301","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.71153846153847,"warning":null},{"id":"9eb087a7abc8d597044257ee","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.717171717171695,"warning":null},{"id":"4d43e7f62daa161d728080e2","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.49490268767377,"warning":null},{"id":"c10be8c55f9429f748beb6b9","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.5374149659864,"warning":null},{"id":"9b204881e5210ee72604f1d6","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.6719576719577,"warning":null},{"id":"78d1af6fd684f7bf363b664e","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.2373670212766,"warning":null},{"id":"49c265837d27d2d2bf8e4efd","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-61.9333333333333,"warning":null},{"id":"cd9a6a9b734e8a05a1d62106","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.9,"warning":null},{"id":"be24a4e27b32bad48a266dab","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.6081871345029,"warning":null},{"id":"37afc57a11a68563f8f2ef72","modelId":"qwen3-4b-instruct-2507","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 4B 2507 Instruct","configuration":"Qwen3 4B 2507 Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.545454545454549,"warning":null},{"id":"96ebe1b6fbf2e0d343e3a601","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":82.6666666666667,"warning":null},{"id":"e5545ac10fac2b93cda828d1","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":37.3333333333333,"warning":null},{"id":"73326f7b5805389fbe29ad4d","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":0,"warning":null},{"id":"e11def07dfa9e995c73e1bac","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":66.6666666666667,"warning":null},{"id":"15a62e80d61f958a5049ebf7","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":6.1631139944393,"warning":null},{"id":"8942418d355ebf98c867b34b","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":49.7959183673469,"warning":null},{"id":"db6ac46ae5700dbe44444c52","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":64.1269841269841,"warning":null},{"id":"7b9744f8d20c3175afe7b407","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":74.3101728723404,"warning":null},{"id":"b3a3f857a9b6c30c32f5e781","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":-60.2666666666667,"warning":null},{"id":"fa0a36c238b7bcee88d55771","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":12.533333333333333,"warning":null},{"id":"9cd71124c6978766c9d6d511","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":25.4385964912281,"warning":null},{"id":"a6478507eb9c1b358f3b4a3c","modelId":"qwen3-4b-thinking-2507","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 4B 2507 (Reasoning)","configuration":"Qwen3 4B 2507 (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-4b-2507-instruct-reasoning","score":1.51515151515152,"warning":null},{"id":"6aec1ffa155c03633da8ddbc","modelId":"qwen3-8b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":0,"warning":null},{"id":"77e57e37efff0800a4ac52e2","modelId":"qwen3-8b","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":24.3333333333333,"warning":null},{"id":"0ee738972827562727062ab4","modelId":"qwen3-8b","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":19,"warning":null},{"id":"810d82044b4b72e1d77b2229","modelId":"qwen3-8b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":0,"warning":null},{"id":"fff734316823fab5da7d9839","modelId":"qwen3-8b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":0,"warning":null},{"id":"470d367830b121eff8410527","modelId":"qwen3-8b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":0.3326067293298021,"warning":null},{"id":"553d1656777ad9dd0b5fbd2e","modelId":"qwen3-8b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":0,"warning":null},{"id":"a71931fca8ea24e37311ccff","modelId":"qwen3-8b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":0,"warning":null},{"id":"ebca0d32e1876250f09920a9","modelId":"qwen3-8b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":1.4000000000000001,"warning":null},{"id":"7855991e76b29f4da4df3c62","modelId":"qwen3-8b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":-48.21,"warning":null},{"id":"9607a0d57cad509ef28ad301","modelId":"qwen3-8b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":45.1515151515152,"warning":null},{"id":"3f121713f6dd60580e421928","modelId":"qwen3-8b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":58.8888888888889,"warning":null},{"id":"f444a8a8c61adea63cb4e058","modelId":"qwen3-8b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":1.95,"warning":null},{"id":"07a8885a5754031517a535e8","modelId":"qwen3-8b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":3.8899999999999997,"warning":null},{"id":"bf779bb3e13563b175f0959e","modelId":"qwen3-8b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":28.571428571428598,"warning":null},{"id":"c0a1791dc8cd498b60abbce8","modelId":"qwen3-8b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":33.469387755102,"warning":null},{"id":"a1b9be1ec2a8c4aba2017b1b","modelId":"qwen3-8b","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":20.2116402116402,"warning":null},{"id":"1c4376001c7ea7ed1c0c72d6","modelId":"qwen3-8b","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":40.634920634920604,"warning":null},{"id":"3598c03b7e5214c93e9c72ca","modelId":"qwen3-8b","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":82.8,"warning":null},{"id":"c030d0371f9e059436a36972","modelId":"qwen3-8b","benchmarkId":"aa-math-500","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":90.4,"warning":null},{"id":"bbb6369c5b97ed3eb6f04b90","modelId":"qwen3-8b","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":64.2785904255319,"warning":null},{"id":"3085e36820ee68bd4a102b22","modelId":"qwen3-8b","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":74.30186170212771,"warning":null},{"id":"7d82a974e643a8d0e079950e","modelId":"qwen3-8b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":-73.95,"warning":null},{"id":"ab22b03c0bef227a91b34986","modelId":"qwen3-8b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":-64.7333333333333,"warning":null},{"id":"f261ae29c7c086c8c29aff0b","modelId":"qwen3-8b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":11.083333333333334,"warning":null},{"id":"5702032aa67edf32f4667d95","modelId":"qwen3-8b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":13.633333333333333,"warning":null},{"id":"f14ac71d6c0d2283748b9d25","modelId":"qwen3-8b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":27.8935185185185,"warning":null},{"id":"c2a1189537534941d9d8a6cf","modelId":"qwen3-8b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":24.8538011695906,"warning":null},{"id":"a2753d2f3b8d5e1cdf181a56","modelId":"qwen3-8b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":27.7777777777778,"warning":null},{"id":"733f2508843a636f98ba5701","modelId":"qwen3-8b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":4.74226804123711,"warning":null},{"id":"e64c7478502eac8baee26e39","modelId":"qwen3-8b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 8B (Non-reasoning)","configuration":"Qwen3 8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct","score":2.27272727272727,"warning":null},{"id":"25b23d6d1d47b76b25448251","modelId":"qwen3-8b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":2.27272727272727,"warning":null},{"id":"90aa8fa4b05c718ddfd307c8","modelId":"qwen3-8b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":2.2471910112359503,"warning":null},{"id":"5c8bf720b7385ebdeb270f69","modelId":"qwen3-8b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3 8B (Reasoning)","configuration":"Qwen3 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8b-instruct-reasoning","score":0,"warning":null},{"id":"a5e53cf2346ee45d0c582775","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.999999999999996,"warning":null},{"id":"8639ec30f3a041013cccdecc","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.6666666666667,"warning":null},{"id":"fdd2d19f3143bc2b076e5415","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"2ae61a29f11e72de77a36c9d","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.616161616161605,"warning":null},{"id":"eec049bcf666ad4a78e12134","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.84615384615385,"warning":null},{"id":"2c80de3cac3aa9982c6fbe1a","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.6530612244898,"warning":null},{"id":"fe3e00ef3fb3a6b9d569105e","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.3174603174603,"warning":null},{"id":"2fb790d8ab3d3930b2000933","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":89.3333333333333,"warning":null},{"id":"1b62d586150de22843349748","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.63663563829789,"warning":null},{"id":"26f52aa824dc36a26b030308","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-50.7333333333333,"warning":null},{"id":"9b267017f4daf531a326f8af","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.400000000000002,"warning":null},{"id":"76caa5bcdaa62e0399c1cbd7","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":34.5029239766082,"warning":null},{"id":"282d2dd5d0f2b0feec3a9bbb","modelId":"qwen3-coder-30b-a3b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 Coder 30B A3B Instruct","configuration":"Qwen3 Coder 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.151515151515198,"warning":null},{"id":"8ece51f27224d12114df7c4a","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.3333333333333,"warning":null},{"id":"0de3521cedd608973feca387","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":45.6666666666667,"warning":null},{"id":"b7bb1715f0274733e3390c12","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"cfa076638384c0501b932837","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":61.818181818181806,"warning":null},{"id":"1ce4d095b186c8dbb2109774","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.54124189063948,"warning":null},{"id":"e9ccf3989be626311c527af4","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.476190476190496,"warning":null},{"id":"c826b7e819def2b96de0b1c1","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.518518518518505,"warning":null},{"id":"69c8dc453e9cd5a94a8bdbac","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-math-500","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":94.19999999999999,"warning":null},{"id":"c490527d77a38dba93e58b40","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.7898936170213,"warning":null},{"id":"0b2034bca18006f3172a087a","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-22.2,"warning":null},{"id":"13d73650a0ce5aadc74813ae","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.683333333333332,"warning":null},{"id":"7c7d9e8d1215597f71eb1e20","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.5672514619883,"warning":null},{"id":"cef9782dc0d168133b8a20ff","modelId":"qwen3-coder-480b-a35b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 Coder 480B A35B Instruct","configuration":"Qwen3 Coder 480B A35B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-480b-a35b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.939393939393902,"warning":null},{"id":"f40b768ea92561cdbff4a1e7","modelId":"qwen3-coder-next","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":403.62,"warning":null},{"id":"4023e37a7772337e40c94a66","modelId":"qwen3-coder-next","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":47,"warning":null},{"id":"c0813bddfc181d56d1e8b8a5","modelId":"qwen3-coder-next","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.0593550691766298,"warning":null},{"id":"b74549c8fd9f209dea3e7bc5","modelId":"qwen3-coder-next","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f590ce5d0d46d0cce0006d8e","modelId":"qwen3-coder-next","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.6,"warning":null},{"id":"7fe0a24c4173fe182703271c","modelId":"qwen3-coder-next","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":513.94,"warning":null},{"id":"b7b3a1fbc5ff2942f395c61f","modelId":"qwen3-coder-next","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.39583333333333,"warning":null},{"id":"f5eee82bc106b4fb51416f3e","modelId":"qwen3-coder-next","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.7373737373737,"warning":null},{"id":"1806e532fa8659866761b296","modelId":"qwen3-coder-next","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.1482854494903,"warning":null},{"id":"4fdb286556f7a859f007abc5","modelId":"qwen3-coder-next","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.2380952380952,"warning":null},{"id":"6fb3b7bf91c4f655e0f9e5f6","modelId":"qwen3-coder-next","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":-62.4,"warning":null},{"id":"9dd3ea750813737b771bb546","modelId":"qwen3-coder-next","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.150000000000002,"warning":null},{"id":"e73b78f68f384f26a35c1f23","modelId":"qwen3-coder-next","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":36.2268518518519,"warning":null},{"id":"70885ee7997cb7f5413f9175","modelId":"qwen3-coder-next","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.5321637426901,"warning":null},{"id":"bb4142bd1725e8cb824ec398","modelId":"qwen3-coder-next","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.36082474226804,"warning":null},{"id":"1c252534d9d58d5a8dc603c1","modelId":"qwen3-coder-next","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.181818181818198,"warning":null},{"id":"0851b3e49721b62cf6fdcad2","modelId":"qwen3-coder-next","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.2022471910112,"warning":null},{"id":"48676396dd68515d0e746b37","modelId":"qwen3-coder-next","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Qwen3 Coder Next","configuration":"Qwen3 Coder Next; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-coder-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"379f3e03c5a75e45500083ec","modelId":"qwen3-max","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.6666666666667,"warning":null},{"id":"d0ee451e3d70f7acbd7160ad","modelId":"qwen3-max","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":75,"warning":null},{"id":"41701e2d8384671332f59b80","modelId":"qwen3-max","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":50,"warning":null},{"id":"873b3fc87079fb7d99451bb6","modelId":"qwen3-max","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":43,"warning":null},{"id":"315935ae2fb0b8d9506630ce","modelId":"qwen3-max","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.857142857142857,"warning":null},{"id":"618ec7df142e33dee8639477","modelId":"qwen3-max","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"4d4b86792ad7edf434ee2ac5","modelId":"qwen3-max","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.28645833333333,"warning":null},{"id":"1ab71f78c9db94e51405eddb","modelId":"qwen3-max","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.3636363636364,"warning":null},{"id":"c2fdb1293b42e7e45cbd183b","modelId":"qwen3-max","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.3636363636364,"warning":null},{"id":"18a603f0cb554eff82286b14","modelId":"qwen3-max","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.1482854494903,"warning":null},{"id":"380908933be5913820054566","modelId":"qwen3-max","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.9091751621872,"warning":null},{"id":"879326ff889e832b42641611","modelId":"qwen3-max","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.0272108843537,"warning":null},{"id":"c69c24d53e30040f4c162f6f","modelId":"qwen3-max","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.1496598639456,"warning":null},{"id":"3659c33ae93ad5a386c4057d","modelId":"qwen3-max","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.7195767195767,"warning":null},{"id":"eb581aed1e808fa587fdb05b","modelId":"qwen3-max","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":65.0793650793651,"warning":null},{"id":"08a91161451f412505e30032","modelId":"qwen3-max","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.83477393617021,"warning":null},{"id":"2b16892f62e2b462f8dead01","modelId":"qwen3-max","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.11735372340429,"warning":null},{"id":"520b99ee426c18dd7f263df5","modelId":"qwen3-max","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":-42.7833333333333,"warning":null},{"id":"c010e27462d3d72e8f96c438","modelId":"qwen3-max","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":-43.55,"warning":null},{"id":"4bbb94159adbed7fb9c2fd95","modelId":"qwen3-max","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.2,"warning":null},{"id":"e392854706b6be7d554ebf84","modelId":"qwen3-max","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.416666666666668,"warning":null},{"id":"12c287fe88bcaa562fa38631","modelId":"qwen3-max","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.7485380116959,"warning":null},{"id":"8e7b03c9d5c0d8dea1a9ebd9","modelId":"qwen3-max","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.2690058479532,"warning":null},{"id":"2bdf79ef92caebbcd2a7db59","modelId":"qwen3-max","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 Max","configuration":"Qwen3 Max; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.4545454545455,"warning":null},{"id":"c942c318ad1475140efa1310","modelId":"qwen3-max","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 Max (Preview)","configuration":"Qwen3 Max (Preview); Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-max-preview; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.6969696969697,"warning":null},{"id":"bb0a0f706bcfacfc8bfbcb0b","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":66.3333333333333,"warning":null},{"id":"39324bf234452599e097140c","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.66666666666669,"warning":null},{"id":"0bebebb1c0d3d8c78f37ab1f","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"b6d6b7200e8855577137d44b","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.8383838383838,"warning":null},{"id":"e3f87d78cfb9be15308174cc","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.645968489341979,"warning":null},{"id":"2dec37e2c4f881a390992157","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.6598639455782,"warning":null},{"id":"7902cc4d3470d6a216813a4e","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.3597883597884,"warning":null},{"id":"02254150ed5397242e653e1c","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.88164893617021,"warning":null},{"id":"2263e85fdc24b011ab25738f","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-59.5,"warning":null},{"id":"c5b9a37885e2ba0e8b3361a4","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.25,"warning":null},{"id":"7ba187ae862e408223ddd229","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.6374269005848,"warning":null},{"id":"32276506d0d1e55800cb28e1","modelId":"qwen3-next-80b-a3b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 Next 80B A3B Instruct","configuration":"Qwen3 Next 80B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.57575757575758,"warning":null},{"id":"8d89e459916593927b09df0d","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":84.3333333333333,"warning":null},{"id":"6f45310222eab628315727a3","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":63.66666666666671,"warning":null},{"id":"7cd92797a65aa1b3546d034a","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":0,"warning":null},{"id":"55316631ea2bf8aab9614283","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":128.46,"warning":null},{"id":"b85018d49d57cf2e8050c2ad","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":81.94270833333333,"warning":null},{"id":"9c43e3dd521102061d12b83c","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":75.8585858585858,"warning":null},{"id":"a632fb2495b87e7d0e9989e0","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":12.5579240037071,"warning":null},{"id":"0fd93cc59b58680c7cd8160d","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":60.6802721088435,"warning":null},{"id":"8725748b17d9c049319bc806","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":78.4126984126984,"warning":null},{"id":"739cc3ae08400828dd7e34ad","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":82.3803191489362,"warning":null},{"id":"9e71a0ae62a979c3a73340e0","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":-51.4166666666667,"warning":null},{"id":"52855ab9f5e4787364e91f02","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":19.03333333333333,"warning":null},{"id":"8d68334084e3a4fcd1a96926","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":41.5204678362573,"warning":null},{"id":"cde368032b5e946e8016b135","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":6.18556701030928,"warning":null},{"id":"8bdb9faad99b2ad6caa3e59e","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":9.84848484848485,"warning":null},{"id":"53f02abbe152e698259000de","modelId":"qwen3-next-80b-a3b-thinking","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3 Next 80B A3B (Reasoning)","configuration":"Qwen3 Next 80B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-next-80b-a3b-reasoning","score":6.741573033707859,"warning":null},{"id":"66e1a70cfbb06ba377a24906","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":52.3333333333333,"warning":null},{"id":"6e1994fe220192d4f648bc00","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"cb01da84527cd626a88d69a5","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"207e511763a5de76dd47573d","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":62.020202020202,"warning":null},{"id":"a52027c783bbeb79f5ee0053","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.6339202965709,"warning":null},{"id":"d2f6bbcba92a230b64695fba","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.156462585034,"warning":null},{"id":"a95606570c645f6594bdd527","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.2222222222222,"warning":null},{"id":"edacfa4d7f499e0fbdf77f56","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.4817154255319,"warning":null},{"id":"23cb074182ab575c2d998986","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.4913294797688,"warning":null},{"id":"96e0b755ad70d1c1e0ae9354","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-69.2666666666667,"warning":null},{"id":"c67b438ddcafcd62ab222e69","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.333333333333334,"warning":null},{"id":"44fe6f288cb432b5362daa11","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.374269005848,"warning":null},{"id":"da85a712a2a928f8a0093ebe","modelId":"qwen3-omni-30b-a3b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 Omni 30B A3B Instruct","configuration":"Qwen3 Omni 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-omni-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.51515151515152,"warning":null},{"id":"f78f496de1e6c875cbbb78f7","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.6666666666667,"warning":null},{"id":"ffe8f1f73796df547f259cd9","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.6666666666667,"warning":null},{"id":"debf36b3ff3a8afa229b244f","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"25e5c20c242f402a50e10b5c","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.2121212121212,"warning":null},{"id":"d83fa0fcf5570fd9e34ce182","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.62650602409639,"warning":null},{"id":"8768eb7e5b3055cae4949a8c","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.653061224489804,"warning":null},{"id":"2b8146bfa8634abeab45959f","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":59.365079365079396,"warning":null},{"id":"94e4190c5036843765fa2396","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.313829787234,"warning":null},{"id":"5e3a87793c1033e86adf1f22","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.5722543352601,"warning":null},{"id":"1648013dead8d8b7e3ccc5b7","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-52.6666666666667,"warning":null},{"id":"d60c61b3b2041a2b188d80da","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.3,"warning":null},{"id":"f872fe1f414e47d1ea0273fc","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.0877192982456,"warning":null},{"id":"a131db2197b9e48ad031e4c0","modelId":"qwen3-vl-235b-a22b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 VL 235B A22B Instruct","configuration":"Qwen3 VL 235B A22B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-235b-a22b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.81818181818182,"warning":null},{"id":"1c9c62a5f652b687c1908560","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.3333333333333,"warning":null},{"id":"ce2d0da396f1bad697146064","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"6bec8c4c730c568a6b901187","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":69.4949494949495,"warning":null},{"id":"5b21bb8a42d86dea2ee62b68","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.302131603336419,"warning":null},{"id":"c8b57cbe46a5a65b9861cf57","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.1292517006803,"warning":null},{"id":"8edf38eeca7d9796ddf0e53e","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.6190476190476,"warning":null},{"id":"7e5b735a0b87bf634aa6456a","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.4461436170213,"warning":null},{"id":"f977ec4daaaca0a87c90c23a","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":62.1387283236994,"warning":null},{"id":"3d168904477178b1120a9f97","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-63.4166666666667,"warning":null},{"id":"d248649872ed9e5e376ce5b7","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.616666666666667,"warning":null},{"id":"4f52b0d13a697ef0d9e0d6b8","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.0058479532164,"warning":null},{"id":"a6bf0592349468ca8091cc9a","modelId":"qwen3-vl-30b-a3b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 VL 30B A3B Instruct","configuration":"Qwen3 VL 30B A3B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.06060606060606,"warning":null},{"id":"a4c021f4566a015478f0f03d","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":82.3333333333333,"warning":null},{"id":"d52a0dfa14f73e70c99eece0","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":0,"warning":null},{"id":"2cde6a74401fadb647be7203","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":72.02020202020199,"warning":null},{"id":"c5ba2f971eba83fe4e4cd779","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":8.943466172381841,"warning":null},{"id":"652d921806364ca7568196c2","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":45.1020408163265,"warning":null},{"id":"f954718896851a2f87b497bd","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":69.73544973544969,"warning":null},{"id":"e26f300586ddb51b29981834","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":80.6682180851064,"warning":null},{"id":"4a35028ffd4c5b0bd78f6503","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":61.849710982659,"warning":null},{"id":"bc8e92ab40faf29f9b4d4907","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":-57.6333333333333,"warning":null},{"id":"d6a9a7a1d698ec0f45749b16","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":16.933333333333334,"warning":null},{"id":"bbcfcfc5a57e209927c93a91","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":19.8830409356725,"warning":null},{"id":"76bffec8df092259b0e2775e","modelId":"qwen3-vl-30b-a3b-thinking","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 VL 30B A3B (Reasoning)","configuration":"Qwen3 VL 30B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-30b-a3b-reasoning","score":5.3030303030303,"warning":null},{"id":"db2809519c432f97129f7f51","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.3333333333333,"warning":null},{"id":"a0b8bc5d77324f5d4a6184e7","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"f932e350d02270272a4699b5","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.0707070707071,"warning":null},{"id":"7346e21ff6039e7851f3cf0c","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.811862835959221,"warning":null},{"id":"5cdf63b8df011f0ff87fb63b","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.1836734693878,"warning":null},{"id":"6fe04ab988c966bcc2cc8917","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.4285714285714,"warning":null},{"id":"d82e08723042532db00fbf64","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.05585106382979,"warning":null},{"id":"8435069f6714805a5bdcb8b3","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.27745664739881,"warning":null},{"id":"47a57d2d2d5bebced36433b6","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-63.3666666666667,"warning":null},{"id":"4eeb25e06a76789d08deef48","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.833333333333334,"warning":null},{"id":"937f3deb6ec8769352a2f898","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.2397660818713,"warning":null},{"id":"788b5a4a18e66ca329d23ccd","modelId":"qwen3-vl-32b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 VL 32B Instruct","configuration":"Qwen3 VL 32B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.33333333333333,"warning":null},{"id":"2e55ba0226be74935acdc897","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":84.6666666666667,"warning":null},{"id":"a4094dd3667ddcfa668fe7e5","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":55.3333333333333,"warning":null},{"id":"d582a1e6ea952bdaa149da51","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":0,"warning":null},{"id":"73203773c02c1803cc040e47","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":73.3333333333333,"warning":null},{"id":"90a23b0985eb873d6da4fbac","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":10.1482854494903,"warning":null},{"id":"a535d8fa36ad7b6f19efccfb","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":59.3877551020408,"warning":null},{"id":"8f152c8b037230620cdf6612","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":73.7566137566138,"warning":null},{"id":"0833bcf585498456cc9d0564","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":81.781914893617,"warning":null},{"id":"72340574c11bdf92b50a5b2f","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":63.410404624277504,"warning":null},{"id":"79f1b817a8bf3ed314092aa1","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":-52.5166666666667,"warning":null},{"id":"6175e18fd0e66b677d391766","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":16.966666666666665,"warning":null},{"id":"91c0534b0a62fc3b88904ed9","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":45.6140350877193,"warning":null},{"id":"96aa9eb5f454deea7dce2ec5","modelId":"qwen3-vl-32b-thinking","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 VL 32B (Reasoning)","configuration":"Qwen3 VL 32B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-32b-reasoning","score":7.57575757575758,"warning":null},{"id":"fdb356426f563f83f2ea4a01","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":37,"warning":null},{"id":"ecb1b4b1085894574d2ede22","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.000000000000002,"warning":null},{"id":"3965d02265b8cd2f3fb53b31","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"048a32652d70416c7f411620","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.0707070707071,"warning":null},{"id":"3411bba2fa134fafbba0bb3a","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":3.6144578313252995,"warning":null},{"id":"0cf5d2bad9846828c180e48e","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.8367346938775,"warning":null},{"id":"18d7349015832a249624fd5c","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.994708994709,"warning":null},{"id":"656a47fb67b2a19fef4e31ec","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.3976063829787,"warning":null},{"id":"5d8e3953b819ae9ecb77cb33","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.8728323699422,"warning":null},{"id":"edb4937174129ff14d58735a","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-75.9666666666667,"warning":null},{"id":"1d167a258f9e93c210f259ed","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.866666666666665,"warning":null},{"id":"1724ab099aea8608156fc956","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.3918128654971,"warning":null},{"id":"ee07bf88a91e4495261d3439","modelId":"qwen3-vl-4b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 VL 4B Instruct","configuration":"Qwen3 VL 4B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"a2016ac901838a157d54e1eb","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":25.6666666666667,"warning":null},{"id":"ecb9cf10e11fd05f5d7d8502","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":21.3333333333333,"warning":null},{"id":"5819e3fd8452136833f35718","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":0,"warning":null},{"id":"57b61b1da9a2c9e3a3acb85f","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":49.3939393939394,"warning":null},{"id":"e959c69c4467ef5bf9ef4072","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":4.5875810936051895,"warning":null},{"id":"45e7c304d53e99009f735f7b","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":36.5986394557823,"warning":null},{"id":"500c1e917c2929e5059ee4d0","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":31.957671957672,"warning":null},{"id":"574bba91af7fe1baced3aeb5","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":69.9800531914894,"warning":null},{"id":"5937d980c6d4c9fbd919d5c9","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":52.0231213872832,"warning":null},{"id":"362b4a14df3cdc84afbc71f4","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":-68.3833333333333,"warning":null},{"id":"be0266afffdd5c4ea60000a9","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":12.216666666666667,"warning":null},{"id":"ca54d8ad8a33c18db1d947ab","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":15.4970760233918,"warning":null},{"id":"bd7f1bd0f1e410e21b9c5b0a","modelId":"qwen3-vl-4b-thinking","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 VL 4B (Reasoning)","configuration":"Qwen3 VL 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-4b-reasoning","score":1.51515151515152,"warning":null},{"id":"a73112ccabc790faf47fc2ac","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-aime-2025","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.3333333333333,"warning":null},{"id":"5bc12952f1891bb0379aa9c7","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.6666666666667,"warning":null},{"id":"1a3f4483f198c01a748200ab","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"6b9f2418d52839baa7a340b2","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.727272727272705,"warning":null},{"id":"2616cf11c1caf39e02f50c19","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.68767377201112,"warning":null},{"id":"fa9589c76a8ee5e83bd7774c","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.312925170068,"warning":null},{"id":"f7d1263f0ece557ae1ad4f4d","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-livecodebench","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.2275132275132,"warning":null},{"id":"68113c3030360a2eb3a4d15a","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-mmlu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.58377659574471,"warning":null},{"id":"adbe6ad89c794eda3ff9c001","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.3410404624277,"warning":null},{"id":"36cb2e462ea1e247f9f8aa78","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-51.8833333333333,"warning":null},{"id":"1cd485ab01810c1389e0f4d6","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.483333333333334,"warning":null},{"id":"5cc01773941f38b549389bb9","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.2397660818713,"warning":null},{"id":"79091a97cdd8c46eb8496f8f","modelId":"qwen3-vl-8b-instruct","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3 VL 8B Instruct","configuration":"Qwen3 VL 8B Instruct; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.27272727272727,"warning":null},{"id":"cc7b91f0e6eb4ba4a8af2d29","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":30.666666666666696,"warning":null},{"id":"6070f6657824dddd8b696352","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":33.3333333333333,"warning":null},{"id":"70e25eae318c15611c90329a","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":0.28571428571428603,"warning":null},{"id":"96396a220f050f9c610a29b7","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":57.8787878787879,"warning":null},{"id":"405328f5d7cbbe8ce56d7b17","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":3.84615384615385,"warning":null},{"id":"8ab8eba51f6cc002414295be","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":39.8639455782313,"warning":null},{"id":"4ecac423e4e3317e14f55ad5","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":35.3439153439153,"warning":null},{"id":"f865331efe53ccc13479e7f9","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":74.85039893617021,"warning":null},{"id":"cffa767ba0a1e16e2db0137e","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":56.647398843930596,"warning":null},{"id":"bc245f925590016dc55fa53a","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":-52.3,"warning":null},{"id":"bce789fbb94ae0834475783e","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":20.433333333333334,"warning":null},{"id":"a0a268dca21cea302e2bdc4b","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":22.5146198830409,"warning":null},{"id":"a6c738ba4e4e4ebda37da9cf","modelId":"qwen3-vl-8b-thinking","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3 VL 8B (Reasoning)","configuration":"Qwen3 VL 8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-vl-8b-reasoning","score":3.78787878787879,"warning":null},{"id":"c79f005bd2c96b3dfac33ffa","modelId":"qwen3.5-0.8b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":8,"warning":null},{"id":"6526c4eb6102f766c4c1062f","modelId":"qwen3.5-0.8b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":9,"warning":null},{"id":"d0bdc890c6710da451dc8600","modelId":"qwen3.5-0.8b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":0,"warning":null},{"id":"60b7d90d07f72044a849d7a4","modelId":"qwen3.5-0.8b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":0,"warning":null},{"id":"6896ec6701b9ac93574cca70","modelId":"qwen3.5-0.8b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":-369.73,"warning":null},{"id":"9e15fe0442b62ce945e0ed3a","modelId":"qwen3.5-0.8b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":-388.11,"warning":null},{"id":"3f1ad55633cc7a58851cd2c9","modelId":"qwen3.5-0.8b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":23.6363636363636,"warning":null},{"id":"3cf0f7313bdb3bb306cf325b","modelId":"qwen3.5-0.8b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":11.1111111111111,"warning":null},{"id":"00910596b4c53cfe534ac0eb","modelId":"qwen3.5-0.8b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":5.09731232622799,"warning":null},{"id":"1c91f30bb30bc36833d46e2a","modelId":"qwen3.5-0.8b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":1.06580166821131,"warning":null},{"id":"dedc8bed9daad7fda2892e8f","modelId":"qwen3.5-0.8b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":21.5646258503401,"warning":null},{"id":"b20cc339a7f979c41b1c036d","modelId":"qwen3.5-0.8b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":21.4965986394558,"warning":null},{"id":"b6fa5f20a494fa9ecedcc7ee","modelId":"qwen3.5-0.8b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":25.6647398843931,"warning":null},{"id":"8aad14fd7116de2c2cfdeeef","modelId":"qwen3.5-0.8b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":25.838150289017296,"warning":null},{"id":"f04c11d1b1e78ba49a626d3e","modelId":"qwen3.5-0.8b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":-88.5833333333333,"warning":null},{"id":"937ba73b35a8ad5eb48af2c0","modelId":"qwen3.5-0.8b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":-54.5166666666667,"warning":null},{"id":"aff956fbf6dee32a521c6a67","modelId":"qwen3.5-0.8b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":4.983333333333333,"warning":null},{"id":"9eb4731b4ea4ea831d5585e4","modelId":"qwen3.5-0.8b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":4.233333333333333,"warning":null},{"id":"c43d78ed5ec033460060cf6c","modelId":"qwen3.5-0.8b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":65.2046783625731,"warning":null},{"id":"ced6927af98788f28fec7064","modelId":"qwen3.5-0.8b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":47.6608187134503,"warning":null},{"id":"fe58f57624f6bfd77e4a529b","modelId":"qwen3.5-0.8b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":0,"warning":null},{"id":"f848e97ce77653c8903415bd","modelId":"qwen3.5-0.8b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":0,"warning":null},{"id":"0557d91e1328599eddeedaa0","modelId":"qwen3.5-0.8b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.5 0.8B (Non-reasoning)","configuration":"Qwen3.5 0.8B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b-non-reasoning","score":0.374531835205993,"warning":null},{"id":"6c477d7a9ed5f790dea8bffe","modelId":"qwen3.5-0.8b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.5 0.8B (Reasoning)","configuration":"Qwen3.5 0.8B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-0-8b","score":0,"warning":null},{"id":"9797fd599cf89cf3d36d814a","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":497.34,"warning":null},{"id":"7570031879abad667cd8e969","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":61.33333333333329,"warning":null},{"id":"fd77aa95b9c29c06a6b169cc","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":76.3333333333333,"warning":null},{"id":"f1e2c1476fab5dba5c841fa6","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":5.481116230292565,"warning":null},{"id":"1434c5402cedf4ac2b4da110","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":0.857142857142857,"warning":null},{"id":"4e4780dee2a81d7e9171107a","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":0.571428571428571,"warning":null},{"id":"dc76f6471810c741012070f3","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":3.4000000000000004,"warning":null},{"id":"0e00f6e43c6a400549e216cd","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":709.24,"warning":null},{"id":"051aa80bb154784791487055","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":814.01,"warning":null},{"id":"4f7ad991264657d538d51832","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":85.88020833333334,"warning":null},{"id":"89cbf8eb01c2697fd35c27a3","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":89.13020833333334,"warning":null},{"id":"712427354a54a3d7928dc9a0","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":82.7272727272727,"warning":null},{"id":"9b2b75d84b4f7c03bc1bc751","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":85.6565656565657,"warning":null},{"id":"cd859ec8921f41c450fecef8","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":15.940685820203901,"warning":null},{"id":"91a6a7e402d1e1cd935677c2","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":25.2085264133457,"warning":null},{"id":"9e4d9fd1a2574d6a5ae7703f","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":50.8163265306122,"warning":null},{"id":"ad4b717ca4a435c0c7b6bf1a","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":75.7142857142857,"warning":null},{"id":"b5b2eded240cc84231225e2e","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":70.2890173410405,"warning":null},{"id":"633304c9460583efb1a9d123","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":74.971098265896,"warning":null},{"id":"13945cfbd8edd56e3c0efd6b","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":-54.9666666666667,"warning":null},{"id":"c7e647c2a4d2f6427799cdf3","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":-41.5,"warning":null},{"id":"2680ab741c6710c6308ef76b","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":19.116666666666667,"warning":null},{"id":"dca6e42139718103b2bc6718","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":24.366666666666667,"warning":null},{"id":"515eeda369aef75d1b3b90a2","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":39.6990740740741,"warning":null},{"id":"c3bfff836c0b94cd19e1ab01","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":84.5029239766082,"warning":null},{"id":"a3b2b0715095d52d58ac063b","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":93.56725146198829,"warning":null},{"id":"3dd3e9d4d8dcd772354db4ac","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":10.3092783505155,"warning":null},{"id":"591bd1ccff3f9c16ac8904ce","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":15.2577319587629,"warning":null},{"id":"c6b2a6fcceb391533b9ccb6d","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":29.545454545454504,"warning":null},{"id":"dbcbfe85658c4ab468fa9ca3","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":31.060606060606098,"warning":null},{"id":"a29311df8979dbac227d11e3","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.5 122B A10B (Non-reasoning)","configuration":"Qwen3.5 122B A10B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b-non-reasoning","score":47.1910112359551,"warning":null},{"id":"928dfca300ef1cd20891bdf3","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":47.565543071160995,"warning":null},{"id":"393b7eb4b3c229026330a49e","modelId":"qwen3.5-122b-a10b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.5 122B A10B (Reasoning)","configuration":"Qwen3.5 122B A10B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-122b-a10b","score":0,"warning":null},{"id":"70ec9dd22c9c9b201c491822","modelId":"qwen3.5-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":64,"warning":null},{"id":"b3259eb613b9008d13bfb2fd","modelId":"qwen3.5-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":77.6666666666667,"warning":null},{"id":"52b6b150197775cfe5c91b8e","modelId":"qwen3.5-27b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"3e51ac2e0bcfc616f60fdbcb","modelId":"qwen3.5-27b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":0.8530612244897959,"warning":null},{"id":"c1822f4f16b355f03b6b133e","modelId":"qwen3.5-27b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":83.82291666666666,"warning":null},{"id":"3a55ccda11cad6f5dc83eb08","modelId":"qwen3.5-27b","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":86.76041666666667,"warning":null},{"id":"4a6498ef78e583463717f583","modelId":"qwen3.5-27b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":84.24242424242419,"warning":null},{"id":"c89518098dd5dccb090f454a","modelId":"qwen3.5-27b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":85.7575757575758,"warning":null},{"id":"78ef779c3fcd29db093d3844","modelId":"qwen3.5-27b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":13.9481000926784,"warning":null},{"id":"4f49cf2685886f6ec4a83da8","modelId":"qwen3.5-27b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":23.9110287303058,"warning":null},{"id":"b21250eb9c4c108f04af43f0","modelId":"qwen3.5-27b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":46.9387755102041,"warning":null},{"id":"9dc289d96ddc9b2f19861742","modelId":"qwen3.5-27b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":75.57823129251699,"warning":null},{"id":"6386fdf31a291d72b80c8bda","modelId":"qwen3.5-27b","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":35.4990583804143,"warning":null},{"id":"8e7a86835734ba4cfd98b502","modelId":"qwen3.5-27b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":70,"warning":null},{"id":"4ae5bffa4ac662eed074e463","modelId":"qwen3.5-27b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":75.028901734104,"warning":null},{"id":"40ff7c5a40a01e803bb1cd8c","modelId":"qwen3.5-27b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":-47.6666666666667,"warning":null},{"id":"55cdfeeabac6031082ecb7d4","modelId":"qwen3.5-27b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":-43.9833333333333,"warning":null},{"id":"6fe875ee0279a564bcd15685","modelId":"qwen3.5-27b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":15.683333333333332,"warning":null},{"id":"8e5f70a0d9b318aca939682c","modelId":"qwen3.5-27b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":20.666666666666668,"warning":null},{"id":"dd74023e1d36e1931e867b89","modelId":"qwen3.5-27b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":87.1345029239766,"warning":null},{"id":"d04bd91f3abbf17a30df7293","modelId":"qwen3.5-27b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":93.859649122807,"warning":null},{"id":"3535372b77b4f10a2aae2f53","modelId":"qwen3.5-27b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 27B (Non-reasoning)","configuration":"Qwen3.5 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b-non-reasoning","score":31.818181818181802,"warning":null},{"id":"87314125e9b141d0f793b80a","modelId":"qwen3.5-27b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 27B (Reasoning)","configuration":"Qwen3.5 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-27b","score":32.5757575757576,"warning":null},{"id":"a7c022f322decc2d18aacdf5","modelId":"qwen3.5-2b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":14.000000000000002,"warning":null},{"id":"340251b326024b7a69c3bf39","modelId":"qwen3.5-2b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":21,"warning":null},{"id":"d6189f1ea70832a412f06340","modelId":"qwen3.5-2b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":0,"warning":null},{"id":"98ea8b4e38a123443727892b","modelId":"qwen3.5-2b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":0,"warning":null},{"id":"2fc909a093ee84fc1354d083","modelId":"qwen3.5-2b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":-366.49,"warning":null},{"id":"430f9d066315c614ba751865","modelId":"qwen3.5-2b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":-45.72,"warning":null},{"id":"1dc539f5096fa85e6092c50a","modelId":"qwen3.5-2b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":43.8383838383838,"warning":null},{"id":"1efb5443137143050e7ed17e","modelId":"qwen3.5-2b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":45.5555555555556,"warning":null},{"id":"248d8fef7e7a90ae0e4e2295","modelId":"qwen3.5-2b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":4.95829471733086,"warning":null},{"id":"7bd48532b9a85b06d476758b","modelId":"qwen3.5-2b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":2.5949953660797,"warning":null},{"id":"67088604dbeb65c302440afb","modelId":"qwen3.5-2b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":29.115646258503403,"warning":null},{"id":"a54f47fe6f74069debb77e36","modelId":"qwen3.5-2b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":31.496598639455797,"warning":null},{"id":"4dfcabb11e935b9d77c34307","modelId":"qwen3.5-2b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":42.6589595375723,"warning":null},{"id":"f426f85b359f8e2fe9579751","modelId":"qwen3.5-2b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":43.0057803468208,"warning":null},{"id":"0aba9d436f02774f10805fd1","modelId":"qwen3.5-2b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":-83.1333333333333,"warning":null},{"id":"4ebd8fc4a160d868162f9b65","modelId":"qwen3.5-2b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":-60.1833333333333,"warning":null},{"id":"05ae862b6429e595ff8d2b84","modelId":"qwen3.5-2b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":7.383333333333334,"warning":null},{"id":"30a8a589ff73daf9b30088ef","modelId":"qwen3.5-2b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":8.066666666666666,"warning":null},{"id":"60624d170a81e918f4879de3","modelId":"qwen3.5-2b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":81.5789473684211,"warning":null},{"id":"3f24913e7ca1a2f286dbdce8","modelId":"qwen3.5-2b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":69.0058479532164,"warning":null},{"id":"20c954bf67b034614b5f10b0","modelId":"qwen3.5-2b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":3.78787878787879,"warning":null},{"id":"050daa4bef2d46ebbccd176d","modelId":"qwen3.5-2b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":3.78787878787879,"warning":null},{"id":"84335a8349dd1b9d4fa8d0d2","modelId":"qwen3.5-2b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.5 2B (Non-reasoning)","configuration":"Qwen3.5 2B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b-non-reasoning","score":0,"warning":null},{"id":"31afa3a5ee4f8ddeb7c59185","modelId":"qwen3.5-2b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.5 2B (Reasoning)","configuration":"Qwen3.5 2B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-2b","score":2.9962546816479403,"warning":null},{"id":"518b3def44f7c280d9c2b344","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":63,"warning":null},{"id":"a6bb3e77e8089018687551e0","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":72,"warning":null},{"id":"0c4fc56597384246a3c510ea","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":0.571428571428571,"warning":null},{"id":"bca3df4c2afb6534b7ab2365","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":0.857142857142857,"warning":null},{"id":"c11008b7d4895bc958e9a160","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":608.32,"warning":null},{"id":"52dac326b14ad8274e860038","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":82.421875,"warning":null},{"id":"5b6def33141d74b866cef6e3","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":86.44270833333333,"warning":null},{"id":"ffa5a2f4c3d54a3e22f9ef1e","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":81.9191919191919,"warning":null},{"id":"1a4eac252ecd64ac0ec9fac3","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":84.5454545454545,"warning":null},{"id":"f4bcadbf6a497277a67985f9","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":13.438368860055599,"warning":null},{"id":"e47d94c493c3d8efa18fed87","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":21.0379981464319,"warning":null},{"id":"4b2a00224e4e77c869bb3525","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":44.4897959183673,"warning":null},{"id":"6429639babbd562974064492","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":72.5170068027211,"warning":null},{"id":"7eb289685e2f9e25fba62890","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":21.5160075329567,"warning":null},{"id":"3db7471df67389c685b88dbc","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":69.19075144508669,"warning":null},{"id":"71b06d1e69d3282835e9802d","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":72.6589595375723,"warning":null},{"id":"f638bda28b9ca9b449743a3e","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":-63.0166666666667,"warning":null},{"id":"eebdfc2b391ba014eacbdef5","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":-48.1,"warning":null},{"id":"cce73caaf54d5224ad13f618","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":15.716666666666669,"warning":null},{"id":"1ef6e17ab8bf9c8ccebbad85","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":20.133333333333333,"warning":null},{"id":"956b26763c1cc19ed48e62ee","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":86.2573099415205,"warning":null},{"id":"ef170a6bbbf2157e05ffd6b9","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":89.1812865497076,"warning":null},{"id":"0fe2b426e7191799fd5ad7b0","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":4.948453608247419,"warning":null},{"id":"23b432199e2d041e6d0644b0","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":10.6060606060606,"warning":null},{"id":"f796f8cd0abdbccf18ef98f6","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 35B A3B (Reasoning)","configuration":"Qwen3.5 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b","score":26.5151515151515,"warning":null},{"id":"ab038381c16df8d1e510bd8a","modelId":"qwen3.5-35b-a3b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.5 35B A3B (Non-reasoning)","configuration":"Qwen3.5 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-35b-a3b-non-reasoning","score":40.8239700374532,"warning":null},{"id":"7a85addb1de948317e2bcefc","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":562.4,"warning":null},{"id":"acf644377690a4fb775a2880","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":15.339233038348098,"warning":null},{"id":"cd62f190593255d149665e0c","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":64.3333333333333,"warning":null},{"id":"dfb69334a23f689f666068a2","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":77.3333333333333,"warning":null},{"id":"135958e9d5c43321eb1d8dbb","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":6.886670023070867,"warning":null},{"id":"75a41b8db9b3590c183008fc","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":0.857142857142857,"warning":null},{"id":"c99d476c058b4a89123fa646","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":1.70612244897959,"warning":null},{"id":"83ea18f7ab43ad1f8b8722de","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":30.88630259623993,"warning":null},{"id":"521a42cd831cbbe5b0e2060b","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":12.2,"warning":null},{"id":"897bff0bd990995acf75035b","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":792.97,"warning":null},{"id":"082b89a396ab2c1e7104c310","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":86.515625,"warning":null},{"id":"7e0dbe0e88c6108996e40748","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":90.03645833333334,"warning":null},{"id":"911a594d4aab2c408e3fd321","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":86.0606060606061,"warning":null},{"id":"bc7c4c940825fac880dcc166","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":89.2929292929293,"warning":null},{"id":"79dce625ee6f0c2ed9c8245b","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":19.8331788693234,"warning":null},{"id":"ba25e9f3d4d437738db9b852","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":28.962001853568097,"warning":null},{"id":"dfd2256c044cc016cd99b343","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":51.632653061224495,"warning":null},{"id":"a3d4369851bced4c6f9e534d","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":78.7755102040816,"warning":null},{"id":"b395c95b8a65bb01a89133e0","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":34.0866290018832,"warning":null},{"id":"86340fbe3bc82bdd6b9d0f7d","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":52.6589595375723,"warning":null},{"id":"721334010c9ce0613c28fb6c","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":77.28323699421969,"warning":null},{"id":"263ebc7960a8a4cb108bd07c","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":-37.9,"warning":null},{"id":"6a3a254fa11a7c706555dff3","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":-30.75,"warning":null},{"id":"8ff8cfd69dedb3a094f26ced","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":24.516666666666666,"warning":null},{"id":"028e196a79298f6522c3e81f","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":30.8,"warning":null},{"id":"b223f7629fdf1388f9bed2a9","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":44.7916666666667,"warning":null},{"id":"3d1b866c502ba3c2b8352a96","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":83.9181286549708,"warning":null},{"id":"a0cc3fbb604ccc3d914119e9","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":95.6140350877193,"warning":null},{"id":"6f9e10389b596e289112755a","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":13.4020618556701,"warning":null},{"id":"1164bcd5a5aa3712ceea50a7","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 397B A17B (Non-reasoning)","configuration":"Qwen3.5 397B A17B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b-non-reasoning","score":35.6060606060606,"warning":null},{"id":"4ce0b8becc8fdf948549228c","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":40.9090909090909,"warning":null},{"id":"642a262fba6eeac8cd6c33a1","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":51.310861423221,"warning":null},{"id":"ee547f794b6f2b28ae9a8028","modelId":"qwen3.5-397b-a17b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.5 397B A17B (Reasoning)","configuration":"Qwen3.5 397B A17B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-397b-a17b","score":0,"warning":null},{"id":"0ea8b264473dc425326ac154","modelId":"qwen3.5-4b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":34.6666666666667,"warning":null},{"id":"897ab118ac1342a924d6112e","modelId":"qwen3.5-4b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":63,"warning":null},{"id":"053597e91079b350b9341d09","modelId":"qwen3.5-4b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":0,"warning":null},{"id":"92b1bbe4bab1236ee4b0635f","modelId":"qwen3.5-4b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":0,"warning":null},{"id":"e91f85c76ca2395e8d16bcd7","modelId":"qwen3.5-4b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":71.2121212121212,"warning":null},{"id":"af08f36e4be4659803ff0480","modelId":"qwen3.5-4b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":77.0707070707071,"warning":null},{"id":"47153236c0ed4b07d1500751","modelId":"qwen3.5-4b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":7.970342910101951,"warning":null},{"id":"642772c958d7cd65f97835cb","modelId":"qwen3.5-4b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":9.91658943466172,"warning":null},{"id":"dce15135abcbf97e2e75e164","modelId":"qwen3.5-4b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":33.265306122449005,"warning":null},{"id":"9f54a0ec86853c159d448797","modelId":"qwen3.5-4b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":51.9727891156463,"warning":null},{"id":"a9e1f9b1809bcd29819a8408","modelId":"qwen3.5-4b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":62.080924855491304,"warning":null},{"id":"a1af20461c5aa98121c402cd","modelId":"qwen3.5-4b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":65.3757225433526,"warning":null},{"id":"e7f8e4594adf9ffcf3ed6e8a","modelId":"qwen3.5-4b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":-74.7333333333333,"warning":null},{"id":"ebe98a5e899b1dcb02f76998","modelId":"qwen3.5-4b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":-58.35,"warning":null},{"id":"ed5f4a1b29b3009ad0ceeed1","modelId":"qwen3.5-4b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":11.216666666666667,"warning":null},{"id":"247c36abe2eb6e790cc417b2","modelId":"qwen3.5-4b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":15.116666666666667,"warning":null},{"id":"d28e385d70e803a765f7fc5c","modelId":"qwen3.5-4b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":87.719298245614,"warning":null},{"id":"50e3c30102eb314283834c71","modelId":"qwen3.5-4b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":92.1052631578947,"warning":null},{"id":"f08a6b8acbba22210e4198c1","modelId":"qwen3.5-4b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":4.32989690721649,"warning":null},{"id":"b2ed8afbd8136fb37d1c54bd","modelId":"qwen3.5-4b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":6.8041237113402095,"warning":null},{"id":"d6fe8b1efef057e8b67e8412","modelId":"qwen3.5-4b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":11.363636363636399,"warning":null},{"id":"3f84ada02ed713f5b4df5e8b","modelId":"qwen3.5-4b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":18.181818181818198,"warning":null},{"id":"5776b4e242b36d7888d749b7","modelId":"qwen3.5-4b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.5 4B (Non-reasoning)","configuration":"Qwen3.5 4B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b-non-reasoning","score":21.3483146067416,"warning":null},{"id":"6b92b80ee9f40384f85cdd13","modelId":"qwen3.5-4b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.5 4B (Reasoning)","configuration":"Qwen3.5 4B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-4b","score":25.842696629213503,"warning":null},{"id":"9b94cf268e2ac4021bbd1e3d","modelId":"qwen3.5-9b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":340.67,"warning":null},{"id":"987f147949570e5f033630d1","modelId":"qwen3.5-9b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":46,"warning":null},{"id":"35ab75c2d9610fd9d6afd707","modelId":"qwen3.5-9b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":70,"warning":null},{"id":"b320268ce575b376a611419d","modelId":"qwen3.5-9b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":1.0473867343065657,"warning":null},{"id":"51cceb0f761a19498f07f43b","modelId":"qwen3.5-9b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":0.571428571428571,"warning":null},{"id":"93e0ee1fea4ee49c409f7f30","modelId":"qwen3.5-9b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":0.28571428571428603,"warning":null},{"id":"3e9c76292ea762a413f6d9f3","modelId":"qwen3.5-9b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":6.6000000000000005,"warning":null},{"id":"5cb7f5c3363e1df057d62795","modelId":"qwen3.5-9b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":419.52,"warning":null},{"id":"c2a233e0c283bc6f848ab7f8","modelId":"qwen3.5-9b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":78.5858585858586,"warning":null},{"id":"9f70624165f1bfb5534f12d8","modelId":"qwen3.5-9b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":80.60606060606061,"warning":null},{"id":"4d9e271b62a791eb7aa355fb","modelId":"qwen3.5-9b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":9.36051899907322,"warning":null},{"id":"91f1614a6570830ed8587c39","modelId":"qwen3.5-9b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":14.921223354958299,"warning":null},{"id":"405394a5fe3ae5cd4ffb5987","modelId":"qwen3.5-9b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":37.8231292517007,"warning":null},{"id":"b268faf0293cf763ce33c6d5","modelId":"qwen3.5-9b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":66.734693877551,"warning":null},{"id":"bee5fa058648448b9abb72f4","modelId":"qwen3.5-9b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":66.7630057803468,"warning":null},{"id":"3271ebc988e5db1d40b6ee92","modelId":"qwen3.5-9b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":69.2485549132948,"warning":null},{"id":"70c7c662608af8f632662f38","modelId":"qwen3.5-9b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":-70.7,"warning":null},{"id":"5686f65f304b0ae905401fe5","modelId":"qwen3.5-9b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":-53.4666666666667,"warning":null},{"id":"37143e5c779711b1534cd2c0","modelId":"qwen3.5-9b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":13.983333333333334,"warning":null},{"id":"bbb1c791dbb001405e0b3785","modelId":"qwen3.5-9b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":16.416666666666664,"warning":null},{"id":"6afd5cfd76652d0557349eb7","modelId":"qwen3.5-9b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":29.5138888888889,"warning":null},{"id":"35b08aa160794079a7760a9a","modelId":"qwen3.5-9b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":85.0877192982456,"warning":null},{"id":"b6ad97d2e6b84fb5bfff38af","modelId":"qwen3.5-9b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":86.8421052631579,"warning":null},{"id":"1ea96f71f7b8acf0e5c92135","modelId":"qwen3.5-9b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":7.01030927835052,"warning":null},{"id":"5a9c2b5e5ea876f9675bbd0c","modelId":"qwen3.5-9b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":18.181818181818198,"warning":null},{"id":"0eb44799cf54377c7b779992","modelId":"qwen3.5-9b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":24.2424242424242,"warning":null},{"id":"9beec241281354166f7f4cdf","modelId":"qwen3.5-9b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.5 9B (Non-reasoning)","configuration":"Qwen3.5 9B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b-non-reasoning","score":21.3483146067416,"warning":null},{"id":"71565ab7921a7bff37fd8271","modelId":"qwen3.5-9b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":29.213483146067396,"warning":null},{"id":"c5d2163d3e991ece89cf3fae","modelId":"qwen3.5-9b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.5 9B (Reasoning)","configuration":"Qwen3.5 9B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-5-9b","score":0.505050505050505,"warning":null},{"id":"c717dab863d029554a546f07","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":52,"warning":null},{"id":"b94f1d2228e900a6c2df44ae","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"4828255fcd543335b20e8178","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.24242424242419,"warning":null},{"id":"5584d2dd8f205f5dfc703398","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.5996292863762696,"warning":null},{"id":"7ed7cbd32da3164f960078dd","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.9591836734694,"warning":null},{"id":"f03d2e976501ce3b51c34286","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":64.73988439306359,"warning":null},{"id":"ce044099dd1a13b985395372","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":-65.0666666666667,"warning":null},{"id":"de3e87fa1b992b3a661f0744","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.649999999999999,"warning":null},{"id":"4305e3ee57626ca07733f9a1","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.5029239766082,"warning":null},{"id":"0e201fd8e88804cf432985b7","modelId":"qwen3.5-omni-flash","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Flash","configuration":"Qwen3.5 Omni Flash; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.33333333333333,"warning":null},{"id":"c82039a80845dab3eadfacd8","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.66666666666671,"warning":null},{"id":"2e75afb685d3a3295079aa9b","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.5632653061224491,"warning":null},{"id":"0f1a5a784a71c9e6676ed3fd","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.6262626262626,"warning":null},{"id":"4d99610acf5319007bb0aeea","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.874884151992601,"warning":null},{"id":"ef689030a893d35e97f6f3d2","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":51.156462585034,"warning":null},{"id":"4db50a094c083851e4fab68c","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.52023121387279,"warning":null},{"id":"62d333463f4ce6b2f02693ee","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":-13.1166666666667,"warning":null},{"id":"047dd20376f5c65c5e1a899a","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.433333333333334,"warning":null},{"id":"75ce6997f9a7d90791f02ea5","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.3040935672515,"warning":null},{"id":"08f342f50a739b3ac9528085","modelId":"qwen3.5-omni-plus","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.5 Omni Plus","configuration":"Qwen3.5 Omni Plus; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/qwen3-5-omni-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.2121212121212,"warning":null},{"id":"0e93ed1c49c2f34e7933895f","modelId":"qwen3.6-27b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":815.82,"warning":null},{"id":"9407d449239c38058843fede","modelId":"qwen3.6-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":66.6666666666667,"warning":null},{"id":"e7a488abcf660ae289f3a24f","modelId":"qwen3.6-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":77.3333333333333,"warning":null},{"id":"69a809df0aeb1b2ecf654222","modelId":"qwen3.6-27b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":11.045223588713588,"warning":null},{"id":"f8f423f444a94a81a9c1a0a8","modelId":"qwen3.6-27b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":0.857142857142857,"warning":null},{"id":"536bcb7f59ec5628dcee1463","modelId":"qwen3.6-27b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":1.1428571428571401,"warning":null},{"id":"448b575dcd1acd2a60fbb4cd","modelId":"qwen3.6-27b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":11,"warning":null},{"id":"b7eb05d6f119d2c5556a1ea3","modelId":"qwen3.6-27b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":963.94,"warning":null},{"id":"9195bde130476ec1bf79e2dd","modelId":"qwen3.6-27b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":985.63,"warning":null},{"id":"9f3b5a5650f0028b75e6627a","modelId":"qwen3.6-27b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":82.9292929292929,"warning":null},{"id":"ab44e7ebb36589f6af440d58","modelId":"qwen3.6-27b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":84.24242424242419,"warning":null},{"id":"511a91b9c33047739bda7f2e","modelId":"qwen3.6-27b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":15.060240963855401,"warning":null},{"id":"f8e8bc36191c7551f6b5cc4d","modelId":"qwen3.6-27b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":23.0769230769231,"warning":null},{"id":"5a0cb4fe77de2668f3a7851d","modelId":"qwen3.6-27b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":45.7142857142857,"warning":null},{"id":"147cff2e7b77dbe7a3ab9a02","modelId":"qwen3.6-27b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":67.5510204081633,"warning":null},{"id":"01d2afdc9f3d7512a2797f1c","modelId":"qwen3.6-27b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":71.73410404624279,"warning":null},{"id":"1f330c240a2e147daf36750e","modelId":"qwen3.6-27b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":74.6242774566474,"warning":null},{"id":"3388bc6451fc50d48d32f80f","modelId":"qwen3.6-27b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":-53.2,"warning":null},{"id":"a670029bf642d55d2f29a869","modelId":"qwen3.6-27b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":-20.0166666666667,"warning":null},{"id":"8c5b10be361e3b52f4be8621","modelId":"qwen3.6-27b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":16.716666666666665,"warning":null},{"id":"3ccb2b0ee8fb9a12d8d0b958","modelId":"qwen3.6-27b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":19.6,"warning":null},{"id":"db7758f0177deb0686de0fde","modelId":"qwen3.6-27b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":42.8240740740741,"warning":null},{"id":"ef1ca5e4c1d0640e21632a3b","modelId":"qwen3.6-27b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":93.56725146198829,"warning":null},{"id":"e8af44f1827fd2b34c9c1f57","modelId":"qwen3.6-27b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":94.15204678362569,"warning":null},{"id":"3bdf7a8c944fd163c8dd47a8","modelId":"qwen3.6-27b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":9.27835051546392,"warning":null},{"id":"b302521289de79298a4c4a11","modelId":"qwen3.6-27b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":16.7010309278351,"warning":null},{"id":"319dd147ad8ab4256d9e920a","modelId":"qwen3.6-27b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":21.2121212121212,"warning":null},{"id":"3e197f94e3c030d6c4c768f4","modelId":"qwen3.6-27b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":34.8484848484849,"warning":null},{"id":"c4e9d38c2e0b145c31db613b","modelId":"qwen3.6-27b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.6 27B (Non-reasoning)","configuration":"Qwen3.6 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b-non-reasoning","score":51.310861423221,"warning":null},{"id":"c666c345c763a37020abfa2b","modelId":"qwen3.6-27b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":60.674157303370805,"warning":null},{"id":"44ce799a8b6dfb3736957ecf","modelId":"qwen3.6-27b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.6 27B (Reasoning)","configuration":"Qwen3.6 27B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-27b","score":0,"warning":null},{"id":"0fcd517f6551a379405a5f87","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":712.84,"warning":null},{"id":"fa3d8f5b093a68f35e1410fa","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":64.3333333333333,"warning":null},{"id":"fc1010f44b620abdf26e9a2e","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":71.6666666666667,"warning":null},{"id":"2e3c0a8de99217d5789e6006","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":5.2085656145058845,"warning":null},{"id":"5e00d1d331c11537440b8cc4","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":0,"warning":null},{"id":"7ade4f0426c6245254429d02","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":0.28571428571428603,"warning":null},{"id":"6744996241ae0d03122039af","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":6.6000000000000005,"warning":null},{"id":"9b4717431639aed364df845a","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":847.36,"warning":null},{"id":"ce2f11141cd23dee77b05aff","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":892.41,"warning":null},{"id":"94ac74e00d580eaeef1daa5e","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":81.7171717171717,"warning":null},{"id":"22f962a63a0b5ff81b811a76","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":84.14141414141419,"warning":null},{"id":"2140cb653135c824c77c5a52","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":13.9017608897127,"warning":null},{"id":"65e97e60b864e0aeedbb1d77","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":22.242817423540302,"warning":null},{"id":"39d90e1de15d33cb62e706b4","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":36.190476190476204,"warning":null},{"id":"4082640dc825a629a68d77ca","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":64.3537414965986,"warning":null},{"id":"e3238199ea24918db578f48a","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":71.0404624277457,"warning":null},{"id":"f021164defbe4c063630dc02","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":75.028901734104,"warning":null},{"id":"eacdde9b4ef2d608c6c5db66","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":-60.0166666666667,"warning":null},{"id":"bd2c4bd715ab130b5a7baa31","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":-22.1833333333333,"warning":null},{"id":"60d26e69fe34430556065e73","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":16.716666666666665,"warning":null},{"id":"794bebda292e3704b05c0a0d","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":18.816666666666666,"warning":null},{"id":"ed5c9e27d2893228f1baa185","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":36.5740740740741,"warning":null},{"id":"309240dbc037d56fbcb6b7b9","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":85.0877192982456,"warning":null},{"id":"5fd472847cba72707397312a","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":95.3216374269006,"warning":null},{"id":"ec8ec75f74ad4b9274f062ed","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":5.36082474226804,"warning":null},{"id":"2c7fa9fd818e2d0e84c1b39f","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":9.27835051546392,"warning":null},{"id":"948bee28800b4658a01a25ed","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":25.7575757575758,"warning":null},{"id":"034278450a24f60e27015b1a","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":34.8484848484849,"warning":null},{"id":"e2153c542957bfe883a91411","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.6 35B A3B (Non-reasoning)","configuration":"Qwen3.6 35B A3B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b-non-reasoning","score":41.5730337078652,"warning":null},{"id":"34bbe6472870bee2df06ea3d","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":44.9438202247191,"warning":null},{"id":"50d69af85898178490384a19","modelId":"qwen3.6-35b-a3b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.6 35B A3B (Reasoning)","configuration":"Qwen3.6 35B A3B (Reasoning); Model source: https://artificialanalysis.ai/models/qwen3-6-35b-a3b","score":0,"warning":null},{"id":"93638a70fa1bd2eb2973ebf3","modelId":"qwen3.6-plus","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":78.3333333333333,"warning":null},{"id":"d22f57bac206134b0895962b","modelId":"qwen3.6-plus","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.85714285714286,"warning":null},{"id":"f358b69750895cfdd85b6237","modelId":"qwen3.6-plus","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":990.26,"warning":null},{"id":"0ea5751656e8fdb09476cde1","modelId":"qwen3.6-plus","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.1818181818182,"warning":null},{"id":"b17e297aaf4c740f79979be4","modelId":"qwen3.6-plus","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":27.8498609823911,"warning":null},{"id":"e50ff97c4425d2ebbf1223b7","modelId":"qwen3.6-plus","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.1700680272109,"warning":null},{"id":"3eb7381b382205fba1f0fe65","modelId":"qwen3.6-plus","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.9768786127168,"warning":null},{"id":"a62bce89fdb4d3abd526eb43","modelId":"qwen3.6-plus","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.883333333333333,"warning":null},{"id":"45436b8af16d8065711b3a43","modelId":"qwen3.6-plus","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":26.38333333333333,"warning":null},{"id":"21437c6d099de4a38389b43d","modelId":"qwen3.6-plus","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":97.6608187134503,"warning":null},{"id":"489ee597fc16833ca2058214","modelId":"qwen3.6-plus","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.8247422680412,"warning":null},{"id":"59391278a6dd9fcc53640072","modelId":"qwen3.6-plus","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.9393939393939,"warning":null},{"id":"08465a135acc9707f92293e4","modelId":"qwen3.6-plus","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.6 Plus","configuration":"Qwen3.6 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-6-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":61.4232209737828,"warning":null},{"id":"9ad3e1f3423393d133f7b393","modelId":"qwen3.7-max","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.75,"warning":null},{"id":"5944f1b2b6aeefb1d9c8d85b","modelId":"qwen3.7-max","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":916.31,"warning":null},{"id":"6f305c1c4e37d27e61f689a4","modelId":"qwen3.7-max","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":79,"warning":null},{"id":"2a657017fa2edaf82956db37","modelId":"qwen3.7-max","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.976890336173604,"warning":null},{"id":"870c96fc00ec0feb2f32dfb8","modelId":"qwen3.7-max","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.4285714285714,"warning":null},{"id":"297e5cb842ccbb88ca618949","modelId":"qwen3.7-max","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":45.00149209191286,"warning":null},{"id":"6e30bc47e35c2d4bf9e1eb0a","modelId":"qwen3.7-max","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.6,"warning":null},{"id":"355999f4a7a549d837d8be6d","modelId":"qwen3.7-max","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":1127.65,"warning":null},{"id":"2307b5abe75cf7cfc52c4986","modelId":"qwen3.7-max","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":92.32323232323229,"warning":null},{"id":"4c0e30a179da19a4e7f3bfa4","modelId":"qwen3.7-max","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.5004633920297,"warning":null},{"id":"1b9110161c8cc2b19d6f2080","modelId":"qwen3.7-max","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.5442176870748,"warning":null},{"id":"3a4875b885ec50b15c8693a6","modelId":"qwen3.7-max","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.4670433145009,"warning":null},{"id":"4355bc5b1225c68cd10d5191","modelId":"qwen3.7-max","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.4833333333333,"warning":null},{"id":"7baa3c2e804282809b3e549f","modelId":"qwen3.7-max","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.133333333333336,"warning":null},{"id":"9ca0c4c478ac4a52d15f4f0c","modelId":"qwen3.7-max","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.537037037037,"warning":null},{"id":"41227f76cdd36be7d5ea2c15","modelId":"qwen3.7-max","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":94.7368421052632,"warning":null},{"id":"d5af817cd3c32f49b5984987","modelId":"qwen3.7-max","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.7525773195876,"warning":null},{"id":"f157407c28400fa2fac4f930","modelId":"qwen3.7-max","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.7575757575758,"warning":null},{"id":"f2e2da41746ec055e02aecf7","modelId":"qwen3.7-max","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":74.53183520599251,"warning":null},{"id":"3196d9aa5b9ad70ac47a6b65","modelId":"qwen3.7-max","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-max; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.51515151515152,"warning":null},{"id":"326cbf0aef11fc0da897cbf3","modelId":"qwen3.7-plus","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":915.27,"warning":null},{"id":"9cedf613da4351040d1e92db","modelId":"qwen3.7-plus","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.4188790560472,"warning":null},{"id":"963077a2374e6ded6adff9ed","modelId":"qwen3.7-plus","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":73,"warning":null},{"id":"0fbe559fc5ce72b2613c6245","modelId":"qwen3.7-plus","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.391220980015497,"warning":null},{"id":"a2ed5103fa0df8f7cfd0375e","modelId":"qwen3.7-plus","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.14285714285714,"warning":null},{"id":"5a1345f8a31fdf95832c6a86","modelId":"qwen3.7-plus","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":40.55505819158461,"warning":null},{"id":"defc9a7127c75101a72a75b1","modelId":"qwen3.7-plus","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":12.2,"warning":null},{"id":"f4f38046908ddfa9a8d6df57","modelId":"qwen3.7-plus","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":767.63,"warning":null},{"id":"ccb7687f63b0c6d5ac1eecd7","modelId":"qwen3.7-plus","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":90,"warning":null},{"id":"79ac3364e000c6cd672fbd91","modelId":"qwen3.7-plus","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.6348470806302,"warning":null},{"id":"4945d0ef68fe56561ca877da","modelId":"qwen3.7-plus","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.9591836734694,"warning":null},{"id":"5d2243ff9337e95a2dd7edc4","modelId":"qwen3.7-plus","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.44444444444444,"warning":null},{"id":"b7a0d9322ace326caf2db310","modelId":"qwen3.7-plus","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.4624277456647,"warning":null},{"id":"1d6fc7a9f857efb4ea8a4d09","modelId":"qwen3.7-plus","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.08333333333333,"warning":null},{"id":"987027dbf8f699071447f03f","modelId":"qwen3.7-plus","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.516666666666666,"warning":null},{"id":"3ea57b2c6129ce72f6450114","modelId":"qwen3.7-plus","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.064814814814795,"warning":null},{"id":"85e6bb415f16c9b56e03fef9","modelId":"qwen3.7-plus","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":92.9824561403509,"warning":null},{"id":"97acc2df5e24a7cda1402355","modelId":"qwen3.7-plus","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":17.5257731958763,"warning":null},{"id":"3f7cbb594499981fc7681ee1","modelId":"qwen3.7-plus","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.969696969697,"warning":null},{"id":"85fb53fb3b36f8e29d58711e","modelId":"qwen3.7-plus","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":61.0486891385768,"warning":null},{"id":"eb42639192e6f3b2becf6ef8","modelId":"qwen3.7-plus","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-7-plus; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.01010101010101,"warning":null},{"id":"d71e3c177c4165c697168c70","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1435.33,"warning":null},{"id":"27df0e5138bd5f8d7c930041","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.3333333333333,"warning":null},{"id":"ff0811d93cce836318787010","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":57.247376545822206,"warning":null},{"id":"190e54337ed7afe809ef9033","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":20,"warning":null},{"id":"8bba6cdadf34a978f2014270","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.35899731423456,"warning":null},{"id":"1c96bec87c5b1f7f05ecaba5","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.6,"warning":null},{"id":"cdfd854b97dcf4f232ac2afe","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":1607.97,"warning":null},{"id":"a99e4925b9bb3d9b381f52ca","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":93.53535353535351,"warning":null},{"id":"1305aa179987856b5864ef92","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":42.4467099165894,"warning":null},{"id":"6ea79d9a87da1399709e4fe4","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"215c0a040e1f469b97f5f0d1","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":4.31666666666667,"warning":null},{"id":"3cb914a9e40ec3c843a24080","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":31.25,"warning":null},{"id":"845e3104f356e8046a04495f","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":54.050925925925895,"warning":null},{"id":"72e55893e37d975afe2e59af","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.0721649484536,"warning":null},{"id":"cc3f9005dc869c1b5547b4f0","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":82.0224719101124,"warning":null},{"id":"db82321f46121c82cac0b9d8","modelId":"qwen3.8-2.4t-a95b","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.8 2.4T A95B","configuration":"Qwen3.8 2.4T A95B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-2-4t-a95b; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.1111111111111,"warning":null},{"id":"d8b4b61760170d1206e4c6e6","modelId":"qwen3.8-27b","benchmarkId":"aa-aa-briefcase","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":1271.57,"warning":null},{"id":"a960887aa78c13effeb9ed2f","modelId":"qwen3.8-27b","benchmarkId":"aa-aa-briefcase","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":1375.49,"warning":null},{"id":"d81f5fc9fd9e25c4123c4b9d","modelId":"qwen3.8-27b","benchmarkId":"aa-aa-briefcase","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":897.66,"warning":null},{"id":"6af578e800c1daab160b76f1","modelId":"qwen3.8-27b","benchmarkId":"aa-aa-briefcase","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":1395.27,"warning":null},{"id":"905f04a3ecbb47ac32e9aa10","modelId":"qwen3.8-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":77.3333333333333,"warning":null},{"id":"e53a307b41f81b2286881536","modelId":"qwen3.8-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":79.6666666666667,"warning":null},{"id":"1ec9d3b365ddb11b5239f4ac","modelId":"qwen3.8-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":69.3333333333333,"warning":null},{"id":"4f4ecdab96e122ef647ed437","modelId":"qwen3.8-27b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":82,"warning":null},{"id":"a8e5c88661bc34dc747db5ff","modelId":"qwen3.8-27b","benchmarkId":"aa-automationbench-aa","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":30.103295303433768,"warning":null},{"id":"141f82568e52fdcbe672747c","modelId":"qwen3.8-27b","benchmarkId":"aa-automationbench-aa","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":40.252101015938635,"warning":null},{"id":"f396ce095a1e473065bf40bb","modelId":"qwen3.8-27b","benchmarkId":"aa-automationbench-aa","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":12.434667722683214,"warning":null},{"id":"48d688abe7c586d5b0e2289d","modelId":"qwen3.8-27b","benchmarkId":"aa-automationbench-aa","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":48.24244368224193,"warning":null},{"id":"9aaef7347f0b2756640efb27","modelId":"qwen3.8-27b","benchmarkId":"aa-critpt","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":0,"warning":null},{"id":"09d27c1b7450c541af7617e6","modelId":"qwen3.8-27b","benchmarkId":"aa-critpt","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":0,"warning":null},{"id":"db23778fb38a738162c3739d","modelId":"qwen3.8-27b","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":0.28571428571428603,"warning":null},{"id":"c008559cd080c973d1ac2371","modelId":"qwen3.8-27b","benchmarkId":"aa-critpt","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":5.42857142857143,"warning":null},{"id":"ea9dbcdb94b4a7795d5ff06e","modelId":"qwen3.8-27b","benchmarkId":"aa-enterprise-ops-gym-aa","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":44.225604297224706,"warning":null},{"id":"e18bba4f18d449fdfa6b8282","modelId":"qwen3.8-27b","benchmarkId":"aa-gdp-pdf","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":10.4,"warning":null},{"id":"f3a1906f290c902728c8edad","modelId":"qwen3.8-27b","benchmarkId":"aa-gdp-pdf","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":12.8,"warning":null},{"id":"589cbe543e0b6d6daf010e1b","modelId":"qwen3.8-27b","benchmarkId":"aa-gdp-pdf","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":5.2,"warning":null},{"id":"d153bc8eb2fe2b184c839bb8","modelId":"qwen3.8-27b","benchmarkId":"aa-gdp-pdf","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":16.400000000000002,"warning":null},{"id":"94f8fbe4059e314ef86adabc","modelId":"qwen3.8-27b","benchmarkId":"aa-gdpval-aa","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":1362.53,"warning":null},{"id":"67e3e22ab8243004e892ff1e","modelId":"qwen3.8-27b","benchmarkId":"aa-gdpval-aa","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":1382.16,"warning":null},{"id":"78d1f56ddc08962d014de2cd","modelId":"qwen3.8-27b","benchmarkId":"aa-gdpval-aa","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":1072.36,"warning":null},{"id":"9018c4b6261daf24ff7c9ed4","modelId":"qwen3.8-27b","benchmarkId":"aa-gdpval-aa","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":1419.88,"warning":null},{"id":"e78bb0eabeba03358b2ef2c5","modelId":"qwen3.8-27b","benchmarkId":"aa-gpqa-diamond","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":84.5454545454545,"warning":null},{"id":"168b0dd5cac8f57579372130","modelId":"qwen3.8-27b","benchmarkId":"aa-gpqa-diamond","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":84.5454545454545,"warning":null},{"id":"149b9aed78a8b393be1ac340","modelId":"qwen3.8-27b","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":81.8181818181818,"warning":null},{"id":"0aff5fd37ca7c568846fa608","modelId":"qwen3.8-27b","benchmarkId":"aa-gpqa-diamond","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":90.5050505050505,"warning":null},{"id":"e0eabf034a041d9241c496b0","modelId":"qwen3.8-27b","benchmarkId":"aa-humanitys-last-exam","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":14.040778498609798,"warning":null},{"id":"f506395c805c00f024f374fa","modelId":"qwen3.8-27b","benchmarkId":"aa-humanitys-last-exam","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":14.0871177015755,"warning":null},{"id":"2b5e2ca909dc461cdc38a556","modelId":"qwen3.8-27b","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":12.1408711770158,"warning":null},{"id":"90f1ca70d689589dcf28ee10","modelId":"qwen3.8-27b","benchmarkId":"aa-humanitys-last-exam","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":33.920296570899,"warning":null},{"id":"a60079b1a185507eca5f2f00","modelId":"qwen3.8-27b","benchmarkId":"aa-mlcr-aa","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":21.6666666666667,"warning":null},{"id":"aabb29256c50d756c3c18f36","modelId":"qwen3.8-27b","benchmarkId":"aa-mmmu-pro","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":73.757225433526,"warning":null},{"id":"28859d444f893d70d220fee5","modelId":"qwen3.8-27b","benchmarkId":"aa-mmmu-pro","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":74.1618497109827,"warning":null},{"id":"135f2e02482f3c68bbcd884a","modelId":"qwen3.8-27b","benchmarkId":"aa-mmmu-pro","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":69.9421965317919,"warning":null},{"id":"8f11c06586bd400f3f105b3a","modelId":"qwen3.8-27b","benchmarkId":"aa-mmmu-pro","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":76.3005780346821,"warning":null},{"id":"00aa46f0b5885d1c20a9313c","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":-26.6666666666667,"warning":null},{"id":"762bcd862dc8b06b020506b2","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":-36.15,"warning":null},{"id":"369ef6b3e1de52e3da9e761d","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":-7.95,"warning":null},{"id":"2b8713e5794a859e545bb625","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":-9.98333333333333,"warning":null},{"id":"2dc85caf2caeaba16bf2341e","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience-accuracy","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":17.133333333333333,"warning":null},{"id":"e1a5a9bdb9b815b408255881","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience-accuracy","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":18.35,"warning":null},{"id":"59fcf8705623283774907cd0","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":8.583333333333334,"warning":null},{"id":"d5c17f619aaaee0373a6bc08","modelId":"qwen3.8-27b","benchmarkId":"aa-omniscience-accuracy","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":15.583333333333332,"warning":null},{"id":"98f72127e05ee7de84bc5d32","modelId":"qwen3.8-27b","benchmarkId":"aa-scicode","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":40.046296296296305,"warning":null},{"id":"287bbebec0f8e0c9081cb90e","modelId":"qwen3.8-27b","benchmarkId":"aa-scicode","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":39.0046296296296,"warning":null},{"id":"f0b6f28261f842ffd221371f","modelId":"qwen3.8-27b","benchmarkId":"aa-scicode","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":36.2268518518519,"warning":null},{"id":"f2649d3f975b64c3e63ec33f","modelId":"qwen3.8-27b","benchmarkId":"aa-scicode","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":46.6435185185185,"warning":null},{"id":"714d8664461eb29681e9f903","modelId":"qwen3.8-27b","benchmarkId":"aa-tau3-banking","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":32.1649484536082,"warning":null},{"id":"78ec7096516c431a180ac5e2","modelId":"qwen3.8-27b","benchmarkId":"aa-tau3-banking","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":47.4226804123711,"warning":null},{"id":"ebdcbf46dd03ef58d97834a3","modelId":"qwen3.8-27b","benchmarkId":"aa-tau3-banking","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":20,"warning":null},{"id":"3d93c33197029f90b82bac70","modelId":"qwen3.8-27b","benchmarkId":"aa-tau3-banking","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":48.041237113402104,"warning":null},{"id":"2fcc8736846efc4a7d4c08cb","modelId":"qwen3.8-27b","benchmarkId":"aa-terminal-bench-science","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":0.476190476190476,"warning":null},{"id":"aea047a6af606ec898927506","modelId":"qwen3.8-27b","benchmarkId":"aa-terminal-bench-science","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":0.952380952380952,"warning":null},{"id":"8292287a3762290043074ada","modelId":"qwen3.8-27b","benchmarkId":"aa-terminal-bench-science","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":0,"warning":null},{"id":"1675c3e6288a38212d1891cc","modelId":"qwen3.8-27b","benchmarkId":"aa-terminal-bench-science","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":0.476190476190476,"warning":null},{"id":"0eb59de01577878fc7073c31","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v2-1","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":67.4157303370786,"warning":null},{"id":"88383c73a051af3aa72a6e5a","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v2-1","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":65.1685393258427,"warning":null},{"id":"06d60c6c4c5a2369cf9b0093","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v2-1","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":49.063670411985,"warning":null},{"id":"f04f49814893dc7159949e60","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v2-1","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":79.7752808988764,"warning":null},{"id":"ac18c882ce39e175a9f275bc","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v4-0","level":"low","sourceLabel":"Qwen3.8 27B (Low)","configuration":"Qwen3.8 27B (Low); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-low","score":2.52525252525253,"warning":null},{"id":"4ddbc37ffc687b1e5936f349","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v4-0","level":"medium","sourceLabel":"Qwen3.8 27B (Medium)","configuration":"Qwen3.8 27B (Medium); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-medium","score":5.05050505050505,"warning":null},{"id":"f2d3f052face4faeeda3179c","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v4-0","level":"non-reasoning","sourceLabel":"Qwen3.8 27B (Non-reasoning)","configuration":"Qwen3.8 27B (Non-reasoning); Model source: https://artificialanalysis.ai/models/qwen3-8-27b-non-reasoning","score":0,"warning":null},{"id":"bda4835dd0c2d6c2f4274e53","modelId":"qwen3.8-27b","benchmarkId":"aa-terminalbench-v4-0","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Qwen3.8 27B (Xhigh); Model source: https://artificialanalysis.ai/models/qwen3-8-27b","score":5.55555555555556,"warning":null},{"id":"6faaafbcb22c6d83fcc7b8a8","modelId":"qwen3.8-27b","benchmarkId":"arc-v2semiprivate","level":"low","sourceLabel":"Qwen3.8-27B (Low)","configuration":"Qwen3.8-27B (Low); Source model ID: alibaba-qwen3-8-27b-low; 120 tasks","score":22.77777777777778,"warning":null},{"id":"1186b47ecb6a01558f0b8b14","modelId":"qwen3.8-27b","benchmarkId":"arc-v2semiprivate","level":"medium","sourceLabel":"Qwen3.8-27B (Medium)","configuration":"Qwen3.8-27B (Medium); Source model ID: alibaba-qwen3-8-27b-medium; 120 tasks","score":13.194444444444445,"warning":null},{"id":"6bcac8a28a611d216f7c2918","modelId":"qwen3.8-27b","benchmarkId":"arc-v2semiprivate","level":"none","sourceLabel":"Qwen3.8-27B (None)","configuration":"Qwen3.8-27B (None); Source model ID: alibaba-qwen3-8-27b-none; 120 tasks","score":1.5277777777777777,"warning":null},{"id":"ecc2d0adbb7e612c9a529eed","modelId":"qwen3.8-27b","benchmarkId":"arc-v2semiprivate","level":"xhigh","sourceLabel":"Qwen3.8-27B (XHigh)","configuration":"Qwen3.8-27B (XHigh); Source model ID: alibaba-qwen3-8-27b-xhigh; 120 tasks","score":42.36111111111111,"warning":null},{"id":"ff398af3902cbbb496663108","modelId":"qwen3.8-flash-next","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":1579.93,"warning":null},{"id":"1af1c14074eae9ab67bac2a5","modelId":"qwen3.8-flash-next","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.6666666666667,"warning":null},{"id":"100eb437023dad366f563363","modelId":"qwen3.8-flash-next","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.9083218951688,"warning":null},{"id":"d7c9d2a0fd1760ad961f399e","modelId":"qwen3.8-flash-next","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.1428571428571,"warning":null},{"id":"e21724aca11f3198eedbc6e7","modelId":"qwen3.8-flash-next","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":15.6,"warning":null},{"id":"fa923c564e87543c1ce2f864","modelId":"qwen3.8-flash-next","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":1627.86,"warning":null},{"id":"e02c83aacb3a00e840a9af72","modelId":"qwen3.8-flash-next","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":92.32323232323229,"warning":null},{"id":"16d3fe3f51d36c136b12f8b5","modelId":"qwen3.8-flash-next","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":38.044485634847106,"warning":null},{"id":"653bd0d5164de68aa04fc91e","modelId":"qwen3.8-flash-next","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":79.7687861271676,"warning":null},{"id":"19beb3218856ab5fe419a3c9","modelId":"qwen3.8-flash-next","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":-9.71666666666667,"warning":null},{"id":"6cd571f906fa29b42e3b182b","modelId":"qwen3.8-flash-next","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":24.5,"warning":null},{"id":"107f8908a0e4a7c541544f2f","modelId":"qwen3.8-flash-next","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.578703703703695,"warning":null},{"id":"e600574b71b21a5cafdd9d49","modelId":"qwen3.8-flash-next","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":45.360824742268,"warning":null},{"id":"58c528b02767f71c2ef600cc","modelId":"qwen3.8-flash-next","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.1423220973783,"warning":null},{"id":"50e7110e3b20dacde3b37e1e","modelId":"qwen3.8-flash-next","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Qwen3.8-Flash-Next","configuration":"Qwen3.8-Flash-Next; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/qwen3-8-flash-next; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.2525252525253,"warning":null},{"id":"ac0ddd5b5ae8b79a4ddc80c8","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-aime-2025","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":84.6666666666667,"warning":null},{"id":"fd77bc1467dfe4486490d845","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":61.33333333333329,"warning":null},{"id":"9e8614db68b3d211029ec88c","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"841cd6af604c06ea92463ab9","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":77.90104166666667,"warning":null},{"id":"58e1dd8e3731504b07c5dfad","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.6262626262626,"warning":null},{"id":"17b50474af561818c94ee00f","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":9.870250231696021,"warning":null},{"id":"8b0760998291d4c24b0a22a4","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.9047619047619,"warning":null},{"id":"ed488162479393e23b14bace","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-livecodebench","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.5079365079365,"warning":null},{"id":"b64d48fbd23f04ac8eebe4df","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-mmlu-pro","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":81.5408909574468,"warning":null},{"id":"12f82b95be07624d144534f1","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":-52.4666666666667,"warning":null},{"id":"044431bb8ae976aa42349a89","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.2,"warning":null},{"id":"1de2302eae9a29fe88d82846","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":49.4152046783626,"warning":null},{"id":"bff8dce7b47e199123fa7411","modelId":"seed-oss-36b-instruct","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Seed-OSS-36B-Instruct","configuration":"Seed-OSS-36B-Instruct; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/seed-oss-36b-instruct; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.81818181818182,"warning":null},{"id":"a7683e908d41bad4b60c7140","modelId":"solar-mini4-260922","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":869.52,"warning":null},{"id":"47bdce7bde5f865c9139beb3","modelId":"solar-mini4-260922","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":83.3333333333333,"warning":null},{"id":"3182dcb826ff8a7d4e609921","modelId":"solar-mini4-260922","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":22.252425682190992,"warning":null},{"id":"1fa53f089bd0da231122e7c7","modelId":"solar-mini4-260922","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.42857142857143,"warning":null},{"id":"dc2aeabd39e81318cbb73ebb","modelId":"solar-mini4-260922","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":5,"warning":null},{"id":"ae3b3f1e5b08e385c4f96ffb","modelId":"solar-mini4-260922","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":1089.14,"warning":null},{"id":"6775465a72cdd9f4c9148198","modelId":"solar-mini4-260922","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.7645968489342,"warning":null},{"id":"c463459225f8ca121c7dc6c7","modelId":"solar-mini4-260922","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":-10.8166666666667,"warning":null},{"id":"0126d37ad452c2adff59ee06","modelId":"solar-mini4-260922","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.416666666666668,"warning":null},{"id":"bff4de9e9b972a0da4c3a48b","modelId":"solar-mini4-260922","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.5694444444444,"warning":null},{"id":"5b70749de87ba5e19bcbfcf4","modelId":"solar-mini4-260922","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Solar Mini 4","configuration":"Solar Mini 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-mini4; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.01010101010101,"warning":null},{"id":"6ed7251fc836a13b8bda6c2c","modelId":"solar-open-100b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":41,"warning":null},{"id":"75cb2fffbf1adac440b9c082","modelId":"solar-open-100b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":0,"warning":null},{"id":"1d3eac54d4adb7c21226a10d","modelId":"solar-open-100b","benchmarkId":"aa-global-mmlu-lite","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":81.58333333333334,"warning":null},{"id":"6cc904fe0b8921985e0fbd2c","modelId":"solar-open-100b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":65.6565656565657,"warning":null},{"id":"b239e9e2e4dc4d96acb069c2","modelId":"solar-open-100b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":10.4263206672845,"warning":null},{"id":"7c6f3969a3bfb58f59e504a9","modelId":"solar-open-100b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":57.687074829931994,"warning":null},{"id":"786415bc21ec9ec2f6bbed5e","modelId":"solar-open-100b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":-54.5,"warning":null},{"id":"c2348de9a5afccbf97deb29c","modelId":"solar-open-100b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":18.53333333333333,"warning":null},{"id":"65366e28efa37bd0fc9304a3","modelId":"solar-open-100b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":48.245614035087705,"warning":null},{"id":"f28b14f230ca174c3535e105","modelId":"solar-open-100b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Solar Open 100B (Reasoning)","configuration":"Solar Open 100B (Reasoning); Model source: https://artificialanalysis.ai/models/solar-open-100b-reasoning","score":2.27272727272727,"warning":null},{"id":"541af6b48e4dd025e681a015","modelId":"solar-open2-250b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.6666666666667,"warning":null},{"id":"e9ccf812e35cb29f130dc31f","modelId":"solar-open2-250b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.71428571428571,"warning":null},{"id":"f87e5a262feeaed2f87defb2","modelId":"solar-open2-250b","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":970.77,"warning":null},{"id":"b4701fb10f02abf0859e9d2b","modelId":"solar-open2-250b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":85.6565656565657,"warning":null},{"id":"f5a8e2ba8b2c9185ba6e9900","modelId":"solar-open2-250b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":28.544949026876697,"warning":null},{"id":"f706a5ffc62ecf11d8aa11fc","modelId":"solar-open2-250b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-1.76666666666667,"warning":null},{"id":"e1c44575202fa0dd5bbff5db","modelId":"solar-open2-250b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.833333333333332,"warning":null},{"id":"793184ef1385ac9eccdc0fef","modelId":"solar-open2-250b","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.032407407407405,"warning":null},{"id":"b72d75ebece38f24171ed19a","modelId":"solar-open2-250b","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.6494845360825,"warning":null},{"id":"b2c76ce1b70fddc006955b7a","modelId":"solar-open2-250b","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Solar Open2 250B","configuration":"Solar Open2 250B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-open2-250b; Mode from evaluator isReasoning metadata; effort budget not specified","score":44.194756554307105,"warning":null},{"id":"ce407757e7278c975f4923a2","modelId":"solar-pro3-260323","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":98.98,"warning":null},{"id":"90113e320325aad453d0207c","modelId":"solar-pro3-260323","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":32.3333333333333,"warning":null},{"id":"e1354b01691573102e86fac0","modelId":"solar-pro3-260323","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.49734242022629577,"warning":null},{"id":"2e642eb8e589b310afece9d1","modelId":"solar-pro3-260323","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"4762ee05df5eb38cf2f0a883","modelId":"solar-pro3-260323","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":1.6,"warning":null},{"id":"64973f1d0a5775767ebb7147","modelId":"solar-pro3-260323","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":264.94,"warning":null},{"id":"087acac2342dac2567f325c0","modelId":"solar-pro3-260323","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":72.42424242424241,"warning":null},{"id":"f6cc07e7f169228137b2d3cf","modelId":"solar-pro3-260323","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.2873030583874,"warning":null},{"id":"6f529a5f931819bad160b6db","modelId":"solar-pro3-260323","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":71.2018140589569,"warning":null},{"id":"cef5e234850298a63380a52c","modelId":"solar-pro3-260323","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.555555555555556,"warning":null},{"id":"4c646f44d8a2f54dd28f759a","modelId":"solar-pro3-260323","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":-53.4,"warning":null},{"id":"5a972a37cc173c738ac9b70a","modelId":"solar-pro3-260323","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":18.5,"warning":null},{"id":"c09d36e66a2605d0626b8874","modelId":"solar-pro3-260323","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.462962962962997,"warning":null},{"id":"5326f847758818bf93448cfe","modelId":"solar-pro3-260323","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":86.2573099415205,"warning":null},{"id":"b761794691d5cb615a77b75a","modelId":"solar-pro3-260323","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":8.65979381443299,"warning":null},{"id":"8b80c8dacf1f2ac3db0ea6ed","modelId":"solar-pro3-260323","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.57575757575758,"warning":null},{"id":"2163d48791cfaa3ebdc1dcd6","modelId":"solar-pro3-260323","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.9850187265918,"warning":null},{"id":"124d3b5b3e2079c76cde89fb","modelId":"solar-pro3-260323","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Solar Pro 3","configuration":"Solar Pro 3; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro-3; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"298d80d4a5399cee230158b2","modelId":"solar-pro4-260806","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":1093.32,"warning":null},{"id":"aefef63584c73c56bed65f21","modelId":"solar-pro4-260806","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":70.6666666666667,"warning":null},{"id":"df90a3ec21d4239743fb2fe4","modelId":"solar-pro4-260806","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.59339457640078,"warning":null},{"id":"24bb299713a3451eeba316c2","modelId":"solar-pro4-260806","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.28571428571429,"warning":null},{"id":"a7f6f1f0d1861502a311e2bc","modelId":"solar-pro4-260806","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":7.000000000000001,"warning":null},{"id":"be1d922acf1db598bfd5815b","modelId":"solar-pro4-260806","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":1105.73,"warning":null},{"id":"524276087f80ab9a7ec587fe","modelId":"solar-pro4-260806","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":89.0909090909091,"warning":null},{"id":"6d4c6e539b85c5b3accf2d74","modelId":"solar-pro4-260806","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.842446709916597,"warning":null},{"id":"fcf8445fb274f138d2087141","modelId":"solar-pro4-260806","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":0.25,"warning":null},{"id":"af138a32895ebcda8487a3df","modelId":"solar-pro4-260806","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":19.416666666666664,"warning":null},{"id":"8535449d0bca6e5ef4206e74","modelId":"solar-pro4-260806","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":48.032407407407405,"warning":null},{"id":"67316d10f432f15fe7c97a55","modelId":"solar-pro4-260806","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":23.298969072164898,"warning":null},{"id":"f4dea894fcffe88ddf4ea8e7","modelId":"solar-pro4-260806","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":57.30337078651681,"warning":null},{"id":"0c517d40c3a43e93fce0d9e4","modelId":"solar-pro4-260806","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/solar-pro4; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"4e4f7f66e12fa41c88b4ae62","modelId":"step-3.7-flash","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.823008849557501,"warning":null},{"id":"d03747357d4b99762e3c2c12","modelId":"step-3.7-flash","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":73.6666666666667,"warning":null},{"id":"05f9d1f4588607425107e0cd","modelId":"step-3.7-flash","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.28571428571429,"warning":null},{"id":"f3f1d0a35cc4c49e68376e7f","modelId":"step-3.7-flash","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":29.065950462548496,"warning":null},{"id":"d26847278c16adb82bf819e3","modelId":"step-3.7-flash","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":856.36,"warning":null},{"id":"aea0451a4896d102b0043d1d","modelId":"step-3.7-flash","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":80.90909090909089,"warning":null},{"id":"9eaec21698f9c763849fc748","modelId":"step-3.7-flash","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":21.408711770157602,"warning":null},{"id":"a23d7538799febe12d9a6925","modelId":"step-3.7-flash","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":67.2789115646258,"warning":null},{"id":"9006b356b7c4957dcf096d9e","modelId":"step-3.7-flash","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.2730696798493,"warning":null},{"id":"c87c7ea70afb04af4f026773","modelId":"step-3.7-flash","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":75.3179190751445,"warning":null},{"id":"f10ed77c5f2a457b4875815a","modelId":"step-3.7-flash","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":-37.2833333333333,"warning":null},{"id":"4cb090aa5d8a78b6a18f80f0","modelId":"step-3.7-flash","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":25.783333333333335,"warning":null},{"id":"31c949451c135ca26418dd9e","modelId":"step-3.7-flash","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":43.8657407407407,"warning":null},{"id":"f6ad7a16a5446adada0a8818","modelId":"step-3.7-flash","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":98.5380116959064,"warning":null},{"id":"8b38eb1018caf6cb002d9ea9","modelId":"step-3.7-flash","benchmarkId":"aa-tau3-banking","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":11.9587628865979,"warning":null},{"id":"d29adfa107dd249d7fdf48ea","modelId":"step-3.7-flash","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":35.6060606060606,"warning":null},{"id":"ee468fde2438f81f9d3a5096","modelId":"step-3.7-flash","benchmarkId":"aa-terminalbench-v2-1","level":"reasoning","sourceLabel":"Step 3.7 Flash","configuration":"Step 3.7 Flash; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-7-flash; Mode from evaluator isReasoning metadata; effort budget not specified","score":39.325842696629195,"warning":null},{"id":"4a88975d662760cd583f9e27","modelId":"step-5-preview","benchmarkId":"aa-aa-analyst-agent","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":35,"warning":null},{"id":"97ce2b12f08ed8f5d9b5e618","modelId":"step-5-preview","benchmarkId":"aa-aa-briefcase","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":1424.32,"warning":null},{"id":"ab9a28e0a53cdcab0dd1cd27","modelId":"step-5-preview","benchmarkId":"aa-apex-agents-aa","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":37.9793510324484,"warning":null},{"id":"d6bca134ef08712b9e071ad1","modelId":"step-5-preview","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":88.3333333333333,"warning":null},{"id":"d147a3085be14474ad98b1ff","modelId":"step-5-preview","benchmarkId":"aa-automationbench-aa","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.956594190962875,"warning":null},{"id":"4adc2d2f72d7512bcefe05d2","modelId":"step-5-preview","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.8571428571429,"warning":null},{"id":"17a82096430134a4793bf474","modelId":"step-5-preview","benchmarkId":"aa-enterprise-ops-gym-aa","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":47.17994628469114,"warning":null},{"id":"e013a02b996a95ebc2ae00cd","modelId":"step-5-preview","benchmarkId":"aa-gdp-pdf","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":14.799999999999999,"warning":null},{"id":"517b7026960a0044c7d05653","modelId":"step-5-preview","benchmarkId":"aa-gdpval-aa","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":1586.63,"warning":null},{"id":"7504c5bd6e9ce1b8e5c7f5e1","modelId":"step-5-preview","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":46.4782205746061,"warning":null},{"id":"f6fdac01ebdf5e190792fdb1","modelId":"step-5-preview","benchmarkId":"aa-itbench-aa","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.5555555555556,"warning":null},{"id":"f47637d5f7cb6e09d0e5407d","modelId":"step-5-preview","benchmarkId":"aa-mlcr-aa","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.6666666666667,"warning":null},{"id":"ff932e1d93554b38481e904e","modelId":"step-5-preview","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":76.35838150289021,"warning":null},{"id":"c6a5ec26e84ae07fde8aefcc","modelId":"step-5-preview","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.3833333333333,"warning":null},{"id":"334ad755e623e4b6693136a6","modelId":"step-5-preview","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":41.53333333333333,"warning":null},{"id":"6142b034e5ac95d3051558ab","modelId":"step-5-preview","benchmarkId":"aa-scicode","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":58.912037037037,"warning":null},{"id":"017170de05d86d07588c3548","modelId":"step-5-preview","benchmarkId":"aa-terminal-bench-science","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":2.38095238095238,"warning":null},{"id":"a329497b5012d24b53ad472e","modelId":"step-5-preview","benchmarkId":"aa-terminalbench-v4-0","level":"reasoning","sourceLabel":"Step 5 Preview","configuration":"Step 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified","score":33.3333333333333,"warning":null},{"id":"43b322ea28a4a6f277552a05","modelId":"step3-vl-10b","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"9132d81889bdde96600d5cfd","modelId":"step3-vl-10b","benchmarkId":"aa-critpt","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"a97b8f1ba39ced23967d346c","modelId":"step3-vl-10b","benchmarkId":"aa-gpqa-diamond","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":68.989898989899,"warning":null},{"id":"faa28dbbc9f9add21df2dc61","modelId":"step3-vl-10b","benchmarkId":"aa-humanitys-last-exam","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":10.8433734939759,"warning":null},{"id":"5cd738dc348ca79f2c279d55","modelId":"step3-vl-10b","benchmarkId":"aa-ifbench","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":50.2040816326531,"warning":null},{"id":"c0c789fed617bbde5f34bffc","modelId":"step3-vl-10b","benchmarkId":"aa-mmmu-pro","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":63.9884393063584,"warning":null},{"id":"0fcf0db8250ffa3a4b7f5ba1","modelId":"step3-vl-10b","benchmarkId":"aa-omniscience","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":-59,"warning":null},{"id":"b442f2f7fef04eefeaf6dde1","modelId":"step3-vl-10b","benchmarkId":"aa-omniscience-accuracy","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":13.266666666666666,"warning":null},{"id":"1d4647242d6328c3454120c9","modelId":"step3-vl-10b","benchmarkId":"aa-tau2-bench","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":16.0818713450292,"warning":null},{"id":"33581c6317d1eb43fcadfe15","modelId":"step3-vl-10b","benchmarkId":"aa-terminalbench-hard","level":"reasoning","sourceLabel":"Step3 VL 10B","configuration":"Step3 VL 10B; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-3-vl-10b; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.3030303030303,"warning":null},{"id":"701a12cb5a469a4de2b81dcc","modelId":"tiny-aya-global","benchmarkId":"aa-artificial-analysis-long-context-reasoning","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"803351c15f6911e960f59a82","modelId":"tiny-aya-global","benchmarkId":"aa-critpt","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"ae198468c720d10bafa7d1ed","modelId":"tiny-aya-global","benchmarkId":"aa-global-mmlu-lite","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":55.13020833333333,"warning":null},{"id":"02b8e27c2385d10ab79bd4fb","modelId":"tiny-aya-global","benchmarkId":"aa-gpqa-diamond","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":30.5050505050505,"warning":null},{"id":"1e3007800bf6ad6160a38a16","modelId":"tiny-aya-global","benchmarkId":"aa-humanitys-last-exam","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":5.23632993512512,"warning":null},{"id":"1438c051d28cc24285d0b24f","modelId":"tiny-aya-global","benchmarkId":"aa-ifbench","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":20.1360544217687,"warning":null},{"id":"41c731de4263a48b615b716d","modelId":"tiny-aya-global","benchmarkId":"aa-omniscience","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":-84.3,"warning":null},{"id":"9fbe07135ac2965896c34113","modelId":"tiny-aya-global","benchmarkId":"aa-omniscience-accuracy","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":6.066666666666666,"warning":null},{"id":"35cc8b451f8300f16237328f","modelId":"tiny-aya-global","benchmarkId":"aa-tau2-bench","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"e244a76401048b5127d71e98","modelId":"tiny-aya-global","benchmarkId":"aa-terminalbench-hard","level":"non-reasoning","sourceLabel":"Tiny Aya Global","configuration":"Tiny Aya Global; Reported effort: non-reasoning; Model source: https://artificialanalysis.ai/models/tiny-aya-global; Mode from evaluator isReasoning metadata; effort budget not specified","score":0,"warning":null},{"id":"0e654cd1466ae894d75cb1e8","modelId":"grok-4.5","benchmarkId":"vulcanbench-v3-07-grok-fable-sol","level":"medium","sourceLabel":"Grok 4.5 (medium)","configuration":"Grok 4.5; source label: Grok 4.5 (medium); medium effort; 23 tasks; evaluation 2026-07-12. See source for repeat counts and wall-clock budget.","score":91.3,"warning":null},{"id":"ffbbc7256f47ee72d321f094","modelId":"grok-4.5","benchmarkId":"vulcanbench-v3-07-grok-fable-sol","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"Grok 4.5; source label: Grok 4.5 (high); high effort; 23 tasks; evaluation 2026-07-12. See source for repeat counts and wall-clock budget.","score":91.3,"warning":null},{"id":"6568b269cb7681ea39a1097c","modelId":"gpt-5.6-sol","benchmarkId":"vulcanbench-v3-07-grok-fable-sol","level":"high","sourceLabel":"GPT-5.6 Sol (high)","configuration":"GPT-5.6 Sol; source label: GPT-5.6 Sol (high); high effort; 23 tasks; evaluation 2026-07-12. See source for repeat counts and wall-clock budget.","score":87,"warning":null},{"id":"55c55b9ae75c2c789bb81880","modelId":"grok-4.5","benchmarkId":"vulcanbench-v3-07-grok-fable-sol","level":"low","sourceLabel":"Grok 4.5 (low)","configuration":"Grok 4.5; source label: Grok 4.5 (low); low effort; 23 tasks; evaluation 2026-07-12. See source for repeat counts and wall-clock budget.","score":82.6,"warning":null},{"id":"82ad7650319190ab8ebf8f1a","modelId":"gpt-5.6-sol","benchmarkId":"vulcanbench-v3-07-grok-fable-sol","level":"medium","sourceLabel":"GPT-5.6 Sol (medium)","configuration":"GPT-5.6 Sol; source label: GPT-5.6 Sol (medium); medium effort; 23 tasks; evaluation 2026-07-12. See source for repeat counts and wall-clock budget.","score":82.6,"warning":null},{"id":"07c58037116b0b06179f7fbf","modelId":"gpt-5.6-sol","benchmarkId":"vulcanbench-v3-07-grok-fable-sol","level":"low","sourceLabel":"GPT-5.6 Sol (low)","configuration":"GPT-5.6 Sol; source label: GPT-5.6 Sol (low); low effort; 23 tasks; evaluation 2026-07-12. See source for repeat counts and wall-clock budget.","score":78.3,"warning":null},{"id":"758fa2580c85027d05e4247c","modelId":"kimi-k3","benchmarkId":"vulcanbench-v3-08-kimi-k3","level":"max","sourceLabel":"Kimi K3 (max, 30-min budget)","configuration":"Kimi K3; source label: Kimi K3 (max, 30-min budget); max effort; 23 tasks; evaluation 2026-07-19. See source for repeat counts and wall-clock budget.","score":73.9,"warning":null},{"id":"a5c39da8a1cf93a9c675f0ee","modelId":"claude-opus-5","benchmarkId":"vulcanbench-v3-10-opus5-effort","level":"low","sourceLabel":"low","configuration":"Claude Opus 5; source label: low; low effort; 23 tasks; evaluation 2026-07-26. See source for repeat counts and wall-clock budget.","score":87,"warning":null},{"id":"7177575f578b9a510d14e532","modelId":"claude-opus-5","benchmarkId":"vulcanbench-v3-10-opus5-effort","level":"medium","sourceLabel":"medium","configuration":"Claude Opus 5; source label: medium; medium effort; 23 tasks; evaluation 2026-07-26. See source for repeat counts and wall-clock budget.","score":82.6,"warning":null},{"id":"f20cb695188c6fb09c7bfde6","modelId":"claude-opus-5","benchmarkId":"vulcanbench-v3-10-opus5-effort","level":"high","sourceLabel":"high","configuration":"Claude Opus 5; source label: high; high effort; 23 tasks; evaluation 2026-07-26. See source for repeat counts and wall-clock budget.","score":78.3,"warning":null},{"id":"724e797747b99ab179cc86f3","modelId":"qwen-3.8-max","benchmarkId":"vulcanbench-v3-12-qwen38-max","level":"low","sourceLabel":"low","configuration":"Qwen3.8-Max; source label: low; low effort; 23 tasks; evaluation 2026-08-04. See source for repeat counts and wall-clock budget.","score":81.2,"warning":null},{"id":"fc3245dd2348d67c6fe183bd","modelId":"qwen-3.8-max","benchmarkId":"vulcanbench-v3-12-qwen38-max","level":"medium","sourceLabel":"medium","configuration":"Qwen3.8-Max; source label: medium; medium effort; 23 tasks; evaluation 2026-08-04. See source for repeat counts and wall-clock budget.","score":71,"warning":null},{"id":"42a596f42d31c722ee3e28df","modelId":"qwen-3.8-max","benchmarkId":"vulcanbench-v3-12-qwen38-max","level":"xhigh","sourceLabel":"xhigh (default)","configuration":"Qwen3.8-Max; source label: xhigh (default); xhigh effort; 23 tasks; evaluation 2026-08-04. Source reports 55.1% and 38/64 completed runs; incomplete trials. Published percentage retained.","score":55.1,"warning":"Source reports 55.1% and 38/64 completed runs; incomplete trials. Published percentage retained."},{"id":"1bef904009c6a794c1dd2f03","modelId":"grok-4.6","benchmarkId":"vulcanbench-v3-14-grok-46-effort","level":"low","sourceLabel":"low","configuration":"Grok 4.6; source label: low; low effort; 23 tasks; evaluation 2026-08-12. See source for repeat counts and wall-clock budget.","score":82.6,"warning":null},{"id":"5f32740262d72bbbdbf60a38","modelId":"grok-4.6","benchmarkId":"vulcanbench-v3-14-grok-46-effort","level":"medium","sourceLabel":"medium","configuration":"Grok 4.6; source label: medium; medium effort; 23 tasks; evaluation 2026-08-12. See source for repeat counts and wall-clock budget.","score":87,"warning":null},{"id":"d99fc9802ae6b1d7987cad39","modelId":"grok-4.6","benchmarkId":"vulcanbench-v3-14-grok-46-effort","level":"high","sourceLabel":"high (default)","configuration":"Grok 4.6; source label: high (default); high effort; 23 tasks; evaluation 2026-08-12. See source for repeat counts and wall-clock budget.","score":73.9,"warning":null},{"id":"47ae4c48e3377c9ee20c9652","modelId":"grok-4.6","benchmarkId":"vulcanbench-v3-14-grok-46-effort","level":"xhigh","sourceLabel":"xhigh","configuration":"Grok 4.6; source label: xhigh; xhigh effort; 23 tasks; evaluation 2026-08-12. See source for repeat counts and wall-clock budget.","score":78.3,"warning":null},{"id":"69dd3899f972d5a522219937","modelId":"qwen3.8-27b","benchmarkId":"vulcanbench-v3-17-qwen38-27b-effort","level":"low","sourceLabel":"low","configuration":"Qwen3.8-27B; source label: low; low effort; 23 tasks; evaluation 2026-08-21. See source for repeat counts and wall-clock budget.","score":82.6,"warning":null},{"id":"d6e683f349eb46ae01149f20","modelId":"qwen3.8-27b","benchmarkId":"vulcanbench-v3-17-qwen38-27b-effort","level":"medium","sourceLabel":"medium","configuration":"Qwen3.8-27B; source label: medium; medium effort; 23 tasks; evaluation 2026-08-21. See source for repeat counts and wall-clock budget.","score":82.6,"warning":null},{"id":"5a6667c8022196c74f1a2ccf","modelId":"qwen3.8-27b","benchmarkId":"vulcanbench-v3-17-qwen38-27b-effort","level":"xhigh","sourceLabel":"xhigh","configuration":"Qwen3.8-27B; source label: xhigh; xhigh effort; 23 tasks; evaluation 2026-08-21. See source for repeat counts and wall-clock budget.","score":73.9,"warning":null},{"id":"f796e96bd180b4dd9bd12eff","modelId":"glm-5.3","benchmarkId":"vulcanbench-v3-18-glm53-zcode-harness","level":"low","sourceLabel":"low","configuration":"GLM 5.3; source label: low; low effort; 23 tasks; evaluation 2026-08-24. Source labels low/high/max; Z.ai API effort labels may be metadata only. Bare-bones API column, not ZCode.","score":78.3,"warning":"Source labels low/high/max; Z.ai API effort labels may be metadata only. Bare-bones API column, not ZCode."},{"id":"d783859010d571f133e77356","modelId":"glm-5.3","benchmarkId":"vulcanbench-v3-18-glm53-zcode-harness","level":"high","sourceLabel":"high","configuration":"GLM 5.3; source label: high; high effort; 23 tasks; evaluation 2026-08-24. Source labels low/high/max; Z.ai API effort labels may be metadata only. Bare-bones API column, not ZCode.","score":73.9,"warning":"Source labels low/high/max; Z.ai API effort labels may be metadata only. Bare-bones API column, not ZCode."},{"id":"3bde5c20893b402a0e56ffb5","modelId":"glm-5.3","benchmarkId":"vulcanbench-v3-18-glm53-zcode-harness","level":"max","sourceLabel":"max","configuration":"GLM 5.3; source label: max; max effort; 23 tasks; evaluation 2026-08-24. Source labels low/high/max; Z.ai API effort labels may be metadata only. Bare-bones API column, not ZCode.","score":65.2,"warning":"Source labels low/high/max; Z.ai API effort labels may be metadata only. Bare-bones API column, not ZCode."},{"id":"1b7ece23e8756e740abb6e39","modelId":"muse-spark-1.2","benchmarkId":"vulcanbench-v3-19-musespark-effort","level":"low","sourceLabel":"low","configuration":"Muse Spark 1.2; source label: low; low effort; 23 tasks; evaluation 2026-08-25. See source for repeat counts and wall-clock budget.","score":87,"warning":null},{"id":"970bc242697fc80101d9ac17","modelId":"muse-spark-1.2","benchmarkId":"vulcanbench-v3-19-musespark-effort","level":"high","sourceLabel":"high","configuration":"Muse Spark 1.2; source label: high; high effort; 23 tasks; evaluation 2026-08-25. See source for repeat counts and wall-clock budget.","score":73.9,"warning":null},{"id":"c5e80648923e0576c68a9c39","modelId":"muse-spark-1.2","benchmarkId":"vulcanbench-v3-19-musespark-effort","level":"xhigh","sourceLabel":"xhigh","configuration":"Muse Spark 1.2; source label: xhigh; xhigh effort; 23 tasks; evaluation 2026-08-25. See source for repeat counts and wall-clock budget.","score":52.2,"warning":null},{"id":"00cd14ef96cfd2e8dfd5e845","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Fable 5 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.003","score":53.480000000000004,"warning":null},{"id":"0181fe7a76a53fbd63e479e8","modelId":"gpt-6-luna","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"GPT-6 Luna (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-luna; variant provider-adapter-low; public-demo task scores are excluded","score":0.2374807559887363,"warning":null},{"id":"01c83861b72d5096082de252","modelId":"claude-opus-4-6","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Opus 4.6 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.003","score":18.310000000000002,"warning":null},{"id":"03d8a27181e69e9adae8ac02","modelId":"grok-4.7","benchmarkId":"terminal4-grokbuild","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Grok Build; effort xhigh; reported margin ±3.5 percentage points, confidence level unspecified; release date 2026-09-21; board checked 2026-10-07","score":37.6,"warning":null},{"id":"05679093da19998c7422e8ae","modelId":"qwen-3.8-max-0902","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"Qwen3.8-Max-0902 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±3.8 percentage points, confidence level unspecified; release date 2026-09-02; board checked 2026-10-07","score":27,"warning":null},{"id":"07002a07d2fb472041035fd7","modelId":"gemini-4-argon","benchmarkId":"frontiercode-1.1-main-chisel","level":"low","sourceLabel":"Gemini 4 Argon (low)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0304","score":47.199999999999996,"warning":null},{"id":"07392b4f7f6b227f47de058d","modelId":"gemini-3.8-flash","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"Gemini 3.8 Flash (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group google-gemini-3-8-flash; variant provider-adapter-low; public-demo task scores are excluded","score":15.071935405559648,"warning":null},{"id":"09d0d6ffaac7cfac027f3198","modelId":"claude-haiku-5-5","benchmarkId":"arc3-standard","level":"max","sourceLabel":"Claude Haiku 5.5 (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group anthropic-claude-haiku-5-5; variant max; public-demo task scores are excluded","score":0.4513711877557703,"warning":null},{"id":"09f19fb85700296ac24a0c0c","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Haiku 5.5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":41.63,"warning":null},{"id":"0a4e83f3b53d6fc01713d497","modelId":"gemini-3.8-flash","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"Gemini 3.8 Flash (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group google-gemini-3-8-flash; variant medium; public-demo task scores are excluded","score":3.9878093830617267,"warning":null},{"id":"0a5efdb346f5b8dbbb92cfe1","modelId":"mistral-large-4","benchmarkId":"frontiercode-1.1-extended-chisel","level":"unspecified","sourceLabel":"Mistral Large 4 (Preview) (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0268","score":40.89,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"0a7287f39f6dbfcfbb63262d","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Sonnet 5.5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0053","score":50.46000000000001,"warning":null},{"id":"0acab77924c8fa3e469de011","modelId":"gemini-3.6-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"low","sourceLabel":"Gemini 3.6 Flash (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":32.769999999999996,"warning":null},{"id":"0af702ab6d9ce3f4e56982ae","modelId":"grok-4.7","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"high","sourceLabel":"Grok 4.7 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0087","score":59.35,"warning":null},{"id":"0c3bc2dca91577d8e531e573","modelId":"gpt-6-sol","benchmarkId":"arc3-standard","level":"none","sourceLabel":"GPT-6 Sol (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-sol; variant none; public-demo task scores are excluded","score":0.2837742350643006,"warning":null},{"id":"0d5e49902497ab349126387e","modelId":"claude-opus-4-6","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Opus 4.6 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":26.640000000000004,"warning":null},{"id":"0d89a61a72b86a5fc22b7ebc","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Fable 5.1 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":61.62,"warning":null},{"id":"0dfcbad58c343133d5960e81","modelId":"claude-opus-5","benchmarkId":"terminal4-claudecode","level":"xhigh","sourceLabel":"Opus 5 (xhigh)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort xhigh; reported margin ±3.2 percentage points, confidence level unspecified; release date 2026-07-24; board checked 2026-10-07","score":53.9,"warning":null},{"id":"0e472c316abe4044c3d6bce0","modelId":"claude-fable-5","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"Fable 5 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±3.8 percentage points, confidence level unspecified; release date 2026-06-09; board checked 2026-10-07","score":44.5,"warning":null},{"id":"0ed985881b62cd0bc3af4eb4","modelId":"gpt-6-astra","benchmarkId":"arc3-standard","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-astra; variant max; public-demo task scores are excluded","score":62.71280210060628,"warning":null},{"id":"0f8151b0e90577da4ff2e3c1","modelId":"mistral-large-4","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"Mistral Large 4 (Preview) (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.02","score":39.43,"warning":null},{"id":"118142e09fa2bec781ac751e","modelId":"glm-5.3","benchmarkId":"frontiercode-1.1-extended-chisel","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":52.629999999999995,"warning":null},{"id":"11d69a3ea3e8e3493db5d23a","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-6 Sol (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":50.55,"warning":null},{"id":"11e28af907b736a7ef5132e0","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-6 Luna (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":51.43,"warning":null},{"id":"12560fec6f79d44c90fd24c7","modelId":"gpt-6-luna","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"GPT-6 Luna (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-luna; variant medium; public-demo task scores are excluded","score":0.19415284639354707,"warning":null},{"id":"125ae37f54ee1d63963a88cc","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":58.120000000000005,"warning":null},{"id":"126b238f0fce5e381a15f5c2","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Opus 4.7 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":28.910000000000004,"warning":null},{"id":"126e948a87af9bad8d67f24c","modelId":"gpt-6-sol","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"GPT-6 Sol (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-sol; variant provider-adapter-low; public-demo task scores are excluded","score":1.7619154311646184,"warning":null},{"id":"1477faeb76f4bf5d802d34b4","modelId":"gpt-6-sol","benchmarkId":"arc3-standard","level":"xhigh","sourceLabel":"GPT-6 Sol (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-sol; variant xhigh; public-demo task scores are excluded","score":1.7976269602783517,"warning":null},{"id":"150f043c8b72655148c79768","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Fable 5.1 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":49.82,"warning":null},{"id":"168f9f1a24e3902156eff89f","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-6.1 Sol (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":60.370000000000005,"warning":null},{"id":"1727937b7ec82bbc16c23336","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0465","score":46.21,"warning":null},{"id":"17937afc9a71c653c9d4f6f2","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":63.629999999999995,"warning":null},{"id":"17980b94e41d75e8340b832a","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-6.1 Sol (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0027","score":60.4,"warning":null},{"id":"18ca9b5ee43fc4850d343f7b","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Opus 4.7 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":27.61,"warning":null},{"id":"190a41f349d6ec138ffcdc33","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0101","score":48.04,"warning":null},{"id":"1927548b0a16cd1479986cbd","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":50.28,"warning":null},{"id":"19fc7dc49aea8a2208bcc28e","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Opus 5.5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":65.2,"warning":null},{"id":"1a14e537a8a814fc2f9bc5d4","modelId":"gpt-6-luna","benchmarkId":"arc3-standard","level":"xhigh","sourceLabel":"GPT-6 Luna (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-luna; variant xhigh; public-demo task scores are excluded","score":0.16413333960356602,"warning":null},{"id":"1cf1455535fc32710e1965cf","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Opus 4.8 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0073","score":55.63,"warning":null},{"id":"1cf9f554d2144da3e03275ee","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-6.1 Sol (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":59.27,"warning":null},{"id":"1ed4c92e4aa05c76200a18bc","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Sonnet 4.6 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":30.44,"warning":null},{"id":"1f3aa6afe3bf07a0c02a7308","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-5.6 Luna (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":41.17,"warning":null},{"id":"1f438db0c3a8d74e269dc9c6","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-5.4-mini (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":20.580000000000002,"warning":null},{"id":"1fb4cc080f6a704550c60899","modelId":"grok-4.6","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"medium","sourceLabel":"Grok 4.6 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":59.62,"warning":null},{"id":"2105c7646f3a139187821073","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-5.6 Terra (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":52.629999999999995,"warning":null},{"id":"212dafc406aab75cb58766c0","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Opus 4.7 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":44.75,"warning":null},{"id":"21c236300e02ca27fc9fa65a","modelId":"gemini-4-argon","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"Gemini 4 Argon (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group google-gemini-4-argon; variant medium; public-demo task scores are excluded","score":46.265668402958575,"warning":null},{"id":"21ee231589903749e98387ba","modelId":"gpt-6-astra","benchmarkId":"arc3-provider-adapter","level":"xhigh","sourceLabel":"GPT-6 Astra (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-astra; variant provider-adapter-xhigh; public-demo task scores are excluded","score":98.44155844155843,"warning":null},{"id":"247bf3dc35d9e67fba3e754a","modelId":"claude-sonnet-5","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"Sonnet 5 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±3.1 percentage points, confidence level unspecified; release date 2026-06-30; board checked 2026-10-07","score":12.4,"warning":null},{"id":"255229c9fb97665746c2f311","modelId":"nemotron-3-ultra","benchmarkId":"frontiercode-1.1-main-chisel","level":"unspecified","sourceLabel":"Nemotron 3 Ultra (none)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":13.61,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"26490bed84e4943e2df858ae","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Opus 5.5 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":63.53,"warning":null},{"id":"2701d728ed945b354a434f26","modelId":"gemini-3.7-flash","benchmarkId":"terminal4-minisweagent","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"Terminal-Bench 4.0.0; 66 tasks; mini-SWE-agent; effort high; reported margin ±2.4 percentage points, confidence level unspecified; release date 2026-08-13; board checked 2026-10-07","score":11.2,"warning":null},{"id":"276a6edb836c21a802685b40","modelId":"gpt-6-astra","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"GPT-6 Astra (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-astra; variant provider-adapter-high; public-demo task scores are excluded","score":99.94571232125699,"warning":null},{"id":"28a8f1e475f0000b4b484058","modelId":"deepseek-v4-pro","benchmarkId":"frontiercode-1.1-extended-mini-swe-agent","level":"unspecified","sourceLabel":"DeepSeek V4 Pro (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0067","score":31.009999999999998,"warning":"DeepSeek V4 Pro checkpoint is unspecified; April and August must remain separate."},{"id":"2ab8119e321a20d7de8cf1c0","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-5.6 Terra (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":37.79,"warning":null},{"id":"2bd0af0abcc178550c64a1ac","modelId":"grok-4.5","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":56.55,"warning":null},{"id":"2c3bec9e4e0eff839c3ff4df","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Fable 5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":62.79,"warning":null},{"id":"2cb65c8958eab0f2022804ae","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-6 Astra (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":50.62,"warning":null},{"id":"2d0594eef125dd86ef42cfcb","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-5.6 Luna (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":54.50000000000001,"warning":null},{"id":"2d2cb54f2cced8ad23e6d7f8","modelId":"gpt-6-astra","benchmarkId":"arc3-provider-adapter","level":"medium","sourceLabel":"GPT-6 Astra (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-astra; variant provider-adapter-medium; public-demo task scores are excluded","score":98.43861309255745,"warning":null},{"id":"2e242cd77be5707ba1963d4c","modelId":"gpt-6.1-sol","benchmarkId":"arc3-provider-adapter","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-1-sol; variant provider-adapter-max; public-demo task scores are excluded","score":96.17739071933946,"warning":null},{"id":"2e2a4bc317bd35c6e0a9acff","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0337","score":59.089999999999996,"warning":null},{"id":"3021bee4352805ca337f7cbc","modelId":"gpt-6.1-sol","benchmarkId":"arc3-standard","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-1-sol; variant max; public-demo task scores are excluded","score":52.72904006820879,"warning":null},{"id":"303f27de5519d24ea6491389","modelId":"nemotron-3-ultra","benchmarkId":"frontiercode-1.1-extended-chisel","level":"unspecified","sourceLabel":"Nemotron 3 Ultra (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":27.21,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"304e04af8619e1665e7baa68","modelId":"grok-4.6","benchmarkId":"frontiercode-1.1-main-grok-build","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.007","score":48.010000000000005,"warning":null},{"id":"308e1d6d59efae59708b2ef0","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-standard","level":"none","sourceLabel":"DeepSeek V4.1 Flash (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group deepseek-v4-1-flash; variant none; public-demo task scores are excluded","score":0.3431694493873915,"warning":null},{"id":"30e4cc08224c496e6f585f72","modelId":"claude-opus-5-5","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"Opus 5.5 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±3.1 percentage points, confidence level unspecified; release date 2026-09-22; board checked 2026-10-07","score":64.8,"warning":null},{"id":"31b2d9d97c184079238926f4","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Haiku 5.5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":55.04,"warning":null},{"id":"32358b5c0d1561f78ac33cd5","modelId":"claude-haiku-5-5","benchmarkId":"arc3-provider-adapter","level":"max","sourceLabel":"Claude Haiku 5.5 (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group anthropic-claude-haiku-5-5; variant provider-adapter-max; public-demo task scores are excluded","score":2.819365772805844,"warning":null},{"id":"33185b87fae17b44533c6d2f","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-6 Luna (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":25.66,"warning":null},{"id":"336011e6d0e94555b099f00f","modelId":"gpt-6-astra","benchmarkId":"arc3-standard","level":"none","sourceLabel":"GPT-6 Astra (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-astra; variant none; public-demo task scores are excluded","score":35.177880317424,"warning":"Source-reported None does not establish a supported Astra API setting."},{"id":"34af973aab00ee6e3f8d8bf8","modelId":"mistral-medium-3.5","benchmarkId":"frontiercode-1.1-extended-chisel","level":"unspecified","sourceLabel":"Mistral 3.5 Medium (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":18.66,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"35e4eb7c3f88bd81a2774079","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Sonnet 5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":49.28,"warning":null},{"id":"35e9f237985104a6fd022a74","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-5.6 Terra (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":38.75,"warning":null},{"id":"364834ea4ff4cde79854901b","modelId":"kimi-k3","benchmarkId":"frontiercode-1.1-main-mini-swe-agent","level":"unspecified","sourceLabel":"Kimi K3 (none)","configuration":"FrontierCode 1.1; main; 100 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0021","score":44.17,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"37e808d11c66296c681ea587","modelId":"gemini-3.7-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"low","sourceLabel":"Gemini 3.7 Flash (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":51.849999999999994,"warning":null},{"id":"391c90fa4b674dc1c3932918","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0067","score":58.940000000000005,"warning":null},{"id":"395047417b94ee3b51adaa55","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Fable 5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.003","score":49.82,"warning":null},{"id":"3b38242e2298df478a3621e4","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-5.6 Terra (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":50.86000000000001,"warning":null},{"id":"3bb49576b7608521da1ae45c","modelId":"claude-opus-4-8","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"Opus 4.8 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±3.6 percentage points, confidence level unspecified; release date 2026-05-28; board checked 2026-10-07","score":23.6,"warning":null},{"id":"3c433cf7df074c5af7c027bd","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-5.6 Sol (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":49.99,"warning":null},{"id":"3c660a8ae862944a3486209b","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-6 Luna (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":37.1,"warning":null},{"id":"3c7cf4ee69520ebbb0843eb2","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":60.69,"warning":null},{"id":"3ca697c43da26b4e6144885f","modelId":"gpt-5.6-terra","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"GPT-5.6 Terra (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-terra; variant medium; public-demo task scores are excluded","score":0.08,"warning":null},{"id":"3cfdd48cd8b66a487f18b2e9","modelId":"gpt-5.6-luna","benchmarkId":"arc3-standard","level":"high","sourceLabel":"GPT-5.6 Luna (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-luna; variant high; public-demo task scores are excluded","score":0.1,"warning":null},{"id":"3d196b1d122191d2b4c8ca49","modelId":"gpt-6-sol","benchmarkId":"arc3-provider-adapter","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-sol; variant provider-adapter-max; public-demo task scores are excluded","score":23.028056734101497,"warning":null},{"id":"3d5e60cff8818bd0c8d67cc0","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-5.6 Luna (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":35.89,"warning":null},{"id":"3f11fa1a09b61e888918e7a3","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-6.1 Sol (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":50.23,"warning":null},{"id":"3f43c11c9388f23914100624","modelId":"gemini-3.8-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":51.239999999999995,"warning":null},{"id":"4089087bc0de296803ff3a88","modelId":"gpt-5.6-sol","benchmarkId":"arc3-standard","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-sol; variant max; public-demo task scores are excluded","score":7.779999999999999,"warning":null},{"id":"4348fbb897a2a59eae46e3f4","modelId":"claude-opus-4-6","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Opus 4.6 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":43.669999999999995,"warning":null},{"id":"44944d3c070d22723450a4bb","modelId":"gemini-4-argon","benchmarkId":"frontiercode-1.1-main-chisel","level":"medium","sourceLabel":"Gemini 4 Argon (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0121","score":47.55,"warning":null},{"id":"4576356c73ac05320b9f32bf","modelId":"gpt-6-astra","benchmarkId":"terminal4-codex","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Codex; effort max; reported margin ±2.8 percentage points, confidence level unspecified; release date 2026-09-03; board checked 2026-10-07","score":58.2,"warning":null},{"id":"462443c5210915015b4c7bf9","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-5.4-mini (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":43.01,"warning":null},{"id":"46f65324a766a154476a00ad","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":40,"warning":null},{"id":"4832e4d29a499174e911f69b","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Sonnet 5.5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0067","score":61.46,"warning":null},{"id":"489ca3f16db7030a704db2f9","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-provider-adapter","level":"max","sourceLabel":"DeepSeek V4.1 Flash (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group deepseek-v4-1-flash; variant provider-adapter-max; public-demo task scores are excluded","score":2.6843916288028566,"warning":null},{"id":"4a70f6ffa3c870b89b0268d2","modelId":"gpt-6-luna","benchmarkId":"arc3-standard","level":"none","sourceLabel":"GPT-6 Luna (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-luna; variant none; public-demo task scores are excluded","score":0.027919455042839488,"warning":null},{"id":"4b326191d83fb7c7b1f5d5b3","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-5.5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":36.6,"warning":null},{"id":"4bacad2a4162155bcf382a46","modelId":"gemini-3.7-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":42.230000000000004,"warning":null},{"id":"4c6eaa166ea13440ca01eb77","modelId":"deepseek-v4.1-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"DeepSeek V4.1 Flash (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0073","score":43.87,"warning":null},{"id":"4c92e3a3655317402e6e0fc6","modelId":"deepseek-v4-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.1911","score":31.7,"warning":null},{"id":"4cfed6e5d5f8f275ec20cb69","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"DeepSeek V4.1 Flash (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group deepseek-v4-1-flash; variant provider-adapter-high; public-demo task scores are excluded","score":3.1518486978853844,"warning":null},{"id":"4e7cfd9b5ba456e592a61884","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Haiku 5.5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":41.89,"warning":null},{"id":"4e9cc5a5d770369cade88aba","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-extended-codex","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":64.48,"warning":null},{"id":"4f12de502332d1a93cca6d84","modelId":"gemini-3.6-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"medium","sourceLabel":"Gemini 3.6 Flash (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":48.02,"warning":null},{"id":"4f8d2f40bdc3160cdf5b7a04","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-5.6 Terra (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":24.11,"warning":null},{"id":"505126abb98e4d099fef8ee7","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Haiku 5.5 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":46.36,"warning":null},{"id":"50b7158c541e4c1686f7c817","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Sonnet 4.6 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":21.099999999999998,"warning":null},{"id":"51d667d1f41dc9e2af12b5f5","modelId":"gemini-4-argon","benchmarkId":"frontiercode-1.1-extended-chisel","level":"medium","sourceLabel":"Gemini 4 Argon (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0134","score":60.5,"warning":null},{"id":"522b46b6ccbbf354137774aa","modelId":"grok-4.7","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"Grok 4.7 (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-7; variant medium; public-demo task scores are excluded","score":1.7056643767835524,"warning":null},{"id":"52da31a04b7d561b7573dc1f","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Opus 5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":63.629999999999995,"warning":null},{"id":"5402da52d181b12603f15f33","modelId":"glm-5.3-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":46.21,"warning":null},{"id":"541e90621a96734aecdd8834","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Opus 4.8 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.008","score":41.02,"warning":null},{"id":"5456762f7405236910bae307","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-6 Sol (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":45.92,"warning":null},{"id":"55349c832862fc4d4b933188","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Fable 5.1 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":50.91,"warning":null},{"id":"56b0314f31e75a7aa19bd3b1","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Opus 5 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0144","score":43.65,"warning":null},{"id":"5957688c8334d48ade93b3d5","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-5.6 Sol (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":46.839999999999996,"warning":null},{"id":"598efd299c0a6cb52c8832b5","modelId":"gpt-6-sol","benchmarkId":"arc3-standard","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-sol; variant max; public-demo task scores are excluded","score":4.623568186131146,"warning":null},{"id":"59a4f2eb5190190599faa9d7","modelId":"gemini-4-argon","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"Gemini 4 Argon (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group google-gemini-4-argon; variant provider-adapter-low; public-demo task scores are excluded","score":97.4591718377776,"warning":null},{"id":"5b4b76daf2a598a3aa8925a7","modelId":"gemini-3.6-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"low","sourceLabel":"Gemini 3.6 Flash (low)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":22.82,"warning":null},{"id":"5b52b90c46791b28a2952476","modelId":"grok-4.7","benchmarkId":"arc3-provider-adapter","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group xai-grok-4-7; variant provider-adapter-xhigh; public-demo task scores are excluded","score":7.016486895838931,"warning":null},{"id":"5bf518719461d570e3d4c0c6","modelId":"gpt-6-luna","benchmarkId":"arc3-provider-adapter","level":"xhigh","sourceLabel":"GPT-6 Luna (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-luna; variant provider-adapter-xhigh; public-demo task scores are excluded","score":0.5072886313635385,"warning":null},{"id":"5c999b15cf49cd9ccd8eb829","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":39.81,"warning":null},{"id":"5ce108c441495c34dbedea2f","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.006","score":46.5,"warning":null},{"id":"5dd2a190643d98ea4f8be204","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Opus 5.5 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0053","score":63.580000000000005,"warning":null},{"id":"5df86550feaab57fdd6e2c5f","modelId":"gpt-6.1-sol","benchmarkId":"arc3-standard","level":"xhigh","sourceLabel":"GPT-6.1 Sol (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-1-sol; variant xhigh; public-demo task scores are excluded","score":39.92932040733721,"warning":null},{"id":"5e17a13831c8ef1944a5a977","modelId":"gemini-4-argon","benchmarkId":"arc3-provider-adapter","level":"medium","sourceLabel":"Gemini 4 Argon (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group google-gemini-4-argon; variant provider-adapter-medium; public-demo task scores are excluded","score":97.3448904168153,"warning":null},{"id":"5e9500f221faa7f9062ad0d2","modelId":"grok-4.7","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"Grok 4.7 (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group xai-grok-4-7; variant provider-adapter-low; public-demo task scores are excluded","score":0.7739582026314942,"warning":null},{"id":"614fa62df1030f50a57ff66b","modelId":"gemini-3.8-flash","benchmarkId":"arc3-standard","level":"low","sourceLabel":"Gemini 3.8 Flash (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group google-gemini-3-8-flash; variant low; public-demo task scores are excluded","score":5.989560493697216,"warning":null},{"id":"618440f872edf631f4f0df6a","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-5.6 Luna (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":38.91,"warning":null},{"id":"61fd47e8efd909e62876c9b4","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Sonnet 5 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.006","score":56.18,"warning":null},{"id":"62ebd59a104b5a031160d534","modelId":"gpt-5.6-sol","benchmarkId":"arc3-standard","level":"low","sourceLabel":"GPT-5.6 Sol (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-sol; variant low; public-demo task scores are excluded","score":0.33,"warning":null},{"id":"62f127539d56d2de5540a715","modelId":"kimi-k3","benchmarkId":"frontiercode-1.1-extended-mini-swe-agent","level":"unspecified","sourceLabel":"Kimi K3 (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0014","score":58.19,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"6312d7db4c52795cad633019","modelId":"gemini-4-argon","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group google-gemini-4-argon; variant provider-adapter-high; public-demo task scores are excluded","score":97.30432688837199,"warning":null},{"id":"6378f4d933273f336b571e3d","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Sonnet 5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":35.199999999999996,"warning":null},{"id":"668028a9e1437683d6c27dea","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Opus 5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":47.99,"warning":null},{"id":"66ba0759f64f15955b12acec","modelId":"gemini-3.8-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"medium","sourceLabel":"Gemini 3.8 Flash (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":53.449999999999996,"warning":null},{"id":"67d76f5adea449a9001331d3","modelId":"gpt-6-astra","benchmarkId":"arc3-provider-adapter","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-astra; variant provider-adapter-max; public-demo task scores are excluded","score":98.55074288592294,"warning":null},{"id":"68d15518ffdac72b611b5888","modelId":"gpt-5.6-luna","benchmarkId":"terminal4-codex","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Codex; effort max; reported margin ±2.8 percentage points, confidence level unspecified; release date 2026-06-26; board checked 2026-10-07","score":17.3,"warning":null},{"id":"6958a5ca37b2c58d4c9c3e9d","modelId":"grok-4.6","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0047","score":61.309999999999995,"warning":null},{"id":"69a8ffd616055a9fca319e41","modelId":"gpt-6.1-sol","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"GPT-6.1 Sol (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-1-sol; variant medium; public-demo task scores are excluded","score":10.568764673656194,"warning":null},{"id":"6aa803cb83233f756479ec7d","modelId":"minimax-m3","benchmarkId":"frontiercode-1.1-main-mini-swe-agent","level":"unspecified","sourceLabel":"MiniMax M3 (none)","configuration":"FrontierCode 1.1; main; 100 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.015","score":14.7,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"6b2c7fe28077ca058f5212f5","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-standard","level":"high","sourceLabel":"DeepSeek V4.1 Flash (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group deepseek-v4-1-flash; variant high; public-demo task scores are excluded","score":0.4846664966204641,"warning":null},{"id":"6c1bd9be1ef1dcf41fece4a8","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-6 Astra (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":57.4,"warning":null},{"id":"6dd3175afce39f643aa12476","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Sonnet 4.6 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":39.86,"warning":null},{"id":"6de659b8ed7b026127969cf3","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-5.6 Terra (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":47.660000000000004,"warning":null},{"id":"6f3efa7c29991f0df2e73360","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":52.09,"warning":null},{"id":"71b1f4840ca9c0df3bc97975","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-6 Luna (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":51.300000000000004,"warning":null},{"id":"72e13f593d1b8f08cf9fdba2","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"DeepSeek V4.1 Flash (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group deepseek-v4-1-flash; variant provider-adapter-low; public-demo task scores are excluded","score":3.264386796625269,"warning":null},{"id":"730b47939e189dc833d4d490","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Opus 5.5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":53.99,"warning":null},{"id":"739cb0ae8374d5569ce1d129","modelId":"glm-5.3","benchmarkId":"frontiercode-1.1-main-chisel","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":40.14,"warning":null},{"id":"746cf85a9f9358e50bf922ef","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-6.1 Sol (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.006","score":49.29,"warning":null},{"id":"74826bf1588713bdce3f6257","modelId":"grok-4.6","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"low","sourceLabel":"Grok 4.6 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":55.02,"warning":null},{"id":"74e9e0d52753aa4e8c4a2c57","modelId":"gemini-4-argon","benchmarkId":"frontiercode-1.1-extended-chisel","level":"low","sourceLabel":"Gemini 4 Argon (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0229","score":60.58,"warning":null},{"id":"750e6cc699e912bef459e45b","modelId":"gemini-3.7-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"medium","sourceLabel":"Gemini 3.7 Flash (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":43.59,"warning":null},{"id":"756c6aa0f40124946baeaf66","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Sonnet 5.5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":29.330000000000002,"warning":null},{"id":"78570d5f6c6db191391b67a4","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-6 Sol (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":59.62,"warning":null},{"id":"7879d0685c077f756cc44783","modelId":"qwen3.7-plus","benchmarkId":"frontiercode-1.1-extended-mini-swe-agent","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.006","score":21.78,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"78cbe45f9b9ed1991de85fd0","modelId":"deepseek-v4-pro","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"DeepSeek V4 Pro 0813 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.106","score":28.549999999999997,"warning":null},{"id":"79f0600c94bfc8d4dbc0b58a","modelId":"gpt-5.6-luna","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"GPT-5.6 Luna (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-luna; variant medium; public-demo task scores are excluded","score":0.16999999999999998,"warning":null},{"id":"7a9e18d2451dd72d99b8c928","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-6.1 Sol (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":45.46,"warning":null},{"id":"7b0f1b3e720e4f33cde4257b","modelId":"gpt-5.6-sol","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"GPT-5.6 Sol (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-sol; variant medium; public-demo task scores are excluded","score":1.0699999999999998,"warning":null},{"id":"7c2bc70ac2ad043aca27ef0b","modelId":"gpt-6-sol","benchmarkId":"arc3-provider-adapter","level":"xhigh","sourceLabel":"GPT-6 Sol (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-sol; variant provider-adapter-xhigh; public-demo task scores are excluded","score":9.453334058864572,"warning":null},{"id":"7d8be1469a7c57ca60300f6b","modelId":"claude-opus-5","benchmarkId":"arc3-standard","level":"high","sourceLabel":"Claude Opus 5 (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group anthropic-claude-opus-5; variant high; public-demo task scores are excluded","score":30.159999999999997,"warning":null},{"id":"7da2111f1593f8d5dbf745f9","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-5.4-mini (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":27.04,"warning":null},{"id":"7db12365ec82e0beae8240ba","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":38.54,"warning":null},{"id":"7f33b674105fc71448617806","modelId":"deepseek-v4-pro","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"DeepSeek V4 Pro 0813 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.076","score":44.45,"warning":null},{"id":"7f8f6fd0bdf1771f869816b9","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Opus 5.5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":65.27,"warning":null},{"id":"7f9263f4d41994356a325281","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Haiku 5.5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":34.760000000000005,"warning":null},{"id":"7fe01339271092343301c357","modelId":"grok-4.5","benchmarkId":"arc3-standard","level":"low","sourceLabel":"Grok 4.5 (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-5; variant low; public-demo task scores are excluded","score":0.26,"warning":null},{"id":"80632f12f66bd4e618b6942e","modelId":"grok-4.7","benchmarkId":"frontiercode-1.1-main-grok-build","level":"high","sourceLabel":"Grok 4.7 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.013","score":47.589999999999996,"warning":null},{"id":"8078e6f76a0f6f0d4f6fd030","modelId":"gpt-6-luna","benchmarkId":"arc3-standard","level":"high","sourceLabel":"GPT-6 Luna (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-luna; variant high; public-demo task scores are excluded","score":0.18024216987988248,"warning":null},{"id":"809502577a06f2501fb320bc","modelId":"gpt-6-luna","benchmarkId":"arc3-provider-adapter","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-luna; variant provider-adapter-max; public-demo task scores are excluded","score":0.5868386891554969,"warning":null},{"id":"8127d6db140d93c6a2a6f0fb","modelId":"glm-5.2","benchmarkId":"frontiercode-1.1-main-mini-swe-agent","level":"unspecified","sourceLabel":"GLM 5.2 (none)","configuration":"FrontierCode 1.1; main; 100 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":24.5,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"8141e776db76fc3152040733","modelId":"glm-5.3","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"GLM-5.3 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±3.2 percentage points, confidence level unspecified; release date 2026-08-14; board checked 2026-10-07","score":41.8,"warning":null},{"id":"8181883cb35a2a27684ba034","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Sonnet 5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":44.46,"warning":null},{"id":"820df5b7bcc37f26c5de08f2","modelId":"gpt-6.1-sol","benchmarkId":"arc3-provider-adapter","level":"medium","sourceLabel":"GPT-6.1 Sol (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-1-sol; variant provider-adapter-medium; public-demo task scores are excluded","score":91.02346102095301,"warning":null},{"id":"825888824227e5b5938c688f","modelId":"gpt-6.1-sol","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"GPT-6.1 Sol (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-1-sol; variant provider-adapter-low; public-demo task scores are excluded","score":82.7870673049169,"warning":null},{"id":"826615a0c654de07c920208e","modelId":"grok-4.5","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"low","sourceLabel":"Grok 4.5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":53.14,"warning":null},{"id":"82d8e132cdcc601f53c1c7d6","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-5.6 Terra (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":36.89,"warning":null},{"id":"84d557cc7cf51b4c7f37d128","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":56.10000000000001,"warning":null},{"id":"86eae0dcefd228356b469c27","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Sonnet 5.5 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":64.42999999999999,"warning":null},{"id":"86f7d9616786fcaa781d9a6e","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Opus 5.5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":47.3,"warning":null},{"id":"87757bb60508b774963ad92d","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-6 Luna (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":39.07,"warning":null},{"id":"89b54ce80dc4f23c8c748073","modelId":"grok-4.7","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0067","score":56.989999999999995,"warning":null},{"id":"8be54fd7c609e5e4264162aa","modelId":"gemini-3.7-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":54.74,"warning":null},{"id":"8bfe922d1f0ae206b1b1ed3f","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-5.6 Sol (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0008","score":58.74,"warning":null},{"id":"8c5df9cae8b71015fb4ed7a5","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-6 Luna (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":37.26,"warning":null},{"id":"8d972c585034853563988a88","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-5.6 Terra (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":33.879999999999995,"warning":null},{"id":"8deadee0327ed0621e7829c8","modelId":"deepseek-v4.1-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"DeepSeek V4.1 Flash (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0049","score":56.620000000000005,"warning":null},{"id":"8e0070188f79ecf6c53cc2c3","modelId":"gemini-3.8-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"medium","sourceLabel":"Gemini 3.8 Flash (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":41.19,"warning":null},{"id":"8eb7c9fa2aadd2ad935aa548","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-6 Astra (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":60.25,"warning":null},{"id":"8f0e651f734226844f960671","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-6 Sol (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":47.699999999999996,"warning":null},{"id":"8f27604b87c7e9ff3a19d10d","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-5.4-mini (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":35.06,"warning":null},{"id":"8f57b7f23dea1aca619ea9db","modelId":"claude-opus-4-6","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Opus 4.6 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":38.65,"warning":null},{"id":"8fa122a380d849f4f7cb625e","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-5.4-mini (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":37.19,"warning":null},{"id":"906266917f34d4a9235d8c64","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-5.5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":51.370000000000005,"warning":null},{"id":"909381319a7974fd682e23e2","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Opus 4.7 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":34.910000000000004,"warning":null},{"id":"917cf346b985df23f01b6f50","modelId":"gpt-6-sol","benchmarkId":"arc3-provider-adapter","level":"none","sourceLabel":"GPT-6 Sol (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-sol; variant provider-adapter-none; public-demo task scores are excluded","score":1.2059164341155284,"warning":null},{"id":"9183ca03c441d129fd6a8992","modelId":"claude-opus-4-6","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Opus 4.6 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":32.93,"warning":null},{"id":"919bc7aa1df6867a3297a658","modelId":"gemini-3.7-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"medium","sourceLabel":"Gemini 3.7 Flash (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":56.269999999999996,"warning":null},{"id":"9300a0e1cb70abbef51d728a","modelId":"claude-sonnet-5-5","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"Sonnet 5.5 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±2.9 percentage points, confidence level unspecified; release date 2026-09-28; board checked 2026-10-07","score":61.8,"warning":null},{"id":"93062cdb2abb0e270b7b4742","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Fable 5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":52.73,"warning":null},{"id":"9450f41db1cd130840ad4f8b","modelId":"glm-5.3-flash","benchmarkId":"terminal4-claudecode","level":"none","sourceLabel":"GLM-5.3-Flash (none)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort none; reported margin ±3.5 percentage points, confidence level unspecified; release date 2026-08-26; board checked 2026-10-07","score":35.8,"warning":null},{"id":"95ede679a7d336017a64ca9c","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-6.1 Sol (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":47.980000000000004,"warning":null},{"id":"968a6828c9b1cd6be6b406e5","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-6 Luna (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":50.14999999999999,"warning":null},{"id":"978ae4621dc780ed9f07a974","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Opus 4.7 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0027","score":51.41,"warning":null},{"id":"984e727c945e5302e01c261a","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-5.5 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":42.96,"warning":null},{"id":"997db17ffbb8901dac4b9051","modelId":"grok-4.7","benchmarkId":"frontiercode-1.1-main-grok-build","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.008","score":45.59,"warning":null},{"id":"99e65bc1e02ebf4fee08b548","modelId":"grok-4.5","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"Grok 4.5 (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-5; variant medium; public-demo task scores are excluded","score":0.32,"warning":null},{"id":"9a91cb47980d10cf43433b6c","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Sonnet 5.5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":43.41,"warning":null},{"id":"9b3960b33adb0f097551e5de","modelId":"gemini-3.7-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"low","sourceLabel":"Gemini 3.7 Flash (low)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":36.91,"warning":null},{"id":"9bb740f8253baa1d17ef962a","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-5.5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":30.55,"warning":null},{"id":"9d0d05ab12b44fcb34eddc63","modelId":"gemini-3.8-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":37.96,"warning":null},{"id":"9edbcfed5036b438899d95f7","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-6 Astra (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":63.07000000000001,"warning":null},{"id":"a00271826c96f0bc443bc92d","modelId":"grok-4.6","benchmarkId":"frontiercode-1.1-main-grok-build","level":"low","sourceLabel":"Grok 4.6 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":40.11,"warning":null},{"id":"a089ed32302364f1e074f032","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Haiku 5.5 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":45.83,"warning":null},{"id":"a0921a61b310776b40a299af","modelId":"gpt-6-astra","benchmarkId":"arc3-standard","level":"high","sourceLabel":"GPT-6 Astra (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-astra; variant high; public-demo task scores are excluded","score":54.81905163612231,"warning":null},{"id":"a0e5300dff64fa35e75d78f1","modelId":"gpt-6-astra","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"GPT-6 Astra (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-astra; variant medium; public-demo task scores are excluded","score":38.58757449733784,"warning":null},{"id":"a17bdfe95dc49e5f3a4a8273","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Haiku 5.5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":48.19,"warning":null},{"id":"a22a021efdc0a7fb24189aa1","modelId":"gpt-6.1-sol","benchmarkId":"arc3-standard","level":"high","sourceLabel":"GPT-6.1 Sol (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-1-sol; variant high; public-demo task scores are excluded","score":26.717284844257062,"warning":null},{"id":"a24f67306456001f98c23e75","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-5.4-mini (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":20.78,"warning":null},{"id":"a3f23bb0246903333cdc1388","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Sonnet 5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":52.71,"warning":null},{"id":"a4313e77019bc32a716730f9","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Sonnet 4.6 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":15.03,"warning":null},{"id":"a445e097ad5e39979d7aa81d","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-5.6 Luna (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":15.379999999999999,"warning":null},{"id":"a47e1f9779f4cb809ad3770e","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-main-codex","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":41.31,"warning":null},{"id":"a6b82e375fc2305418bc413e","modelId":"mistral-large-4","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"Mistral Large 4 (Preview) (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.016","score":53.21,"warning":null},{"id":"a7483c339d410fadf3eda32a","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-standard","level":"max","sourceLabel":"DeepSeek V4.1 Flash (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group deepseek-v4-1-flash; variant max; public-demo task scores are excluded","score":0.8516646657220222,"warning":null},{"id":"a80e2e3eda4d53bebbc45613","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-6 Astra (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":48.83,"warning":null},{"id":"a825739422e65a8fe0204db4","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Fable 5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":64.25,"warning":null},{"id":"a889d1ce384914e2806e70c7","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-5.6 Luna (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":50.519999999999996,"warning":null},{"id":"aa2c038898f84e93e7834f06","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-5.5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.003","score":40.56,"warning":null},{"id":"aca1164cd0ccc0467f609f01","modelId":"gemini-3.8-flash","benchmarkId":"arc3-standard","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group google-gemini-3-8-flash; variant high; public-demo task scores are excluded","score":10.36627922392312,"warning":null},{"id":"ae189fcaebaa21a3fe020a5f","modelId":"deepseek-v4-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.2547","score":18.8,"warning":null},{"id":"afbecdd463048a0daf4430c6","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Fable 5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":48.010000000000005,"warning":null},{"id":"b03ff6ec0f4cd41acef948c0","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-6.1 Sol (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":58.089999999999996,"warning":null},{"id":"b1f2936eb97f6064636f0b09","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Opus 4.7 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":35.17,"warning":null},{"id":"b1f91824de3ef9411c7d0e7a","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-5.6 Sol (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":35.44,"warning":null},{"id":"b21f635f0c1b98c573a8cec0","modelId":"gpt-6-astra","benchmarkId":"arc3-standard","level":"low","sourceLabel":"GPT-6 Astra (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-astra; variant low; public-demo task scores are excluded","score":17.452190401133805,"warning":null},{"id":"b2edb78e5bc7a5d9debd2c89","modelId":"claude-fable-5-1","benchmarkId":"terminal4-claudecode","level":"max","sourceLabel":"Fable 5.1 (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Claude Code; effort max; reported margin ±3.8 percentage points, confidence level unspecified; release date 2026-09-01; board checked 2026-10-07","score":57.9,"warning":null},{"id":"b36e1936d9f1400c7a8303c6","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Sonnet 5 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.005","score":42.730000000000004,"warning":null},{"id":"b3e20f0dbe09ce936f270b9e","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-5.6 Sol (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":45.06,"warning":null},{"id":"b3ef1bc9e68ff51e017bc544","modelId":"deepseek-v4-pro","benchmarkId":"frontiercode-1.1-main-mini-swe-agent","level":"unspecified","sourceLabel":"DeepSeek V4 Pro (none)","configuration":"FrontierCode 1.1; main; 100 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.01","score":17.64,"warning":"DeepSeek V4 Pro checkpoint is unspecified; April and August must remain separate."},{"id":"b3f107f9e75ddc6769d67214","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-6 Astra (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":50.94,"warning":null},{"id":"b4505193e0e1f1e917b5cabb","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Fable 5.1 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":61.35,"warning":null},{"id":"b4c8ae568c04a23f30f0b3fd","modelId":"gemini-3.8-flash","benchmarkId":"terminal4-minisweagent","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"Terminal-Bench 4.0.0; 66 tasks; mini-SWE-agent; effort high; reported margin ±3.4 percentage points, confidence level unspecified; release date 2026-09-02; board checked 2026-10-07","score":19.1,"warning":null},{"id":"b54fdd6462a56cf9fbbff457","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Fable 5 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":64.94,"warning":null},{"id":"b56938a2e49c9671f1657923","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Opus 4.8 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0033","score":59.45,"warning":null},{"id":"b587e42c0ff834839006d87f","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":59.809999999999995,"warning":null},{"id":"b5b2f35a4688fb053d31b0b3","modelId":"mistral-large-4","benchmarkId":"frontiercode-1.1-main-chisel","level":"unspecified","sourceLabel":"Mistral Large 4 (Preview) (none)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0357","score":27.87,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"b6b9c826a49e8b6040e7ec1c","modelId":"gpt-6-sol","benchmarkId":"arc3-standard","level":"high","sourceLabel":"GPT-6 Sol (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-sol; variant high; public-demo task scores are excluded","score":0.8994389911152361,"warning":null},{"id":"b7467edb46036b2935325b6c","modelId":"grok-4.5","benchmarkId":"frontiercode-1.1-extended-grok-build","level":"medium","sourceLabel":"Grok 4.5 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; grok-build; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":56.07,"warning":null},{"id":"b7d96566d6b6b34d9e314f79","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":24.310000000000002,"warning":null},{"id":"b8e9f7152e70e6a96533cb76","modelId":"gpt-6-luna","benchmarkId":"arc3-provider-adapter","level":"medium","sourceLabel":"GPT-6 Luna (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-luna; variant provider-adapter-medium; public-demo task scores are excluded","score":0.31612009238177263,"warning":null},{"id":"b911c4d9f13d8188e2de86a1","modelId":"glm-5.3-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":31.830000000000002,"warning":null},{"id":"b95d4bdbd7d9a7c813c65587","modelId":"gemini-4-argon","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0103","score":45.87,"warning":null},{"id":"b99056688b2b172aff603206","modelId":"gpt-6-astra","benchmarkId":"arc3-provider-adapter","level":"low","sourceLabel":"GPT-6 Astra (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-astra; variant provider-adapter-low; public-demo task scores are excluded","score":98.02838165829648,"warning":null},{"id":"b9bb4a797f74eb3290bd59f2","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Opus 5.5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0067","score":60.33,"warning":null},{"id":"b9cd81a3a14751517d70b67b","modelId":"grok-4.7","benchmarkId":"arc3-standard","level":"low","sourceLabel":"Grok 4.7 (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-7; variant low; public-demo task scores are excluded","score":0.24696281908456538,"warning":null},{"id":"ba391b95e256d194ae244169","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Opus 5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0021","score":41.949999999999996,"warning":null},{"id":"ba416adf6ebdd6333f9b4700","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-standard","level":"low","sourceLabel":"DeepSeek V4.1 Flash (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group deepseek-v4-1-flash; variant low; public-demo task scores are excluded","score":0.34514247315937174,"warning":null},{"id":"ba8ff35e294d29a743fa5c65","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Opus 4.8 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.005","score":45.49,"warning":null},{"id":"bb19199d941a9b7b0ad30caa","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Sonnet 5.5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":36.480000000000004,"warning":null},{"id":"bbfde4d573412eed7593e750","modelId":"gpt-6-luna","benchmarkId":"arc3-standard","level":"low","sourceLabel":"GPT-6 Luna (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-luna; variant low; public-demo task scores are excluded","score":0.03291205821227175,"warning":null},{"id":"bc2becdddfac48a99983235f","modelId":"gpt-5.6-terra","benchmarkId":"terminal4-codex","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Codex; effort max; reported margin ±3.3 percentage points, confidence level unspecified; release date 2026-06-26; board checked 2026-10-07","score":21.5,"warning":null},{"id":"bc4e683318a11b6833ff21c6","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Opus 5.5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":54.64,"warning":null},{"id":"bc72c2627601d13388cbfc75","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-main-codex","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":53.26,"warning":null},{"id":"bced9c84bedddd36c1d85054","modelId":"claude-opus-4-6","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Opus 4.6 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":23.669999999999998,"warning":null},{"id":"bd377043899ff0ae0b462963","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Sonnet 4.6 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":37.49,"warning":null},{"id":"be5258d2d7a78bcec434049d","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":47.49,"warning":null},{"id":"bef3e60369b8392a3a3820b6","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-6 Sol (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":57.11000000000001,"warning":null},{"id":"bf209e5b4b2da520e58b7863","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Fable 5.1 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":48.730000000000004,"warning":null},{"id":"bf764fb9af20b8b9fff21eb8","modelId":"gpt-5.6-terra","benchmarkId":"frontiercode-1.1-extended-codex","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":55.84,"warning":null},{"id":"bfa114d125622b9e97e9b404","modelId":"gemini-4-argon","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0137","score":59.38,"warning":null},{"id":"bfe462efb459731aff29651b","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Fable 5.1 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":50.339999999999996,"warning":null},{"id":"c02f418374e14634d8639cb0","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":51.57000000000001,"warning":null},{"id":"c0a21b210841792f09e6eaa9","modelId":"gemini-3.6-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"medium","sourceLabel":"Gemini 3.6 Flash (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":34.37,"warning":null},{"id":"c16b09ec865ee083d20ed15f","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":59.599999999999994,"warning":null},{"id":"c2cebdbb75fc439d384a1645","modelId":"qwen3.7-plus","benchmarkId":"frontiercode-1.1-main-mini-swe-agent","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (none)","configuration":"FrontierCode 1.1; main; 100 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.009","score":10.2,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"c30c3f0dc62b17786a57f7cc","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Opus 5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":58.45,"warning":null},{"id":"c3ea3e1488fb895ac3920c7b","modelId":"grok-4.7","benchmarkId":"arc3-standard","level":"high","sourceLabel":"Grok 4.7 (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-7; variant high; public-demo task scores are excluded","score":1.7332202934667684,"warning":null},{"id":"c49f9aec89220786fdfb9d2f","modelId":"grok-4.7","benchmarkId":"arc3-provider-adapter","level":"medium","sourceLabel":"Grok 4.7 (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group xai-grok-4-7; variant provider-adapter-medium; public-demo task scores are excluded","score":8.373637088154913,"warning":null},{"id":"c551f728a54b7fd7fc18686b","modelId":"gpt-6.1-sol","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"GPT-6.1 Sol (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-1-sol; variant provider-adapter-high; public-demo task scores are excluded","score":94.99362560053225,"warning":null},{"id":"c6fcba70fe006c20bead04c2","modelId":"grok-4.7","benchmarkId":"arc3-standard","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-7; variant xhigh; public-demo task scores are excluded","score":1.8162526536637038,"warning":null},{"id":"c722f195ec4769f24887bfb4","modelId":"gpt-6-sol","benchmarkId":"arc3-standard","level":"low","sourceLabel":"GPT-6 Sol (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-sol; variant low; public-demo task scores are excluded","score":0.12649234250931635,"warning":null},{"id":"c73f184464b158eb37866926","modelId":"gemini-4-argon","benchmarkId":"arc3-standard","level":"low","sourceLabel":"Gemini 4 Argon (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group google-gemini-4-argon; variant low; public-demo task scores are excluded","score":49.79401633943577,"warning":null},{"id":"c7981128f2cb3799b53b66ec","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Opus 4.8 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":50.83,"warning":null},{"id":"c9b43e92020f700c0d052db9","modelId":"grok-4.5","benchmarkId":"frontiercode-1.1-main-grok-build","level":"low","sourceLabel":"Grok 4.5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":37.86,"warning":null},{"id":"c9e1fea0c53568fa87547d39","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-5.6 Sol (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":39.93,"warning":null},{"id":"c9ea93dd77a44efafd8b2823","modelId":"mistral-medium-3.5","benchmarkId":"frontiercode-1.1-main-chisel","level":"unspecified","sourceLabel":"Mistral 3.5 Medium (none)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.006","score":8,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"ca8bd8b35520f3df29860eb1","modelId":"gemini-3.6-flash","benchmarkId":"frontiercode-1.1-extended-chisel","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":47.58,"warning":null},{"id":"cba8b60609ec8ba2d384bac4","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"xhigh","sourceLabel":"Claude Opus 5 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0097","score":56.87,"warning":null},{"id":"cc50f474e4bed017002a068b","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-5.6 Luna (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":25.740000000000002,"warning":null},{"id":"cf038268638355842fabacec","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":60.550000000000004,"warning":null},{"id":"d1018d5f46b97e159d2f0a02","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":55.059999999999995,"warning":null},{"id":"d2bdad5d6d457b8f28281e8f","modelId":"glm-5.2","benchmarkId":"frontiercode-1.1-extended-mini-swe-agent","level":"unspecified","sourceLabel":"GLM 5.2 (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":40.050000000000004,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"d45e011559e01576731fce1b","modelId":"gpt-6-sol","benchmarkId":"arc3-provider-adapter","level":"medium","sourceLabel":"GPT-6 Sol (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-sol; variant provider-adapter-medium; public-demo task scores are excluded","score":4.930246053628116,"warning":null},{"id":"d4b90ed9c9a977ac490e95b4","modelId":"gemini-3.6-flash","benchmarkId":"frontiercode-1.1-main-chisel","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"FrontierCode 1.1; main; 100 tasks; chisel; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":33.69,"warning":null},{"id":"d4de4a5ced8821eb86459630","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-5.5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":45.550000000000004,"warning":null},{"id":"d521363d88a93e4d46434c3c","modelId":"gpt-5.6-sol","benchmarkId":"terminal4-codex","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Codex; effort max; reported margin ±3.8 percentage points, confidence level unspecified; release date 2026-06-26; board checked 2026-10-07","score":37.3,"warning":null},{"id":"d667c01aa76eb58a13006db0","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"medium","sourceLabel":"GPT-5.6 Sol (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":54.67999999999999,"warning":null},{"id":"d69d0f9d7823430df03fa93d","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":49.27,"warning":null},{"id":"d802102582079f98ea102e70","modelId":"grok-4.5","benchmarkId":"terminal4-grokbuild","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Grok Build; effort high; reported margin ±2.6 percentage points, confidence level unspecified; release date 2026-07-16; board checked 2026-10-07","score":12.4,"warning":null},{"id":"d828110f6fc35356fb3fac50","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Opus 5.5 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.008","score":54.43,"warning":null},{"id":"d8ae0e163cd2b653449b803b","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-6 Astra (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":45.269999999999996,"warning":null},{"id":"d9ff43c2ee3145f93fa639e1","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Opus 4.7 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":43.6,"warning":null},{"id":"da08a3f2efd54d7cb540441b","modelId":"gpt-5.6-luna","benchmarkId":"arc3-standard","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-luna; variant max; public-demo task scores are excluded","score":0.18,"warning":null},{"id":"dabbb2909497af3a1a745de6","modelId":"gpt-6-sol","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"GPT-6 Sol (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-sol; variant provider-adapter-high; public-demo task scores are excluded","score":5.855663760579018,"warning":null},{"id":"dae492a117175343395ae164","modelId":"claude-fable-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Fable 5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":60.8,"warning":null},{"id":"daf9afd286691adc70b10a5b","modelId":"gpt-5.6-terra","benchmarkId":"arc3-standard","level":"high","sourceLabel":"GPT-5.6 Terra (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-terra; variant high; public-demo task scores are excluded","score":0.49,"warning":null},{"id":"db8204b5e5c7174d88f5e8a4","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-5.4-mini (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":9.25,"warning":null},{"id":"dbcf7bd488b0b930033a6cf7","modelId":"grok-4.7","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"Grok 4.7 (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group xai-grok-4-7; variant provider-adapter-high; public-demo task scores are excluded","score":10.047981644646995,"warning":null},{"id":"dbd587172dfdcad52e7e40e2","modelId":"claude-opus-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"xhigh","sourceLabel":"Claude Opus 5.5 (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":51.42,"warning":null},{"id":"dc357b19b6bea4e41d45b423","modelId":"gpt-5.6-luna","benchmarkId":"arc3-standard","level":"low","sourceLabel":"GPT-5.6 Luna (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-luna; variant low; public-demo task scores are excluded","score":0.16999999999999998,"warning":null},{"id":"dcbdda875af63c9d559ccae4","modelId":"grok-4.6","benchmarkId":"arc3-standard","level":"xhigh","sourceLabel":"Grok 4.6 (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-6; variant xhigh; public-demo task scores are excluded","score":2.11,"warning":null},{"id":"dd27c86aaec46afa668fb3ed","modelId":"grok-4.5","benchmarkId":"arc3-standard","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group xai-grok-4-5; variant high; public-demo task scores are excluded","score":0.3,"warning":null},{"id":"ddfef6cc7d43c0b169169e45","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"low","sourceLabel":"Claude Opus 5 (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0041","score":55.75,"warning":null},{"id":"de359ceaf316774036c9cd92","modelId":"grok-4.5","benchmarkId":"frontiercode-1.1-main-grok-build","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":42.44,"warning":null},{"id":"dea717013f27270a23b74f90","modelId":"gpt-6-astra","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-6 Astra (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":62.12,"warning":null},{"id":"e0e844ac4fa5c437d7f9afe6","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Sonnet 5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":39.35,"warning":null},{"id":"e146b9770f760fcaf7655c2c","modelId":"gemini-3.8-flash","benchmarkId":"arc3-provider-adapter","level":"medium","sourceLabel":"Gemini 3.8 Flash (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group google-gemini-3-8-flash; variant provider-adapter-medium; public-demo task scores are excluded","score":24.202718056957025,"warning":null},{"id":"e182040a6b20731b3805613d","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Fable 5.1 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":63.6,"warning":null},{"id":"e2c3b944130f3a5e4b806d94","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Opus 4.8 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":35.14,"warning":null},{"id":"e46aa5212aad528a859b9813","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"medium","sourceLabel":"GPT-6 Luna (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":35.53,"warning":null},{"id":"e554b93428a677182f0b4906","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"low","sourceLabel":"Claude Sonnet 5 (low)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":28.71,"warning":null},{"id":"e5f0edaa48b4e6162542d4d5","modelId":"minimax-m3","benchmarkId":"frontiercode-1.1-extended-mini-swe-agent","level":"unspecified","sourceLabel":"MiniMax M3 (none)","configuration":"FrontierCode 1.1; extended; 150 tasks; mini-swe-agent; source effort none; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0107","score":28.54,"warning":"The evaluator reports no named effort control. This does not identify a non-reasoning configuration."},{"id":"e61670a23ffd1d53d4575c95","modelId":"gpt-5.6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-5.6 Sol (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":59.98,"warning":null},{"id":"e638e44f7a00fc2c77bd320a","modelId":"gpt-5.4-mini","benchmarkId":"frontiercode-1.1-main-codex","level":"high","sourceLabel":"GPT-5.4-mini (high)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.001","score":23.080000000000002,"warning":null},{"id":"e78de1fe1b2913b438e8bfb4","modelId":"gpt-5.6-luna","benchmarkId":"frontiercode-1.1-extended-codex","level":"low","sourceLabel":"GPT-5.6 Luna (low)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":29.020000000000003,"warning":null},{"id":"e7a90181ff22c2632f726d32","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-extended-codex","level":"high","sourceLabel":"GPT-5.5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":53.99,"warning":null},{"id":"e83c78ee104f6ac791b27a26","modelId":"grok-4.6","benchmarkId":"frontiercode-1.1-main-grok-build","level":"medium","sourceLabel":"Grok 4.6 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.006","score":46.06,"warning":null},{"id":"e93116c3e86f122b9ce063bf","modelId":"gpt-6-astra","benchmarkId":"arc3-provider-adapter","level":"none","sourceLabel":"GPT-6 Astra (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-astra; variant provider-adapter-none; public-demo task scores are excluded","score":96.72021831785531,"warning":"Source-reported None does not establish a supported Astra API setting."},{"id":"e97f395dde5500c89744b1bc","modelId":"gpt-5.6-terra","benchmarkId":"arc3-standard","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-terra; variant max; public-demo task scores are excluded","score":0.8,"warning":null},{"id":"e98091f1432b7fd7d85239e6","modelId":"gpt-6.1-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":47.57,"warning":null},{"id":"ea4aa12006abea91afb9c17c","modelId":"gemini-3.8-flash","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group google-gemini-3-8-flash; variant provider-adapter-high; public-demo task scores are excluded","score":35.00424335075243,"warning":null},{"id":"eaf538a9130f43391ba85bbf","modelId":"grok-4.5","benchmarkId":"frontiercode-1.1-main-grok-build","level":"medium","sourceLabel":"Grok 4.5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; grok-build; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":41.93,"warning":null},{"id":"ebc8a87de8e88db13b42eefb","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Haiku 5.5 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":58.43000000000001,"warning":null},{"id":"ec3c20ed5f0db0ac92661a10","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"low","sourceLabel":"GPT-6 Sol (low)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort low; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":37.28,"warning":null},{"id":"ecfb916e25d392c524f62e21","modelId":"gpt-6-luna","benchmarkId":"arc3-standard","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-luna; variant max; public-demo task scores are excluded","score":0.10360997760319558,"warning":null},{"id":"ed4a5360238dda5d9ed009f8","modelId":"grok-4.6","benchmarkId":"terminal4-grokbuild","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Grok Build; effort high; reported margin ±3.1 percentage points, confidence level unspecified; release date 2026-08-12; board checked 2026-10-07","score":20.3,"warning":null},{"id":"edc50b3b264c08f97d553277","modelId":"deepseek-v4.1-flash","benchmarkId":"arc3-provider-adapter","level":"none","sourceLabel":"DeepSeek V4.1 Flash (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group deepseek-v4-1-flash; variant provider-adapter-none; public-demo task scores are excluded","score":0.4830935313474596,"warning":null},{"id":"ee65b4047c01418e7d2991a8","modelId":"gpt-6.1-sol","benchmarkId":"terminal4-codex","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Codex; effort max; reported margin ±3.1 percentage points, confidence level unspecified; release date 2026-09-29; board checked 2026-10-07","score":58.2,"warning":null},{"id":"ef3184b2503a57f0077f8932","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":62.029999999999994,"warning":null},{"id":"efecb437a4948047ef73bc35","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-6 Sol (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0013","score":59.13,"warning":null},{"id":"f019be82c0b7b5fcd5e1527b","modelId":"gpt-5.6-terra","benchmarkId":"arc3-standard","level":"low","sourceLabel":"GPT-5.6 Terra (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-terra; variant low; public-demo task scores are excluded","score":0.01,"warning":null},{"id":"f069bb43a49e2e73d5a979fd","modelId":"claude-sonnet-4-6","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Sonnet 4.6 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":23.47,"warning":null},{"id":"f0c644f9ca0c8b80a60b8f6e","modelId":"claude-opus-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Opus 5 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.006","score":53.38,"warning":null},{"id":"f1474950f0a1512219e1d401","modelId":"claude-fable-5-1","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Fable 5.1 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":62.68,"warning":null},{"id":"f1b6cbed605d02c34efca6af","modelId":"gpt-6.1-sol","benchmarkId":"arc3-standard","level":"low","sourceLabel":"GPT-6.1 Sol (low)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-1-sol; variant low; public-demo task scores are excluded","score":3.9243846587314177,"warning":null},{"id":"f1edd38ed72382bd57766812","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"medium","sourceLabel":"Claude Opus 4.8 (medium)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.004","score":55.37,"warning":null},{"id":"f1f80cdfdf2fedc08b3c8e7d","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":53.93,"warning":null},{"id":"f2c8f1aa900334eca9b374f2","modelId":"claude-haiku-5-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Haiku 5.5 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":55.879999999999995,"warning":null},{"id":"f38a139c5a520dd21aaf9132","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.003","score":42.36,"warning":null},{"id":"f55940d0d324a594d2e7cb02","modelId":"gpt-5.6-sol","benchmarkId":"arc3-standard","level":"high","sourceLabel":"GPT-5.6 Sol (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-5-6-sol; variant high; public-demo task scores are excluded","score":2.15,"warning":null},{"id":"f6c40a31975168e070d2cb4c","modelId":"claude-sonnet-5","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0053","score":54.85,"warning":null},{"id":"f7f350725fc81f9c8eb65b9d","modelId":"gpt-6-sol","benchmarkId":"frontiercode-1.1-main-codex","level":"xhigh","sourceLabel":"GPT-6 Sol (xhigh)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.002","score":48.449999999999996,"warning":null},{"id":"f8386742f94346ed03acace3","modelId":"gpt-6-luna","benchmarkId":"arc3-provider-adapter","level":"none","sourceLabel":"GPT-6 Luna (none)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-luna; variant provider-adapter-none; public-demo task scores are excluded","score":0.04644044435594966,"warning":null},{"id":"fa5c692bcb2cab6ed2609402","modelId":"gpt-5.5","benchmarkId":"frontiercode-1.1-extended-codex","level":"xhigh","sourceLabel":"GPT-5.5 (xhigh)","configuration":"FrontierCode 1.1; extended; 150 tasks; codex; source effort xhigh; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0027","score":56.66,"warning":null},{"id":"fa6ed88eb7e904aa6955ebfb","modelId":"claude-opus-4-7","benchmarkId":"frontiercode-1.1-extended-claude-code","level":"high","sourceLabel":"Claude Opus 4.7 (high)","configuration":"FrontierCode 1.1; extended; 150 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.0007","score":51.12,"warning":null},{"id":"fa884927bb3db5f6e643e385","modelId":"gpt-6-luna","benchmarkId":"frontiercode-1.1-main-codex","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"FrontierCode 1.1; main; 100 tasks; codex; source effort max; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":42.42,"warning":null},{"id":"fac6171dfd16505cbe59bdf4","modelId":"gpt-6.1-sol","benchmarkId":"arc3-provider-adapter","level":"xhigh","sourceLabel":"GPT-6.1 Sol (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-1-sol; variant provider-adapter-xhigh; public-demo task scores are excluded","score":96.37493342875182,"warning":null},{"id":"fb309a77f991373cd200cb6f","modelId":"gpt-6-luna","benchmarkId":"terminal4-codex","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Codex; effort max; reported margin ±2.7 percentage points, confidence level unspecified; release date 2026-09-22; board checked 2026-10-07","score":16.4,"warning":null},{"id":"fb4c2a49491c2edb9bfecada","modelId":"gpt-6-luna","benchmarkId":"arc3-provider-adapter","level":"high","sourceLabel":"GPT-6 Luna (high)","configuration":"ARC-AGI-3 Semi-Private verified summary; Provider Adapter harness; source group openai-gpt-6-luna; variant provider-adapter-high; public-demo task scores are excluded","score":0.37847731587690087,"warning":null},{"id":"fc2e2593cadcc2e0ec53f01d","modelId":"gpt-6-sol","benchmarkId":"terminal4-codex","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Codex; effort max; reported margin ±3.2 percentage points, confidence level unspecified; release date 2026-09-22; board checked 2026-10-07","score":49.4,"warning":null},{"id":"fd19565d5d4088b6dfc79064","modelId":"claude-opus-4-8","benchmarkId":"frontiercode-1.1-main-claude-code","level":"medium","sourceLabel":"Claude Opus 4.8 (medium)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort medium; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0.005","score":40.46,"warning":null},{"id":"fdf647585fd6368c5653e162","modelId":"claude-sonnet-5-5","benchmarkId":"frontiercode-1.1-main-claude-code","level":"high","sourceLabel":"Claude Sonnet 5.5 (high)","configuration":"FrontierCode 1.1; main; 100 tasks; claude-code; source effort high; weighted rubric score, not pass rate; unfair internet runs zeroed by evaluator; flag rate 0","score":49.38,"warning":null},{"id":"fe590a70de3d448e4da4dc99","modelId":"muse-spark-1.3","benchmarkId":"terminal4-musecode","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"Terminal-Bench 4.0.0; 66 tasks; Muse Code; effort xhigh; reported margin ±2.9 percentage points, confidence level unspecified; release date 2026-09-02; board checked 2026-10-07","score":14.5,"warning":null},{"id":"feb7d0ef7a4bc1a3de0566c8","modelId":"gpt-6-sol","benchmarkId":"arc3-standard","level":"medium","sourceLabel":"GPT-6 Sol (medium)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-sol; variant medium; public-demo task scores are excluded","score":0.418612306617149,"warning":null},{"id":"ff9e159dda586ac4a0349062","modelId":"gpt-6-astra","benchmarkId":"arc3-standard","level":"xhigh","sourceLabel":"GPT-6 Astra (xhigh)","configuration":"ARC-AGI-3 Semi-Private verified summary; Standard harness; source group openai-gpt-6-astra; variant xhigh; public-demo task scores are excluded","score":59.342954091460555,"warning":null},{"id":"01e44fc757e76a75628ede29","modelId":"claude-opus-5","benchmarkId":"scale-sweatlas-qna-claudecode","level":"xhigh","sourceLabel":"Opus 5 (Claude Code) xHigh","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-08-04T16:11:53.000Z; reported margin ±5.01 percentage points; No contamination annotation reported (not proof of absence)","score":63.17,"warning":null},{"id":"07b714836f4d50d2545a5455","modelId":"gpt-5.4","benchmarkId":"scale-sweatlas-tw-codex","level":"xhigh","sourceLabel":"GPT-5.4 (Codex) xHigh","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-26T18:24:53.000Z; reported margin ±6.04 percentage points; No contamination annotation reported (not proof of absence)","score":44.36,"warning":null},{"id":"0e5a24ad5a3aa10c27a55eb7","modelId":"claude-fable-5-1","benchmarkId":"scale-sweatlas-tw-claudecode","level":"xhigh","sourceLabel":"Fable-5.1 (Claude Code) xHigh*","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-09-02T19:02:52.000Z; reported margin ±5.33 percentage points; No contamination annotation reported (not proof of absence)","score":67.04,"warning":"Starred protocol annotation requires row-specific review before ranking."},{"id":"0eb870f9f6496683cda2ff7a","modelId":"gpt-6-astra","benchmarkId":"scale-sweatlas-tw-codex","level":"xhigh","sourceLabel":"GPT 6 Astra (Codex) xHigh","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-09-09T19:19:03.000Z; reported margin ±5.91 percentage points; No contamination annotation reported (not proof of absence)","score":50.74,"warning":null},{"id":"129ef34aa55c39c9f65a8288","modelId":"claude-fable-5-1","benchmarkId":"scale-sweatlas-refactoring-claudecode","level":"xhigh","sourceLabel":"Fable-5.1 (Claude Code) xHigh","configuration":"70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.; published row 2026-09-02T21:52:32.000Z; reported margin ±6.52 percentage points; No contamination annotation reported (not proof of absence)","score":56.67,"warning":null},{"id":"1ce57fce710e63f4fde5f000","modelId":"claude-opus-4-6","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"non-reasoning","sourceLabel":"claude-opus-4-6 (Non-Thinking)","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2026-02-17T17:05:29.000Z; reported margin ±1.54 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":19,"warning":null},{"id":"22863bca1313db0d07725e03","modelId":"claude-fable-5","benchmarkId":"scale-sweatlas-qna-claudecode","level":"xhigh","sourceLabel":"Fable-5 (Claude Code) xHigh*","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-02T20:20:06.000Z; reported margin ±5 percentage points; No contamination annotation reported (not proof of absence); Board footnote: elevated refusals on benign tasks are scored zero by the evaluator and remain in the published score.","score":39,"warning":null},{"id":"22f827b71cc6c1cb3e229440","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929-thinking","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2025-10-02T17:29:42.000Z; reported margin ±1.47 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. ","score":14.09,"warning":null},{"id":"25753b4cde30cf31c9192afb","modelId":"gpt-5.1","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"thinking","sourceLabel":"gpt-5.1-thinking","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2025-11-26T20:24:19.000Z; reported margin ±1.67 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":23.68,"warning":null},{"id":"294120a57d522f24c679c51f","modelId":"gpt-5.6-sol","benchmarkId":"scale-sweatlas-qna-codex","level":"xhigh","sourceLabel":"GPT-5.6-Sol (Codex) xHigh*","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-02T20:20:06.000Z; reported margin ±5 percentage points; No contamination annotation reported (not proof of absence); Board footnote: elevated refusals on benign tasks are scored zero by the evaluator and remain in the published score.","score":46,"warning":null},{"id":"2b65412a084daf79b87e4b34","modelId":"gpt-5.1","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"thinking","sourceLabel":"gpt-5.1-thinking","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2025-11-26T20:28:20.000Z; reported margin ±1.82 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":24.65,"warning":null},{"id":"308c1446bec689a5b5f02280","modelId":"gpt-6-astra","benchmarkId":"scale-sweatlas-refactoring-codex","level":"xhigh","sourceLabel":"GPT 6 Astra (Codex) xHigh","configuration":"70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.; published row 2026-09-09T19:16:22.000Z; reported margin ±6.43 percentage points; No contamination annotation reported (not proof of absence)","score":59.05,"warning":null},{"id":"3203239d5cb0c63824c2a86d","modelId":"claude-fable-5","benchmarkId":"scale-enigma_eval-scaleofficialevaluation","level":"high","sourceLabel":"claude-fable-5-high","configuration":"Scale EnigmaEval puzzle-hunt accuracy; source-reported reasoning effort.; published row 2026-03-23T21:18:50.000Z; reported margin ±2.8 percentage points; No contamination annotation reported (not proof of absence)","score":39.28,"warning":null},{"id":"34bc44371787dfc935087f65","modelId":"gpt-5.4","benchmarkId":"scale-sweatlas-tw-miniswe","level":"xhigh","sourceLabel":"GPT-5.4 (Mini-SWE) xHigh","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-26T18:24:57.000Z; reported margin ±6 percentage points; No contamination annotation reported (not proof of absence)","score":40,"warning":null},{"id":"3668be0998ee9f9dbe1cb4b4","modelId":"claude-opus-4-6","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"max","sourceLabel":"claude-opus-4-6-thinking-max","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2026-02-17T17:06:06.000Z; reported margin ±2.03 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":36.24,"warning":null},{"id":"3836c4d32eeadb2f5ad4990c","modelId":"gpt-5.5","benchmarkId":"scale-sweatlas-refactoring-codex","level":"xhigh","sourceLabel":"GPT-5.5 (Codex) xHigh","configuration":"70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.; published row 2026-05-06T01:44:12.000Z; reported margin ±6.76 percentage points; No contamination annotation reported (not proof of absence)","score":44.79,"warning":null},{"id":"38da8dccc01df7999a09d824","modelId":"claude-opus-4-5-20251101","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"thinking","sourceLabel":"claude-opus-4-5-20251101-thinking","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2025-11-26T20:27:42.000Z; reported margin ±1.86 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":26.32,"warning":null},{"id":"445b1683e6ecba6cbf9f044a","modelId":"gpt-5.5","benchmarkId":"scale-sweatlas-qna-codex","level":"xhigh","sourceLabel":"GPT 5.5 (Codex) xHigh","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-05-07T15:50:31.000Z; reported margin ±5.08 percentage points; No contamination annotation reported (not proof of absence)","score":45.43,"warning":null},{"id":"488e21e807236e118e34fcf6","modelId":"claude-fable-5","benchmarkId":"scale-sweatlas-refactoring-claudecode","level":"xhigh","sourceLabel":"Fable-5 (Claude Code) xHigh","configuration":"70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.; published row 2026-06-11T17:55:16.000Z; reported margin ±6.76 percentage points; No contamination annotation reported (not proof of absence)","score":54.76,"warning":null},{"id":"595d203626f42c7eaf274a4d","modelId":"gpt-5.4","benchmarkId":"scale-sweatlas-refactoring-codex","level":"xhigh","sourceLabel":"GPT 5.4 (Codex) xHigh","configuration":"70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.; published row 2026-05-06T01:44:16.000Z; reported margin ±6.76 percentage points; No contamination annotation reported (not proof of absence)","score":44.29,"warning":null},{"id":"5982871718a1d5284ac2e401","modelId":"claude-opus-4-6","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"max","sourceLabel":"claude-opus-4-6-thinking-max","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2026-02-17T17:04:32.000Z; reported margin ±1.86 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":34.44,"warning":null},{"id":"5d521eed35c933b8c0cf6ca3","modelId":"gpt-5.6-sol","benchmarkId":"scale-enigma_eval-scaleofficialevaluation","level":"high","sourceLabel":"gpt-5.6-sol-high","configuration":"Scale EnigmaEval puzzle-hunt accuracy; source-reported reasoning effort.; published row 2026-03-25T18:48:38.000Z; reported margin ±2.8 percentage points; No contamination annotation reported (not proof of absence)","score":37.12,"warning":null},{"id":"60804685f644d5cf0ad52320","modelId":"gemini-3.5-flash","benchmarkId":"scale-enigma_eval-scaleofficialevaluation","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"Scale EnigmaEval puzzle-hunt accuracy; source-reported reasoning effort.; published row 2025-11-19T23:53:27.000Z; reported margin ±2.45 percentage points; No contamination annotation reported (not proof of absence)","score":25.41,"warning":null},{"id":"60b0d52dd406e2a990549468","modelId":"gpt-6-astra","benchmarkId":"scale-sweatlas-qna-codex","level":"xhigh","sourceLabel":"GPT 6 Astra (Codex) xHigh*","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-09-09T19:17:58.000Z; reported margin ±4.88 percentage points; No contamination annotation reported (not proof of absence)","score":59.14,"warning":"Starred protocol annotation requires row-specific review before ranking."},{"id":"6c79741b5a7ecef0ff4d0346","modelId":"muse-spark-1.1","benchmarkId":"scale-sweatlas-qna-miniswe","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Mini-SWE-Agent) xHigh","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-07-09T21:00:32.000Z; reported margin ±5.08 percentage points; No contamination annotation reported (not proof of absence)","score":42.2,"warning":null},{"id":"72c40b8902056dacea8cafd8","modelId":"claude-sonnet-5-5","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"xhigh","sourceLabel":"Sonnet 5.5 (xhigh)","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2026-10-12T23:23:20.000Z; reported margin ±1.95 percentage points; No contamination annotation reported (not proof of absence)","score":46.12,"warning":null},{"id":"79db12d910dc9b17468fc07c","modelId":"gpt-5.4","benchmarkId":"scale-sweatlas-qna-codex","level":"xhigh","sourceLabel":"GPT 5.4 (Codex) xHigh","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-09T16:27:54.000Z; reported margin ±5.1 percentage points; No contamination annotation reported (not proof of absence)","score":40.8,"warning":null},{"id":"8141013d16a793653a973079","modelId":"claude-opus-4-8","benchmarkId":"scale-enigma_eval-scaleofficialevaluation","level":"xhigh","sourceLabel":"claude-opus-4-8-xhigh","configuration":"Scale EnigmaEval puzzle-hunt accuracy; source-reported reasoning effort.; published row 2026-03-10T21:12:05.000Z; reported margin ±2.41 percentage points; No contamination annotation reported (not proof of absence)","score":23.51,"warning":null},{"id":"8154d53aea81c7edfd9708c9","modelId":"muse-spark-1.1","benchmarkId":"scale-sweatlas-tw-miniswe","level":"xhigh","sourceLabel":"Muse Spark 1.1 (Mini-SWE-Agent) xHigh","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-07-09T21:02:58.000Z; reported margin ±5.96 percentage points; No contamination annotation reported (not proof of absence)","score":41.48,"warning":null},{"id":"8adcbb20f22ee3ada097baa2","modelId":"claude-fable-5-1","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"xhigh","sourceLabel":"Fable 5.1 (xhigh)","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2026-09-03T18:19:45.000Z; reported margin ±2.1 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":46.8,"warning":null},{"id":"8b77418f10a0d2c871504615","modelId":"claude-fable-5-1","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"xhigh","sourceLabel":"Fable 5.1 (xhigh)","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2026-09-03T18:22:21.000Z; reported margin ±2 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":46.5,"warning":null},{"id":"9a7a2837429ecbebf2d957b1","modelId":"claude-opus-4-6","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"non-reasoning","sourceLabel":"claude-opus-4-6 (Non-Thinking)","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2026-02-17T17:06:51.000Z; reported margin ±1.67 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. ","score":19.37,"warning":null},{"id":"a71c17d0e87547e9f30a79b2","modelId":"claude-opus-4-5-20251101","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"thinking","sourceLabel":"claude-opus-4-5-20251101-thinking","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2025-11-26T20:23:24.000Z; reported margin ±1.7 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":25.2,"warning":null},{"id":"b406780519719b231c94f076","modelId":"claude-opus-5","benchmarkId":"scale-sweatlas-tw-claudecode","level":"xhigh","sourceLabel":"Opus 5 (Claude Code) xHigh","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-08-04T16:09:52.000Z; reported margin ±5.58 percentage points; No contamination annotation reported (not proof of absence)","score":62.22,"warning":null},{"id":"ba265c887f15f603db7f5f80","modelId":"claude-opus-4-8","benchmarkId":"scale-sweatlas-qna-claudecode","level":"xhigh","sourceLabel":"Opus 4.8 (Claude Code) xhigh","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-06-08T20:50:16.000Z; reported margin ±4.93 percentage points; No contamination annotation reported (not proof of absence)","score":57.26,"warning":null},{"id":"bfc3139a28a8c37b769ae912","modelId":"claude-fable-5-1","benchmarkId":"scale-sweatlas-qna-claudecode","level":"xhigh","sourceLabel":"Fable-5.1 (Claude Code) xHigh*","configuration":"124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-02T20:57:11.000Z; reported margin ±4.85 percentage points; No contamination annotation reported (not proof of absence)","score":59.95,"warning":"Starred protocol annotation requires row-specific review before ranking."},{"id":"c457ac399b0c7c755bf0adbb","modelId":"o3-mini","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"high","sourceLabel":"o3 mini (high)","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2025-04-10T19:29:02.000Z; reported margin ±1.44 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":13.37,"warning":null},{"id":"caa92a3ee80886991f302d07","modelId":"o3-mini","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"medium","sourceLabel":"o3 mini (medium)","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2025-04-10T19:29:02.000Z; reported margin ±1.28 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":10.31,"warning":null},{"id":"cbc2db363833abfb6c9f650d","modelId":"gemini-3.1-pro","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"high","sourceLabel":"gemini-3.1-pro-preview (thinking high)","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2026-04-10T15:51:06.000Z; reported margin ±1.96 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":46.44,"warning":null},{"id":"cd7d373c2c6d5a8b1501e762","modelId":"gpt-5.5","benchmarkId":"scale-sweatlas-tw-codex","level":"xhigh","sourceLabel":"GPT-5.5 (Codex) xHigh","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-05-07T16:02:36.000Z; reported margin ±5.96 percentage points; No contamination annotation reported (not proof of absence)","score":42.59,"warning":null},{"id":"d11e9cc7b5eb05e0e97dbb6d","modelId":"claude-opus-4-8","benchmarkId":"scale-sweatlas-tw-claudecode","level":"xhigh","sourceLabel":"Opus 4.8 (Claude Code) xhigh","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-06-08T20:53:39.000Z; reported margin ±5.93 percentage points; No contamination annotation reported (not proof of absence)","score":49.63,"warning":null},{"id":"da0ddcdb318167ad7f56f77a","modelId":"gemini-3.1-pro","benchmarkId":"scale-enigma_eval-scaleofficialevaluation","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"Scale EnigmaEval puzzle-hunt accuracy; source-reported reasoning effort.; published row 2025-11-06T22:52:23.000Z; reported margin ±2.71 percentage points; No contamination annotation reported (not proof of absence)","score":36.78,"warning":null},{"id":"f621460ea8cfcfd9a6843901","modelId":"gemini-3.1-pro","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"high","sourceLabel":"gemini-3.1-pro-preview (thinking high)","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2026-04-10T15:51:36.000Z; reported margin ±2.11 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":47.31,"warning":null},{"id":"fc3ef3bbe098fd394ff0d2c8","modelId":"gpt-5.6-sol","benchmarkId":"scale-sweatlas-tw-codex","level":"xhigh","sourceLabel":"GPT-5.6-Sol (Codex) xHigh*","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-07-28T19:54:06.000Z; reported margin ±6 percentage points; No contamination annotation reported (not proof of absence)","score":45.9,"warning":"Starred protocol annotation requires row-specific review before ranking."},{"id":"fdcbe9f44fc9ad3996aa935f","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"scale-humanitys_last_exam-scaleofficialevaluation","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929-thinking","configuration":"Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2025-10-02T17:27:45.000Z; reported margin ±1.35 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.","score":13.72,"warning":null},{"id":"ffc0fa711ef42f89ea97e2b2","modelId":"claude-fable-5","benchmarkId":"scale-sweatlas-tw-claudecode","level":"xhigh","sourceLabel":"Fable-5 (Claude Code) xHigh*","configuration":"90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-06-11T17:41:27.000Z; reported margin ±5.8 percentage points; No contamination annotation reported (not proof of absence); Board footnote: elevated refusals on benign tasks are scored zero by the evaluator and remain in the published score.","score":55.6,"warning":null},{"id":"ffd05899fca843c2a297d41a","modelId":"claude-sonnet-5-5","benchmarkId":"scale-humanitys_last_exam_text_only-scaleofficialevaluation","level":"xhigh","sourceLabel":"Sonnet 5.5 (xhigh)","configuration":"Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2026-10-12T23:23:22.000Z; reported margin ±2.1 percentage points; No contamination annotation reported (not proof of absence)","score":46.62,"warning":null},{"id":"panel-ea6e6bc84b19cbbf13457f52","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":99,"warning":null},{"id":"panel-6be887e1f44cda444a386e7b","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":99,"warning":null},{"id":"panel-e9b2bd8993aef948df3383be","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":97,"warning":null},{"id":"panel-4fb37fe7592ba9a5d9869871","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":98,"warning":null},{"id":"panel-c8dfdb832c728882285c9b8a","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":99,"warning":null},{"id":"panel-b3c92aa9680c3988ba1d0614","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":97,"warning":null},{"id":"panel-47f44e2cf6028a348add9648","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":98,"warning":null},{"id":"panel-c2408725a1995f2f76a38c0a","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":98,"warning":null},{"id":"panel-d19da72f8f3dcc7122fa23c7","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":99.01,"warning":null},{"id":"panel-727e4929c5b77b8d68e9616f","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":76,"warning":null},{"id":"panel-cac9ea3d75b90f9f8e22b2f6","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":98,"warning":null},{"id":"panel-3ffaae790b7e3d3e07e6bb93","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":98,"warning":null},{"id":"panel-3b87ecad0c1e9decd51acdca","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":98,"warning":null},{"id":"panel-9ee205c15ad677241207d058","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":98,"warning":null},{"id":"panel-4ef118c3fb039ddcd97965c7","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":98,"warning":null},{"id":"panel-51cb8fa9a9f00859384ed176","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":95,"warning":null},{"id":"panel-f31c0e7a8f4e830bbd7f64c2","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":98,"warning":null},{"id":"panel-9705c757dfdbd2feabac032e","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":98,"warning":null},{"id":"panel-96b3bfd402ebd013b9937af0","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":99,"warning":null},{"id":"panel-fa6c72914cd97c024b6e360a","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":97,"warning":null},{"id":"panel-277c823db47aca935fcfe21d","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":98,"warning":null},{"id":"panel-04f87e2aec3acc7b9cc4f944","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":97,"warning":null},{"id":"panel-79cea26fdffed0236a191bec","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":98,"warning":null},{"id":"panel-8523e2132a25c138914b8459","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":97,"warning":null},{"id":"panel-d24bd2a1ac9eaf9d9494ddaf","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":98,"warning":null},{"id":"panel-0a6ba849134e0075a7f8d9bb","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":98,"warning":null},{"id":"panel-7dbc7ec6565a0f02cf88e3df","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":98,"warning":null},{"id":"panel-3f8750e2188da7c334cd0e4e","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-6-astra-max; checkpoint=gpt-6-astra","score":98,"warning":null},{"id":"panel-162244688f7fe4a357ea05af","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-6-luna-max; checkpoint=gpt-6-luna","score":98,"warning":null},{"id":"panel-939d376b1bd6b6aac363a93d","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-6-sol-max; checkpoint=gpt-6-sol","score":98,"warning":null},{"id":"panel-1650a6167549bdf4f5cb314d","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":98,"warning":null},{"id":"panel-341082a702dc7b549c99482c","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":98,"warning":null},{"id":"panel-633e9a68cee7aa3fb4b67979","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; grok-4.7-xhigh; checkpoint=grok-4.7","score":98,"warning":null},{"id":"panel-c150be94540718c3712937db","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; mistral-large-4-high; checkpoint=mistral-large-4","score":98,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-d1592c494c0bc37b96c4b13e","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":82,"warning":null},{"id":"panel-5db4666fe0d22748d017fabf","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":89,"warning":null},{"id":"panel-e21e4fca1c3e3b98e1bc5a8c","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":99,"warning":null},{"id":"panel-e689a062d192adbe930392e1","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-AMPS_Hard","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; muse-wonder-1.0-max; checkpoint=muse-wonder","score":99,"warning":null},{"id":"panel-662dfda40adfc5a2caa14820","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":80.435,"warning":null},{"id":"panel-641076241a9a5fb731ab01d0","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":82.609,"warning":null},{"id":"panel-e1167db4817e25e6513bfed0","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":78.261,"warning":null},{"id":"panel-76ab901e6799fbf03e76ba13","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":73.913,"warning":null},{"id":"panel-d40fdcb405336ff7b0641bc7","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":80.435,"warning":null},{"id":"panel-c7a6b32798ef20c70836ed4e","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":76.087,"warning":null},{"id":"panel-d622ba2d226d053585bbe3b8","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":78.261,"warning":null},{"id":"panel-335d542464864769d1ca1bdf","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":84.783,"warning":null},{"id":"panel-8dfe6e73a7070dd2ca731ec5","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":82.609,"warning":null},{"id":"panel-5c66ecc29c060d251f32641f","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-code_completion","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":78.261,"warning":null},{"id":"panel-090d54d836ee129f1e5e6609","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":78.261,"warning":null},{"id":"panel-e8160d00199e2caee698cb7e","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":86.957,"warning":null},{"id":"panel-c8d178a76e0240427dcb6fa4","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":84.783,"warning":null},{"id":"panel-d3127514aeabe252e43231e1","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":78.261,"warning":null},{"id":"panel-1a02f6b51825ce2ca2fe0cc6","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":76.087,"warning":null},{"id":"panel-607345e5f467b95cf251d99d","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":76.087,"warning":null},{"id":"panel-37f739dcc05b39207d9d2861","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":78.261,"warning":null},{"id":"panel-c117f17c51149f80018795a6","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":76.087,"warning":null},{"id":"panel-3c33dcbb8642d300cb0bd6fa","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":71.739,"warning":null},{"id":"panel-b6daeef5e5290527aac86f54","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":76.087,"warning":null},{"id":"panel-32aa6b03fa6b1e2b8df86c7b","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":80.435,"warning":null},{"id":"panel-5babfb32a1f20b6e99a59e4e","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":75.556,"warning":null},{"id":"panel-ab5cc700d3ab0b9b4f6bbc85","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":67.391,"warning":null},{"id":"panel-778414f94d6574b36b6dab61","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":82.609,"warning":null},{"id":"panel-f33cd6ce63480f276b68ccdd","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":86.957,"warning":null},{"id":"panel-c08eb2b8012e609064d33ada","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":84.783,"warning":null},{"id":"panel-09a415daaa3ba1bf4182a0f7","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":80.435,"warning":null},{"id":"panel-e91cb820ea7d82b6f4b05678","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-6-astra-max; checkpoint=gpt-6-astra","score":80.435,"warning":null},{"id":"panel-130a9ddad8f163763bec2487","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-6-luna-max; checkpoint=gpt-6-luna","score":80.435,"warning":null},{"id":"panel-c27d70d91ae7bc152ff8560e","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-6-sol-max; checkpoint=gpt-6-sol","score":80.435,"warning":null},{"id":"panel-9c281ac89011e4ff9905b616","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":80.435,"warning":null},{"id":"panel-e06d613ec8098e04af891dfd","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":82.609,"warning":null},{"id":"panel-7abf85f1de127761dfb4a243","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; grok-4.7-xhigh; checkpoint=grok-4.7","score":78.261,"warning":null},{"id":"panel-394dd3bdef92e8dcd0886faf","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-code_completion","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; mistral-large-4-high; checkpoint=mistral-large-4","score":78.261,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-7e01c14d29d1b99014cf62ea","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":78.261,"warning":null},{"id":"panel-31199fc389abf8c37e97a1ef","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":80.435,"warning":null},{"id":"panel-e762d325f86c557f5d10186f","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-code_completion","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":80.435,"warning":null},{"id":"panel-25fa2338b585e4d5e24aaf70","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-code_completion","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_completion; muse-wonder-1.0-max; checkpoint=muse-wonder","score":78.261,"warning":null},{"id":"panel-8a3152b9d28a7530dac28f3d","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":91.549,"warning":null},{"id":"panel-32989ffe22dc5dd32f1b1f83","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":90.141,"warning":null},{"id":"panel-fdfc67b3f048413353a05a7f","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":77.465,"warning":null},{"id":"panel-dfa26defa6fc362dab03f04d","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":78.873,"warning":null},{"id":"panel-e7ab002c6f6098191728383d","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":78.873,"warning":null},{"id":"panel-bfd076ce374bb30a922457d6","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":80.282,"warning":null},{"id":"panel-3eeb212b0c0bb0dc2e478ffa","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":85.915,"warning":null},{"id":"panel-e79c5684c29cd6b303b8c12e","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":78.873,"warning":null},{"id":"panel-6a2c68e12f2d8caab35e1330","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":80.282,"warning":null},{"id":"panel-8e215924ae06b331fbc47123","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-code_generation","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":80.282,"warning":null},{"id":"panel-794d65e981554bcb3f0e92e4","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":83.099,"warning":null},{"id":"panel-eed881eb93fdef4c09afaa59","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":95.775,"warning":null},{"id":"panel-b038ec2a234f6cf8021743af","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":92.958,"warning":null},{"id":"panel-2d4783c770d432f985707cfd","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":74.648,"warning":null},{"id":"panel-4dca8fee9d98c7ddb0581ad6","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":80.282,"warning":null},{"id":"panel-4d93b113c1ffbccf10888778","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":76.056,"warning":null},{"id":"panel-8326dae2469e1a9f74d18cfa","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":77.465,"warning":null},{"id":"panel-7fc55c6792c58d44fdf6c60a","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":81.69,"warning":null},{"id":"panel-00e8a0af372be006ba3d084d","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":73.239,"warning":null},{"id":"panel-da3a0ed56d6a64f954659fa2","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":76.056,"warning":null},{"id":"panel-e93f929cb0853d27c95d04b3","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":74.648,"warning":null},{"id":"panel-054ddbbb05cbeafd60110ea3","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":67.692,"warning":null},{"id":"panel-28232fc4ce54814ba09b4d62","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":74.286,"warning":null},{"id":"panel-52c6e1962b976902f29d09a8","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":81.69,"warning":null},{"id":"panel-d63e27e1819ee0b204cf74ed","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":78.873,"warning":null},{"id":"panel-ada0ae4411fcde2f346dd40e","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":83.099,"warning":null},{"id":"panel-ca7c8b7e35ea3b5ded4662c3","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":76.056,"warning":null},{"id":"panel-353e8b3bee4e37d087e93bed","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-6-astra-max; checkpoint=gpt-6-astra","score":80.282,"warning":null},{"id":"panel-b89337a7e3004fdcade93906","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-6-luna-max; checkpoint=gpt-6-luna","score":77.465,"warning":null},{"id":"panel-992cc8609e0891af67ab03f2","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-6-sol-max; checkpoint=gpt-6-sol","score":83.099,"warning":null},{"id":"panel-3c8f097ca9e3961912b24ad1","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":80.282,"warning":null},{"id":"panel-cad3a85bb94203d52bfb643e","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":78.873,"warning":null},{"id":"panel-eb1630d6571f33ec35b4857c","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; grok-4.7-xhigh; checkpoint=grok-4.7","score":76.056,"warning":null},{"id":"panel-9d17bc866a773505d636d87b","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-code_generation","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; mistral-large-4-high; checkpoint=mistral-large-4","score":76.056,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-6baf6c794bf4aaf644c7b7c3","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":76.056,"warning":null},{"id":"panel-90609ba61b59695b58129751","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":74.648,"warning":null},{"id":"panel-4d93f01d1e8e2402aaeb5616","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-code_generation","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":81.69,"warning":null},{"id":"panel-ebadb48e181f168f04a307b1","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-code_generation","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-code_generation; muse-wonder-1.0-max; checkpoint=muse-wonder","score":83.099,"warning":null},{"id":"panel-cfb0683f76806bf4fb0265d0","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":99.333,"warning":null},{"id":"panel-ad218815ebcfbb1c50b5859c","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":99.333,"warning":null},{"id":"panel-b53a255d5de9fcfe2c3e7b11","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":96,"warning":null},{"id":"panel-8b1d0526c4e8240505f76ea3","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":93.833,"warning":null},{"id":"panel-d3216c623cce0c30a3549af3","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":99.333,"warning":null},{"id":"panel-02715a3959bcec12a2118d77","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":99.333,"warning":null},{"id":"panel-fd74c890d8cbdc78ca4926f7","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":92.333,"warning":null},{"id":"panel-de5882ae22989b74ec6dc0e7","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":99.333,"warning":null},{"id":"panel-da44c7f3cf2a9599579e7cea","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":99.333,"warning":null},{"id":"panel-30450bbe2cd94e48badcae38","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-connections","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":99.333,"warning":null},{"id":"panel-4f59778a92b6497855298360","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":95.5,"warning":null},{"id":"panel-f21749cd94a53dcbe2801102","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":99.333,"warning":null},{"id":"panel-f43d0549e279bd88fe7b0c13","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":98.5,"warning":null},{"id":"panel-6fb4bc1b60f839623c5d4381","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":100,"warning":null},{"id":"panel-828ad3db06596510a3a5b361","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":100,"warning":null},{"id":"panel-85e5f51bb280d552f8a1344b","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":92,"warning":null},{"id":"panel-99585dbf3dfb57868d6e94e8","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":100,"warning":null},{"id":"panel-f4dc253f7a805058432fb239","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":100,"warning":null},{"id":"panel-4d73f7faa8106fea92f3fb4d","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":100,"warning":null},{"id":"panel-77f63d80977191cf2c481d68","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":99,"warning":null},{"id":"panel-424fc5747345bdb81336e06a","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":100,"warning":null},{"id":"panel-7ad1f49952e339692c304819","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":98.98,"warning":null},{"id":"panel-342d090b1be63195b7fa706e","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":94.388,"warning":null},{"id":"panel-f8a70bbbcd0f698be7bc1a43","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":100,"warning":null},{"id":"panel-f55d9d7148753278f84c2df4","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":96.5,"warning":null},{"id":"panel-014e5b85ba751dcc8149108e","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":100,"warning":null},{"id":"panel-de29cc4390253be5d2777e81","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":93,"warning":null},{"id":"panel-c1b3d7f5a165e5b35bf2018b","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-6-astra-max; checkpoint=gpt-6-astra","score":100,"warning":null},{"id":"panel-d2a5a8202d5f6238cc578ece","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-6-luna-max; checkpoint=gpt-6-luna","score":100,"warning":null},{"id":"panel-650b58aae7b8903d928b42f6","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-6-sol-max; checkpoint=gpt-6-sol","score":100,"warning":null},{"id":"panel-06f1e4ca42c0eb2409ff341e","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":100,"warning":null},{"id":"panel-4fa5248a7daa6a060a9809cb","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":100,"warning":null},{"id":"panel-44a5c4225ed2f6e960b849ed","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; grok-4.7-xhigh; checkpoint=grok-4.7","score":98,"warning":null},{"id":"panel-8723236a70407e8aab03a7e1","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-connections","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; mistral-large-4-high; checkpoint=mistral-large-4","score":48,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-1cba090af30c55faf508a232","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":98,"warning":null},{"id":"panel-bbd53ca4e345a1af051f9838","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":100,"warning":null},{"id":"panel-d86a15e1a604826048816160","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-connections","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":94,"warning":null},{"id":"panel-74766c587c80763ddfc4e0ea","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-connections","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-connections; muse-wonder-1.0-max; checkpoint=muse-wonder","score":96,"warning":null},{"id":"panel-190855818b7e1222a48e2eae","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":91.36,"warning":null},{"id":"panel-fd99f744a0c951b883b6eaea","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":90.555,"warning":null},{"id":"panel-d98b9807899d7d05e70fa109","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":10.638,"warning":null},{"id":"panel-5469acf9d5d94024f3406a17","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":71.047,"warning":null},{"id":"panel-9fbf2f2774c52b4c65e96782","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":79.363,"warning":null},{"id":"panel-39728f4c4f0069938bbfdf20","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":63.467,"warning":null},{"id":"panel-1b8be849869ca37987fe3c9f","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":89.521,"warning":null},{"id":"panel-dd863a0d16624af984380aa0","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":52.905,"warning":null},{"id":"panel-2aa437a8829dbf56b4b5d351","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":77.571,"warning":null},{"id":"panel-9a9cff8685184797af933836","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":92.683,"warning":null},{"id":"panel-dc52f3d94a18a75b842b27ef","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":74.741,"warning":null},{"id":"panel-5f85117b9a7433d053758ad0","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":40.616,"warning":null},{"id":"panel-e8b537d191eadd3fd111ef9d","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":90.936,"warning":null},{"id":"panel-b70f7219a2898aed69af5a1b","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":85.239,"warning":null},{"id":"panel-c0f6c7ecaaece53d8096cb1a","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":47.994,"warning":null},{"id":"panel-27b4ef02c7a063ec88c3ddc7","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":21.095,"warning":null},{"id":"panel-56f7440caf2be4e507c56b62","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":43.824,"warning":null},{"id":"panel-c7924ab86a4819c0c3bc7f8f","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":60.373,"warning":null},{"id":"panel-147c821e2fbfd731f109691d","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":20.612,"warning":null},{"id":"panel-a2b9872bb59a3f7ac3aff0c0","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":88.913,"warning":null},{"id":"panel-686153cb0b44a71199a0619d","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":86.19,"warning":null},{"id":"panel-8afc38f69537e97bbf2c5a67","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":62.972,"warning":null},{"id":"panel-7b24c6ef6a4016a1ce0ec99a","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":54.301,"warning":null},{"id":"panel-8f65ef6ded7e915b5df1386a","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":88.823,"warning":null},{"id":"panel-ebc66a99e1b9d323c3fa4812","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":86.504,"warning":null},{"id":"panel-1f6eb2d767a08535a94cb732","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":90.214,"warning":null},{"id":"panel-cf82c90989f66444fdc5693d","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":87.67,"warning":null},{"id":"panel-ec9906abde481a797aa4cc9a","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-6-astra-max; checkpoint=gpt-6-astra","score":90.555,"warning":null},{"id":"panel-d3589d5cd927ea089932c917","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-6-luna-max; checkpoint=gpt-6-luna","score":70.494,"warning":null},{"id":"panel-c774e7bdb7e27ce19f429492","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-6-sol-max; checkpoint=gpt-6-sol","score":89.549,"warning":null},{"id":"panel-105cc43b0c2557e180d20087","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":90.555,"warning":null},{"id":"panel-0718fdbd75f567adc417839f","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":91.289,"warning":null},{"id":"panel-277ec4301d4c151d341badad","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; grok-4.7-xhigh; checkpoint=grok-4.7","score":85.727,"warning":null},{"id":"panel-495c60b884a90bb39ef8b8f5","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; mistral-large-4-high; checkpoint=mistral-large-4","score":80.529,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-b0e641583182f918dc7c7607","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":71.086,"warning":null},{"id":"panel-dbe6aa071a22ccaa40e097c9","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":82.548,"warning":null},{"id":"panel-9b4dc6b2d88a35de49975d62","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":90.504,"warning":null},{"id":"panel-c69e0aad188d1ebf54683cc3","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-consecutive_events","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-consecutive_events; muse-wonder-1.0-max; checkpoint=muse-wonder","score":98.525,"warning":null},{"id":"panel-1330ccf27b12cf2611c21dbe","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":97,"warning":null},{"id":"panel-65c52ed61e13ff7b3b9fb1e6","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":99,"warning":null},{"id":"panel-3a82e3756cd78153ccb995d8","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":67,"warning":null},{"id":"panel-53cc7a2ac6ed3984ee7771ed","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":89,"warning":null},{"id":"panel-1ee6676d7e81096dff4bb6b9","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":78,"warning":null},{"id":"panel-fb548e320f80168b36d88c76","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":73,"warning":null},{"id":"panel-cf976805e5a88ef22e6c1aa3","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":84,"warning":null},{"id":"panel-5610f87659e4a35f39fc8224","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":89,"warning":null},{"id":"panel-f717d2864a445467e54f0414","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":97,"warning":null},{"id":"panel-e9fc47dbe181087a7a1a77db","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":90,"warning":null},{"id":"panel-130f53a4fe944096e9fb6b61","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":88,"warning":null},{"id":"panel-96fd55540df8e4a9431ff751","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":97,"warning":null},{"id":"panel-de353052d22f8f8da48668e6","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":100,"warning":null},{"id":"panel-4f12ebc87581a8b3fead6ca5","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":78,"warning":null},{"id":"panel-6ffc749529292fd222d18cba","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":68,"warning":null},{"id":"panel-f573c5db18f106fe09304de4","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":26,"warning":null},{"id":"panel-45e2f893822c593c4ecd6e28","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":63,"warning":null},{"id":"panel-78e254545a2ae1a666fe3269","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":88,"warning":null},{"id":"panel-20246bf3e3306bd736a3c80c","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":80,"warning":null},{"id":"panel-2c75d7123d8291e5be81a00b","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":92,"warning":null},{"id":"panel-ead1203dc409c40234d5fee4","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":93,"warning":null},{"id":"panel-4305d4d3278517373e3f5b3e","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":37,"warning":null},{"id":"panel-7934928ca32eaec604963700","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":82.828,"warning":null},{"id":"panel-87b260b8aea4065b7af8575f","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":99,"warning":null},{"id":"panel-bb3ced238358ea1c61780784","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":70,"warning":null},{"id":"panel-e42b32e0c02a97ebf85ee088","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":100,"warning":null},{"id":"panel-a9f487811c15538616e240dc","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":95,"warning":null},{"id":"panel-1a8d0032f7d3827254000690","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-6-astra-max; checkpoint=gpt-6-astra","score":100,"warning":null},{"id":"panel-3909bf096095708a59780a61","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-6-luna-max; checkpoint=gpt-6-luna","score":74,"warning":null},{"id":"panel-9640dd63ff82a5cef1d24a23","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-6-sol-max; checkpoint=gpt-6-sol","score":99,"warning":null},{"id":"panel-cabfe320c3555420565cb068","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":100,"warning":null},{"id":"panel-9534f83822d33e0bed8ebd90","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":99,"warning":null},{"id":"panel-d6dfa32303cdf3f250dd3ec2","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; grok-4.7-xhigh; checkpoint=grok-4.7","score":96,"warning":null},{"id":"panel-881e35e28677c446d8b38378","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; mistral-large-4-high; checkpoint=mistral-large-4","score":90,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-ff7deaad8514a735c60215cd","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":83,"warning":null},{"id":"panel-939b2284f37f61d8eb5f67b6","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":91,"warning":null},{"id":"panel-ea3620ceae6503098460020c","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":96,"warning":null},{"id":"panel-c67eba384acc4ed30aef4730","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-integrals_with_game","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-integrals_with_game; muse-wonder-1.0-max; checkpoint=muse-wonder","score":98,"warning":null},{"id":"panel-9e636b17bbcae275ecd8ee17","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":68.182,"warning":null},{"id":"panel-dfcdc9973f2a0d00f0fc919c","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":68.182,"warning":null},{"id":"panel-10926a682b532e80cfc809da","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":68.182,"warning":null},{"id":"panel-9e877a28272b2b5cdd278ea6","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":59.091,"warning":null},{"id":"panel-e7184a5ecfae339d037a2b4e","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":59.091,"warning":null},{"id":"panel-6414676946387e29fa05c0a8","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":63.636,"warning":null},{"id":"panel-33d654882b80b1889204d2e4","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":63.636,"warning":null},{"id":"panel-af8e1e7afb22e13a8161b804","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":68.182,"warning":null},{"id":"panel-1915dc0e10000252e26496f2","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":77.273,"warning":null},{"id":"panel-239cfd267c6875696e83c9cb","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-javascript","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":54.545,"warning":null},{"id":"panel-e6a2705861cdce46400bf7cb","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":68.182,"warning":null},{"id":"panel-2699ab6f506006b4a0a59751","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":77.273,"warning":null},{"id":"panel-f89d19344115b6d796a79459","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":36.364,"warning":null},{"id":"panel-58aac2c2dde4fbb0c2408d75","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":59.091,"warning":null},{"id":"panel-f9834bac2c8a90d6c7ce4e5e","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":63.636,"warning":null},{"id":"panel-5c58f2a3098155c426bac27d","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":59.091,"warning":null},{"id":"panel-f2c256a295bd7932f236f885","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":63.636,"warning":null},{"id":"panel-36628f2999915365afee11c0","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":68.182,"warning":null},{"id":"panel-eaf227f5b755476d39c81ae7","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":72.727,"warning":null},{"id":"panel-49c4b3d4d02e77f4afa7ec01","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":59.091,"warning":null},{"id":"panel-53e2e2b8cbdb216a093034bf","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":68.182,"warning":null},{"id":"panel-422822cff2d0c951a07e107d","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":50,"warning":null},{"id":"panel-ed27fdf022a823e0abcb4c54","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":63.636,"warning":null},{"id":"panel-1de01be8339431f506375fa6","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":63.636,"warning":null},{"id":"panel-0468cbf93870ad38beab1059","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":63.636,"warning":null},{"id":"panel-e366b9965c11ca7171a26c5e","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":63.636,"warning":null},{"id":"panel-b227ac7dafc453027f546347","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":68.182,"warning":null},{"id":"panel-d0d17e92e62bfc0685fa0240","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-6-astra-max; checkpoint=gpt-6-astra","score":63.636,"warning":null},{"id":"panel-4c1f14f11ae12bd98a05bdf0","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-6-luna-max; checkpoint=gpt-6-luna","score":63.636,"warning":null},{"id":"panel-c6743dc3d8cf2ac56248541e","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-6-sol-max; checkpoint=gpt-6-sol","score":63.636,"warning":null},{"id":"panel-206dfa02fc2828678b6d2faa","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":63.636,"warning":null},{"id":"panel-98bda5caa8ea0f4e5a0a90eb","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":63.636,"warning":null},{"id":"panel-47a51d6b96d0cd6547d125e9","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; grok-4.7-xhigh; checkpoint=grok-4.7","score":63.636,"warning":null},{"id":"panel-ff03c9088659ae53d42267dd","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-javascript","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; mistral-large-4-high; checkpoint=mistral-large-4","score":68.182,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-6bdc2206bd256d34c7fd7650","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":77.273,"warning":null},{"id":"panel-c37413eeb9c7e46ddfc253d2","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":72.727,"warning":null},{"id":"panel-e2b90ad9ff025816b82efbba","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-javascript","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":77.273,"warning":null},{"id":"panel-54e4e1da32d5723495e597ea","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-javascript","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-javascript; muse-wonder-1.0-max; checkpoint=muse-wonder","score":68.182,"warning":null},{"id":"panel-fb6b440bc03b13ca1c9f354c","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":78,"warning":null},{"id":"panel-8f0343e8b72f79cdda804ff7","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":86,"warning":null},{"id":"panel-8411ac86a2cb78bc80d4bda9","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":66,"warning":null},{"id":"panel-40395aaf36b8d1b1d9231de1","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":56,"warning":null},{"id":"panel-5391f4e5c06b72826fee1ad5","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":68,"warning":null},{"id":"panel-f1f80cee7d1e762d01154e92","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":80,"warning":null},{"id":"panel-4172bd2caec615ab3eff3b0c","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":70,"warning":null},{"id":"panel-c3668fe9c1af9e1f815034dc","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":78,"warning":null},{"id":"panel-8b775d5d773456b6ab271308","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":86,"warning":null},{"id":"panel-d4d1042248b561833a92f2c4","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":70,"warning":null},{"id":"panel-e8a89b55eb0543d94e2cc816","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":86,"warning":null},{"id":"panel-09c8816501726571273693f9","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":82,"warning":null},{"id":"panel-9ffecef92abd29eaad4796ce","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":80,"warning":null},{"id":"panel-4d524cc2ad780da119be2cf9","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":72,"warning":null},{"id":"panel-79730153cbde8d293d44213a","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":74,"warning":null},{"id":"panel-873971713d4ce0742cfd0e18","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":68,"warning":null},{"id":"panel-19da7fc7d58c0bfc47ec9a8c","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":76,"warning":null},{"id":"panel-e23f33b25b7acb309969ee70","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":74,"warning":null},{"id":"panel-4000faefeda3951617e07ec9","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":82,"warning":null},{"id":"panel-50a0a1f23eeb2eceab0f4c4a","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":74,"warning":null},{"id":"panel-ce06ee813be216293e7004cf","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":68,"warning":null},{"id":"panel-603dd81fba10b50320567b1e","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":60,"warning":null},{"id":"panel-76844476c511fb7c5531e3e8","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":64,"warning":null},{"id":"panel-ad21b2b9b3a285f669bede29","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":76,"warning":null},{"id":"panel-c2188e8931d7269e78bdf635","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":80,"warning":null},{"id":"panel-51a4f42f816be6130de2f015","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":82,"warning":null},{"id":"panel-f9e413513f7da7d9e60b6964","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":76,"warning":null},{"id":"panel-60bacf6417f2d976a9991114","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-6-astra-max; checkpoint=gpt-6-astra","score":88,"warning":null},{"id":"panel-a54e6a5bb487cd66b30504f9","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-6-luna-max; checkpoint=gpt-6-luna","score":62,"warning":null},{"id":"panel-6a1c2759c6c12ce64fea9574","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-6-sol-max; checkpoint=gpt-6-sol","score":74,"warning":null},{"id":"panel-98c9ec191b9d3d1c99b07882","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":86,"warning":null},{"id":"panel-1aa9d89adde00d6a19211df8","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":82,"warning":null},{"id":"panel-ed2c7ad29478e2c8b59b4217","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; grok-4.7-xhigh; checkpoint=grok-4.7","score":66,"warning":null},{"id":"panel-0dc5b51301907163a7deaa45","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; mistral-large-4-high; checkpoint=mistral-large-4","score":68,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-eb8e759c8756c57a6678b38c","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":78,"warning":null},{"id":"panel-4a402a6edf729fb609faca1b","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":88,"warning":null},{"id":"panel-a608739b95a5826c41cbb0bc","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":76,"warning":null},{"id":"panel-7ce2e0d0b83d48ac917a163a","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-logic_with_navigation","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-logic_with_navigation; muse-wonder-1.0-max; checkpoint=muse-wonder","score":80,"warning":null},{"id":"panel-bdae59106882e6ce30423560","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":95.098,"warning":null},{"id":"panel-67e783f1c412af0bc138d97d","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":97.059,"warning":null},{"id":"panel-27ec8000077cd9fe4b496b08","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":98.039,"warning":null},{"id":"panel-81f5007ab231db05774cfcbf","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":97.059,"warning":null},{"id":"panel-f4611dea0f98ae60341b3497","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":95.098,"warning":null},{"id":"panel-9af7f61af6179d5eac0c51be","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":95.098,"warning":null},{"id":"panel-9266dd9fd15700366b37c0ed","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":98.039,"warning":null},{"id":"panel-3dd977ca39f5c684439cf98e","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":98.039,"warning":null},{"id":"panel-919a5557dc2a750353058ef4","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":94.118,"warning":null},{"id":"panel-61b37a37ffa13a56ab533193","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-math_comp","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":94.118,"warning":null},{"id":"panel-a2e219e671bd391642993295","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":95.098,"warning":null},{"id":"panel-b674a6016d5e1443779b0676","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":97.059,"warning":null},{"id":"panel-df84ed0b0745f8d56df7f0c4","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":97.059,"warning":null},{"id":"panel-dd35ef007b87680fc5d395d0","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":96.078,"warning":null},{"id":"panel-65f8d6e0729b9c5c42dba5c9","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":95.098,"warning":null},{"id":"panel-d44be38166189727bb5717a2","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":92.157,"warning":null},{"id":"panel-ef97c9d27235b86d73fda913","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":94.118,"warning":null},{"id":"panel-6fcbe883dd2cc95616b54d6f","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":96.078,"warning":null},{"id":"panel-51fe9f895027fea4c0ae2759","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":95.098,"warning":null},{"id":"panel-fbcbde600561c3d3696be83c","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":95.098,"warning":null},{"id":"panel-14b8986f6045509b1190f001","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":94.118,"warning":null},{"id":"panel-1eb69e1e8bf07bc6c2dd0181","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":94.118,"warning":null},{"id":"panel-21ace49cde007a4ad2f4e1f5","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":95.098,"warning":null},{"id":"panel-f845a78cc2cb4939b0d457d3","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":96.078,"warning":null},{"id":"panel-336ccdda131c10a4ceb1c659","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":92.157,"warning":null},{"id":"panel-b2d8d393da701a3f89530fc2","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":95.098,"warning":null},{"id":"panel-a3408fd345658854b48ac897","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":95.098,"warning":null},{"id":"panel-49cb0eef818478e18b9fc73e","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-6-astra-max; checkpoint=gpt-6-astra","score":97.059,"warning":null},{"id":"panel-9a1afabb04048d30c0e13610","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-6-luna-max; checkpoint=gpt-6-luna","score":94.118,"warning":null},{"id":"panel-af919c82b4251252f3687e2e","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-6-sol-max; checkpoint=gpt-6-sol","score":97.059,"warning":null},{"id":"panel-868d8090ed72e0e3c6a8932c","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":97.059,"warning":null},{"id":"panel-cc5db73d28404ad81a50c02b","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":97.059,"warning":null},{"id":"panel-ca92bdf8905c066c45d0aae8","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; grok-4.7-xhigh; checkpoint=grok-4.7","score":97.059,"warning":null},{"id":"panel-3025e1d8b41e854f23a4f5af","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-math_comp","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; mistral-large-4-high; checkpoint=mistral-large-4","score":97.059,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-9a71dad62d416416d1786c72","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":96.078,"warning":null},{"id":"panel-b4abe05d293dd2f434bf065b","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":97.059,"warning":null},{"id":"panel-eb4ce80a488879c6c664dde7","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-math_comp","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":97.059,"warning":null},{"id":"panel-fbf6974aa029bb3328653ff9","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-math_comp","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-math_comp; muse-wonder-1.0-max; checkpoint=muse-wonder","score":97.059,"warning":null},{"id":"panel-43a711c2e4bf1032ca789322","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":92.845,"warning":null},{"id":"panel-6c18356f3e605ccf870bc0e4","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":92.968,"warning":null},{"id":"panel-745e660722dbdecf76e8a91a","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":83.701,"warning":null},{"id":"panel-de0d0749dcfb4777722fe6dd","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":86.951,"warning":null},{"id":"panel-09c708f2367d5cdaf9f13aaf","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":89.458,"warning":null},{"id":"panel-0d7a1d9d2a51aa4564298f91","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":92.17,"warning":null},{"id":"panel-9569698400df06830af314ab","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":91.376,"warning":null},{"id":"panel-141f8c45edac9dc4154ac08a","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":92.227,"warning":null},{"id":"panel-35992b4dd228235b40982727","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":92.796,"warning":null},{"id":"panel-7d9064929591d93f9978bec5","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-olympiad","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":87.858,"warning":null},{"id":"panel-6ade069d452324b81f97d2d0","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":90.671,"warning":null},{"id":"panel-63b9e159cd4a333dab85ab7e","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":92.468,"warning":null},{"id":"panel-d5f0b4a0f674e5fc2bfbad55","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":91.72,"warning":null},{"id":"panel-8272dd6e6a4ec9a8e6cbdc9f","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":92.102,"warning":null},{"id":"panel-43e3fa367afc5689d6e1a375","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":91.877,"warning":null},{"id":"panel-c72b7981866ef805a49330cf","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":81.801,"warning":null},{"id":"panel-c399a15bbae382124505b488","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":90.492,"warning":null},{"id":"panel-b0cd868f837f3c8d3ba378c0","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":91.794,"warning":null},{"id":"panel-0ebcff26b3dd1ee910c0e7d8","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":92.159,"warning":null},{"id":"panel-e236e8906de2ac79a50c15e6","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":88.566,"warning":null},{"id":"panel-387ee41f45c34973406f6a99","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":91.474,"warning":null},{"id":"panel-1a9f74dc3ba9d5be73f6de46","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":85.73,"warning":null},{"id":"panel-261505d025f9d0d994bd7c1c","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":87.982,"warning":null},{"id":"panel-f7afb6dfa5602fcbcc94b9b9","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":91.351,"warning":null},{"id":"panel-f35f8aae0d8d575b94fd76cf","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":88.646,"warning":null},{"id":"panel-5f7b8f51d17c740d7ce23328","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":91.693,"warning":null},{"id":"panel-2ffe687d54dc27361f384c80","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":91.535,"warning":null},{"id":"panel-c033816ccaa5e32d8fc2fa0d","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-6-astra-max; checkpoint=gpt-6-astra","score":92.166,"warning":null},{"id":"panel-1cc510640bb52a05cf58bcdd","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-6-luna-max; checkpoint=gpt-6-luna","score":90.376,"warning":null},{"id":"panel-e71f2c3822cf33588c373ceb","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-6-sol-max; checkpoint=gpt-6-sol","score":91.365,"warning":null},{"id":"panel-faaffb73fad007de19cd846d","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":92.268,"warning":null},{"id":"panel-a505d83051f6bd95f151f80d","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":91.855,"warning":null},{"id":"panel-7c74c5d6ab8aa9e5bf2c203d","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; grok-4.7-xhigh; checkpoint=grok-4.7","score":91.653,"warning":null},{"id":"panel-5342d5279a0ed1a3e891243e","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-olympiad","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; mistral-large-4-high; checkpoint=mistral-large-4","score":89.28,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-5f2ebc6018d88a7989407282","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":87.501,"warning":null},{"id":"panel-7d6cb0c0f64f505b9a71b6ff","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":87.755,"warning":null},{"id":"panel-ae4a0aa81365bb0ac625fdee","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-olympiad","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":91.741,"warning":null},{"id":"panel-d89fbdd1388539c6a6c35abd","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-olympiad","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-olympiad; muse-wonder-1.0-max; checkpoint=muse-wonder","score":91.823,"warning":null},{"id":"panel-2be4cdc484550578181f30a1","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":77.65,"warning":null},{"id":"panel-7b21ddb3a6195cfe20acdf0f","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":71.517,"warning":null},{"id":"panel-e6cb4829222c03f428608ea2","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":50.217,"warning":null},{"id":"panel-f422f1b40639d5881abb1dea","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":62.5,"warning":null},{"id":"panel-b7c7abb64ed26c981fa97ecb","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":65.667,"warning":null},{"id":"panel-53612080416a8a2c52255475","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":62.817,"warning":null},{"id":"panel-369a3b2d9f40c49f3783f859","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":65.767,"warning":null},{"id":"panel-d1b3a97faa4ffcecfd7ad8f3","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":67.867,"warning":null},{"id":"panel-1099925f01eec7849c621575","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":65.733,"warning":null},{"id":"panel-c1f148c80800c7bb48207f5e","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-paraphrase","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":59.4,"warning":null},{"id":"panel-47835dba414723cbd89102e1","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":61.25,"warning":null},{"id":"panel-84f8ea1d78c778ab372af271","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":48.217,"warning":null},{"id":"panel-79dc1e8d9d1fe27308af31e5","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":62.283,"warning":null},{"id":"panel-6730b76b0480a9ddba9ceb60","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":79.867,"warning":null},{"id":"panel-54d0390b13496c156fa846ab","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":75.517,"warning":null},{"id":"panel-e759c56964927bda6b798d1b","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":65,"warning":null},{"id":"panel-217316b453a05c8a737a5311","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":74.317,"warning":null},{"id":"panel-0973b8fdc8a954215ca48936","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":78.067,"warning":null},{"id":"panel-8e7c5485b183951578bf6f63","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":80,"warning":null},{"id":"panel-c0d042721929915c8645b57d","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":64.1,"warning":null},{"id":"panel-6212f1adcca2531c22bd97dc","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":63.433,"warning":null},{"id":"panel-20cb55fec83d1270c8d422e2","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":58.883,"warning":null},{"id":"panel-6ffae1ead5b7afc06bdb440c","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":66.649,"warning":null},{"id":"panel-67db00b7f19613b66461ba0f","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":71.733,"warning":null},{"id":"panel-f4035a99f61947c9e2d3a096","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":47.817,"warning":null},{"id":"panel-aee7349a79f0e6aea04d06f6","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":71.517,"warning":null},{"id":"panel-b9a05d343e2978f3dcffdc61","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":57.033,"warning":null},{"id":"panel-963e7b44cd565be1b3cc9695","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-6-astra-max; checkpoint=gpt-6-astra","score":78.25,"warning":null},{"id":"panel-a9246dca57ff39f1de17943d","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-6-luna-max; checkpoint=gpt-6-luna","score":49.7,"warning":null},{"id":"panel-650ca7621d72b9b4104ca4d2","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-6-sol-max; checkpoint=gpt-6-sol","score":70.083,"warning":null},{"id":"panel-98a29b927add46b750fbd539","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":72.583,"warning":null},{"id":"panel-60acbd5637c42225d81622fb","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":71.033,"warning":null},{"id":"panel-ee701308d97e330a8b4249fa","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; grok-4.7-xhigh; checkpoint=grok-4.7","score":78.05,"warning":null},{"id":"panel-8ed268ed656d254b108ce78f","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-paraphrase","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; mistral-large-4-high; checkpoint=mistral-large-4","score":63.967,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-64d9137045cdef710bd3ca92","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":69.4,"warning":null},{"id":"panel-d10f6d2cdabb5ddb7e4899d2","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":74.3,"warning":null},{"id":"panel-d4235d275b78c1d09ff5b633","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-paraphrase","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":78.617,"warning":null},{"id":"panel-ca12f749d0bae5cb876bb122","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-paraphrase","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-paraphrase; muse-wonder-1.0-max; checkpoint=muse-wonder","score":74.067,"warning":null},{"id":"panel-03a1e315c4f027dcfd8299b7","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":78.719,"warning":null},{"id":"panel-d60952da8f212b02c5b80e5a","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":75.17,"warning":null},{"id":"panel-3c1ca099d1b432a26355e864","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":27.884,"warning":null},{"id":"panel-f97c4ecafec217f90271e795","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":42.715,"warning":null},{"id":"panel-bd2cb017dd4aabc76d11d0f5","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":66.45,"warning":null},{"id":"panel-5db103e8c0d0940f23c1c437","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":66.476,"warning":null},{"id":"panel-9884a6b97b5686266ebc97bd","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":61.409,"warning":null},{"id":"panel-4ed8370e5706671ee5d2dd82","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":61.645,"warning":null},{"id":"panel-bf356ff8ba3d33e34d5d7695","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":74.731,"warning":null},{"id":"panel-57c9d66e18c45c4f050a462a","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":56.975,"warning":null},{"id":"panel-a7b7df43b105c928e66b1c2f","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":53.411,"warning":null},{"id":"panel-e31cb53231e2b27cca93912b","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":48.619,"warning":null},{"id":"panel-218411603a1f37301a8e1439","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":67.789,"warning":null},{"id":"panel-0662502669451fa91c907b36","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":74.128,"warning":null},{"id":"panel-e5935fb8a328a72c57a5ecc4","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":67.753,"warning":null},{"id":"panel-7393786a27fc26872419dcaa","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":49.461,"warning":null},{"id":"panel-f128b281d5845ab277063785","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":63.693,"warning":null},{"id":"panel-b18397a1aab14698db6506d5","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":66.365,"warning":null},{"id":"panel-21d2715c30bd186f37bdd468","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":73.378,"warning":null},{"id":"panel-91d118503479005ce69deb47","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":58.427,"warning":null},{"id":"panel-6fda03e93661f6c58ea5e7e4","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":65.901,"warning":null},{"id":"panel-776124249647ccf9c51344bb","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":43.875,"warning":null},{"id":"panel-62a295778554ce1fab5fb5a5","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":39.133,"warning":null},{"id":"panel-5990349629b3aae8b03c2734","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":74.089,"warning":null},{"id":"panel-e05a7d89da690e9010835fcf","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":51.201,"warning":null},{"id":"panel-76fb0a8d12e96d2ae0671a45","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":79.051,"warning":null},{"id":"panel-32a3607710de8dca34022897","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":69.678,"warning":null},{"id":"panel-299ef4c22b4342670d0dae08","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-6-astra-max; checkpoint=gpt-6-astra","score":86.292,"warning":null},{"id":"panel-eb0ce253202e2da98ce21572","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-6-luna-max; checkpoint=gpt-6-luna","score":57.483,"warning":null},{"id":"panel-30e4eb8001c89f21516b2f27","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-6-sol-max; checkpoint=gpt-6-sol","score":77.896,"warning":null},{"id":"panel-0c904491b31b75d49fe877cd","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":84.378,"warning":null},{"id":"panel-b3e8d273a7fc98d8e5ffb54f","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":81.935,"warning":null},{"id":"panel-a091c9ab2a910352398c4d11","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; grok-4.7-xhigh; checkpoint=grok-4.7","score":72.416,"warning":null},{"id":"panel-853412c6f0ef9c6d08529d83","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; mistral-large-4-high; checkpoint=mistral-large-4","score":40.927,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-3cfdcdd691354a0a5029cdd2","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":63.026,"warning":null},{"id":"panel-b100b1155d4fad7945551d5b","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":61.724,"warning":null},{"id":"panel-01d19abcae8228176928ff95","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":68.381,"warning":null},{"id":"panel-18a7565596dfecd68f676d6a","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-plot_unscrambling","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-plot_unscrambling; muse-wonder-1.0-max; checkpoint=muse-wonder","score":74.797,"warning":null},{"id":"panel-a2de81d0984e710daa10737b","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":65,"warning":null},{"id":"panel-03d6d7332196ab2b7940014a","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":70,"warning":null},{"id":"panel-f85b737595936346a3c0a262","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":60,"warning":null},{"id":"panel-8412193eccfe41b3c21b69c8","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":55,"warning":null},{"id":"panel-344cfa519ab6003234bf074c","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":40,"warning":null},{"id":"panel-b8667c26ba0685da759190a2","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":50,"warning":null},{"id":"panel-662364bb53927d6d913898d5","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":45,"warning":null},{"id":"panel-75a12802fa02fbeea400c4cb","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":50,"warning":null},{"id":"panel-679a950ffb35122c670ad6ba","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":75,"warning":null},{"id":"panel-c9dec434280ecd1032922f0a","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-python","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":50,"warning":null},{"id":"panel-667bd298eef009be052a0928","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":60,"warning":null},{"id":"panel-806f7f43c0689fa946699788","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":35,"warning":null},{"id":"panel-7b1e4167fe37b5060d24255b","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":35,"warning":null},{"id":"panel-728ecc9a397966fe315b4b0a","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":50,"warning":null},{"id":"panel-f0a38c4adeea44226aae53f0","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":50,"warning":null},{"id":"panel-0e6208d14f6ce63317052579","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":50,"warning":null},{"id":"panel-8d9e48f418b10c02e9fa3a61","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":40,"warning":null},{"id":"panel-eaef80a81b2207fafc6cf592","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":60,"warning":null},{"id":"panel-aef8a69ac35010bb70d01430","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":50,"warning":null},{"id":"panel-09e4a65f3e19547b97c9aaf4","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":55,"warning":null},{"id":"panel-cd52404d183578937e22deac","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":50,"warning":null},{"id":"panel-e6821f14580e75e469839a58","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":55,"warning":null},{"id":"panel-26dd5319f2452d0d3e5920b6","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":50,"warning":null},{"id":"panel-843ad8b9e75ccb2e2e0290aa","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":55,"warning":null},{"id":"panel-3b0a9ae45b36b943f938ebee","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":45,"warning":null},{"id":"panel-c932d96633f436c5ab84fc2a","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":55,"warning":null},{"id":"panel-e1f754b895ac409b7ff2d1f9","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":50,"warning":null},{"id":"panel-d5ad12f947dad7a22a4caa26","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-6-astra-max; checkpoint=gpt-6-astra","score":65,"warning":null},{"id":"panel-7d074912e4b3edceef7ca746","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-6-luna-max; checkpoint=gpt-6-luna","score":50,"warning":null},{"id":"panel-cd26b25408142e12f6dfdbca","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-6-sol-max; checkpoint=gpt-6-sol","score":55,"warning":null},{"id":"panel-b9819b2ddfd0ed655f9230ba","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":60,"warning":null},{"id":"panel-673b3503bf541b6e478878f4","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":60,"warning":null},{"id":"panel-4e9e5ef059d5f6b13b1104ae","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; grok-4.7-xhigh; checkpoint=grok-4.7","score":55,"warning":null},{"id":"panel-3eff14e4aa41826f6862f62c","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-python","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; mistral-large-4-high; checkpoint=mistral-large-4","score":60,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-cb4d12433d4ef2158a779de3","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":55,"warning":null},{"id":"panel-1c622171ac4a61c0905126a8","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":50,"warning":null},{"id":"panel-60b5fcefe36e164667e674a2","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-python","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":55,"warning":null},{"id":"panel-20dee3cbee18a22247270619","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-python","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-python; muse-wonder-1.0-max; checkpoint=muse-wonder","score":60,"warning":null},{"id":"panel-c8fc51de7487c1b2d258c0f2","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":72.033,"warning":null},{"id":"panel-e497283877b2504635abf48e","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":70.167,"warning":null},{"id":"panel-f3c8cb30728a0d584b76d951","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":50.25,"warning":null},{"id":"panel-f1c6d5c59e5f097944e2a365","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":62.783,"warning":null},{"id":"panel-5df71994be4789f38c407267","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":54.95,"warning":null},{"id":"panel-19218562af1e33a655a35d87","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":59.15,"warning":null},{"id":"panel-74ec4158f36d43fe6107c6cb","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":65.5,"warning":null},{"id":"panel-69f8e076de1d08be8612d652","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":62,"warning":null},{"id":"panel-c4700c0736a7c76bf94c8d04","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":61.583,"warning":null},{"id":"panel-7395d6919946c2adfaf7424b","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-simplify","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":62.933,"warning":null},{"id":"panel-91e73f1dddb884058280f326","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":60.65,"warning":null},{"id":"panel-7c561f5d4d3417205b165aa0","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":50.85,"warning":null},{"id":"panel-202a3f3a56717b37d9bf457e","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":72.367,"warning":null},{"id":"panel-0e57c643f0410832cbc9fb48","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":71.05,"warning":null},{"id":"panel-7b35407610dfe74743495040","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":69.1,"warning":null},{"id":"panel-5c1bdcb24909d99616ed5841","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":68.017,"warning":null},{"id":"panel-ed8c7db295ec15f18a8163ae","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":75.7,"warning":null},{"id":"panel-c238833b50dc00f09d833ac6","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":74.617,"warning":null},{"id":"panel-117a0e497718ad4b2bcd3af5","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":75.4,"warning":null},{"id":"panel-790eb22013b7deb683b85849","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":54.083,"warning":null},{"id":"panel-6f5160d5e1a3874da330f091","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":70,"warning":null},{"id":"panel-7f34159560b42fbf4f72e29d","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":54.067,"warning":null},{"id":"panel-218cd2b7f3fa4ba48808e4fc","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":62.75,"warning":null},{"id":"panel-7f78f88694ccc33fb7e0c3d3","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":66.317,"warning":null},{"id":"panel-332b86a089da785cdea3f402","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":58.667,"warning":null},{"id":"panel-1bbc8f3b917da240bd97fec6","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":70.517,"warning":null},{"id":"panel-949d2db642838c34e2d944f3","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":61.433,"warning":null},{"id":"panel-759a23b0d450b5e3fb8ee226","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-6-astra-max; checkpoint=gpt-6-astra","score":70.533,"warning":null},{"id":"panel-a2680a88d56e864a7c5930cf","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-6-luna-max; checkpoint=gpt-6-luna","score":55.433,"warning":null},{"id":"panel-ca1a4d244e65b320405a7125","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-6-sol-max; checkpoint=gpt-6-sol","score":69.283,"warning":null},{"id":"panel-b82a3101ea67adcac68072dc","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":71.967,"warning":null},{"id":"panel-2056196c29b5798acd9f7a87","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":68.167,"warning":null},{"id":"panel-f6561ed3d1bce5caa9f84e40","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; grok-4.7-xhigh; checkpoint=grok-4.7","score":69.967,"warning":null},{"id":"panel-1601a1d64a7a9e15152dab7f","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-simplify","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; mistral-large-4-high; checkpoint=mistral-large-4","score":66.183,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-fd205044a78778402333b4ab","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":65.7,"warning":null},{"id":"panel-c5de3c6d1577f440b4c8c784","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":68.733,"warning":null},{"id":"panel-21fd85624e4a16f80678eeb6","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-simplify","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":72.517,"warning":null},{"id":"panel-9900ad797a426c46f91cdc10","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-simplify","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-simplify; muse-wonder-1.0-max; checkpoint=muse-wonder","score":67.05,"warning":null},{"id":"panel-983329bcfa432e4151bc9bc2","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":96,"warning":null},{"id":"panel-a7d2871bbc6917780c8d21b3","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":100,"warning":null},{"id":"panel-309d0bc9c87d6f4cd495bdb2","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":92,"warning":null},{"id":"panel-d2ba8db92d11d77e230396b8","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":94,"warning":null},{"id":"panel-870d52baedb6499434a24a51","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":96,"warning":null},{"id":"panel-a794216ca56ac67f785dad20","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":98,"warning":null},{"id":"panel-5d21ba95ec8fdb8b8e301fac","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":100,"warning":null},{"id":"panel-c8c35a24dc4b0f79d4adbb2f","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":98,"warning":null},{"id":"panel-ff77a6c3a9e65b6ccb8e3a7c","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":100,"warning":null},{"id":"panel-a0dacead0b0fff9e77e638df","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-spatial","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":100,"warning":null},{"id":"panel-6e17221392c2b73cb835e55c","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":100,"warning":null},{"id":"panel-663d8d1a55edbd6bbc306613","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":98,"warning":null},{"id":"panel-faad84623223eca51434985f","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":98,"warning":null},{"id":"panel-6af50c99811132886f95297a","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":98,"warning":null},{"id":"panel-779fc91b62607caad532fe71","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":96,"warning":null},{"id":"panel-6a0b5beb7d2564d7dfeee8dd","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":90,"warning":null},{"id":"panel-62c1d3bad9ee9ff062c24b7c","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":100,"warning":null},{"id":"panel-bcfd094cfd1370b8184eb917","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":100,"warning":null},{"id":"panel-e0bda6cdff53e878f032c374","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":100,"warning":null},{"id":"panel-8cb5768c95cfc9d9ee948a32","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":96,"warning":null},{"id":"panel-52f73ccb84c0dad0faf4db74","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":98,"warning":null},{"id":"panel-cbdc096fb05b58e68fa0a78b","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":88,"warning":null},{"id":"panel-5871d8133517ae7eab0f50d6","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":82,"warning":null},{"id":"panel-a54de73dfe62d0d252c262de","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":98,"warning":null},{"id":"panel-6b6298d5105594af69577e2b","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":96,"warning":null},{"id":"panel-aa139a442c1d74acd762316c","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":100,"warning":null},{"id":"panel-d65e406870344792f02f74dc","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":100,"warning":null},{"id":"panel-29daa78a29277b8e3c4e1935","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-6-astra-max; checkpoint=gpt-6-astra","score":98,"warning":null},{"id":"panel-f8b05e805333b62ec847be0b","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-6-luna-max; checkpoint=gpt-6-luna","score":96,"warning":null},{"id":"panel-b1ec8369fb5b5f6055dca378","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-6-sol-max; checkpoint=gpt-6-sol","score":96,"warning":null},{"id":"panel-ab04896af2769f8333c21634","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":98,"warning":null},{"id":"panel-5014e38f5ba7cc04a17abdda","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":98,"warning":null},{"id":"panel-13ea0487b82427d49282c61a","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; grok-4.7-xhigh; checkpoint=grok-4.7","score":94,"warning":null},{"id":"panel-82eddc3d36552642e464eff7","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-spatial","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; mistral-large-4-high; checkpoint=mistral-large-4","score":94,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-79c0b925029bc6018d77d2de","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":100,"warning":null},{"id":"panel-315c13eaf06cba3126de4186","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":96,"warning":null},{"id":"panel-6c3961f5f7f2f61d3927bbf6","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-spatial","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":100,"warning":null},{"id":"panel-dca032e1e3b8025ad77f46f2","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-spatial","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-spatial; muse-wonder-1.0-max; checkpoint=muse-wonder","score":96,"warning":null},{"id":"panel-2af525469b096e71b83c05f7","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":71.7,"warning":null},{"id":"panel-cceb36ce7791a1eb1562144d","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":78.1,"warning":null},{"id":"panel-d6c3c007f75e79b95f868f27","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":64.533,"warning":null},{"id":"panel-b1947c445edfa98ec575d12d","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":73.6,"warning":null},{"id":"panel-e87a024bc711a8588ca5c4a5","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":65.85,"warning":null},{"id":"panel-ba137339713465f06d96bfe2","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":66.133,"warning":null},{"id":"panel-fb153f17dcaa7568a31cc10e","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":69.4,"warning":null},{"id":"panel-f24c3abaf143433907c54872","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":80.6,"warning":null},{"id":"panel-3e98d27984c5af805c065745","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":61.317,"warning":null},{"id":"panel-97a9005f0a93edb4193e1c8a","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-story_generation","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":68.9,"warning":null},{"id":"panel-6b40d917d05afe59030e43e3","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":62.317,"warning":null},{"id":"panel-88551f37720f4b4b9b9edf71","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":69.917,"warning":null},{"id":"panel-a7560c71dfddbecf455c3873","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":77.083,"warning":null},{"id":"panel-1daa1fc476d17b814e17715a","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":77.25,"warning":null},{"id":"panel-8db1335a3d1c4eb91f13f494","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":75.2,"warning":null},{"id":"panel-c11b52096eae273aa28d6c38","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":67.467,"warning":null},{"id":"panel-3b96713def9718e318008623","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":75.6,"warning":null},{"id":"panel-386333b19f9f51f693fcf5f9","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":82.25,"warning":null},{"id":"panel-717fe68688b1cd88d3865fa7","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":82.75,"warning":null},{"id":"panel-eaa5fc61415b8aa88e767880","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":65.7,"warning":null},{"id":"panel-930496933f2794492ab365b2","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":79.017,"warning":null},{"id":"panel-2d9a5f1ed5367f0741ff26d2","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":66.633,"warning":null},{"id":"panel-7a8291feea5ecd53861d115b","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":67.653,"warning":null},{"id":"panel-ecbe45e5b2d5de3f475a6b9a","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":72.55,"warning":null},{"id":"panel-cc1676839b16b4cae9c2b786","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":69.067,"warning":null},{"id":"panel-f5913ad96576103aa18ce7e0","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":75.917,"warning":null},{"id":"panel-acc39098a765b02240f927e9","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":66.833,"warning":null},{"id":"panel-ea75961dc8f3284bcade8e4d","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-6-astra-max; checkpoint=gpt-6-astra","score":73.983,"warning":null},{"id":"panel-c846f00e2443a2d618049b21","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-6-luna-max; checkpoint=gpt-6-luna","score":63.283,"warning":null},{"id":"panel-ae6cd93f49044aec4a1f1cb9","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-6-sol-max; checkpoint=gpt-6-sol","score":66.933,"warning":null},{"id":"panel-2458c11ec127d17cfab9c56b","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":70.817,"warning":null},{"id":"panel-8a5f565b32271328e4e7566f","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":70.567,"warning":null},{"id":"panel-f3a299d30f60c54be14032b5","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; grok-4.7-xhigh; checkpoint=grok-4.7","score":74.917,"warning":null},{"id":"panel-f7389adfe8fd745b3a0404ca","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-story_generation","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; mistral-large-4-high; checkpoint=mistral-large-4","score":67.5,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-643a2b102ca96cc1c855d0d9","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":69.15,"warning":null},{"id":"panel-6e422918805d22eb5f088d67","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":73.65,"warning":null},{"id":"panel-42a2d7c7037fb97d8e799de8","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-story_generation","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":77.483,"warning":null},{"id":"panel-d3299870468757c818165bba","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-story_generation","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-story_generation; muse-wonder-1.0-max; checkpoint=muse-wonder","score":74.167,"warning":null},{"id":"panel-dc7ca75625f95fde3437e823","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":81.7,"warning":null},{"id":"panel-6c20b285e58d1c77b49b2468","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":72.167,"warning":null},{"id":"panel-0f24c7415b7607f609a36fa6","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":53.633,"warning":null},{"id":"panel-f24cbe13af649d665688ad22","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":66.983,"warning":null},{"id":"panel-d11017bb0a00fabdb4412bad","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":63.717,"warning":null},{"id":"panel-f627cd7090aff302ea1688bd","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":65.15,"warning":null},{"id":"panel-2feb2a233295800339a769e5","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":66.283,"warning":null},{"id":"panel-520b3b60cfad57f258de5139","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":77.667,"warning":null},{"id":"panel-f6473e04c058c6372d74263c","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":66.433,"warning":null},{"id":"panel-10dfde1dfd55d1050db26ccd","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-summarize","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":61.65,"warning":null},{"id":"panel-1a215fce1b280d87fa9122ee","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":71.217,"warning":null},{"id":"panel-1cbfecf4550a313cdb271f27","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":58.167,"warning":null},{"id":"panel-7e0c8ddba0c82132b57a806b","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":70.333,"warning":null},{"id":"panel-0ae689afa56fc1952d84c292","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":88.233,"warning":null},{"id":"panel-69d89ddfd6815cc3c5feb368","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":82.583,"warning":null},{"id":"panel-82b98df2935ac4c2e6f10aba","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":68.467,"warning":null},{"id":"panel-422f1455cd480504d8bec31c","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":75.85,"warning":null},{"id":"panel-d4152f1599eec1bd1bc2db33","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":84.767,"warning":null},{"id":"panel-71ca1e47d78479714d961f87","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":87.5,"warning":null},{"id":"panel-14610010800749fab1e3696a","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":63.2,"warning":null},{"id":"panel-4844de203c7156b6d863ace0","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":68.417,"warning":null},{"id":"panel-8c3e5d2106f12b3a8b88ef1b","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":59.633,"warning":null},{"id":"panel-7c91d12d311b446ed2b720d6","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":71.767,"warning":null},{"id":"panel-cf744fd7b4c62f6358037653","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":72.317,"warning":null},{"id":"panel-2c494e13cd9d39187d4dfaf9","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":64.933,"warning":null},{"id":"panel-350f2da94844393f21c11f7e","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":69.433,"warning":null},{"id":"panel-eea8659463585ccb5e9e0a4b","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":73.167,"warning":null},{"id":"panel-947aa825ea0ab05af778ef0a","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-6-astra-max; checkpoint=gpt-6-astra","score":79.55,"warning":null},{"id":"panel-68616e51e7d88b4fb0f87fa9","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-6-luna-max; checkpoint=gpt-6-luna","score":55.317,"warning":null},{"id":"panel-bb0c10a5a40fdfb2c6ba888b","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-6-sol-max; checkpoint=gpt-6-sol","score":68,"warning":null},{"id":"panel-2508ce88f4bb55a342194fa6","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":81.25,"warning":null},{"id":"panel-e5cfaf3ae482dfaae2ff0314","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":75.533,"warning":null},{"id":"panel-731812b88cf83b8e27677bc4","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; grok-4.7-xhigh; checkpoint=grok-4.7","score":78.167,"warning":null},{"id":"panel-129bf3385c08a43a5f9c9647","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-summarize","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; mistral-large-4-high; checkpoint=mistral-large-4","score":61.633,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-87b456d84438ff05a853f952","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":74.3,"warning":null},{"id":"panel-9e2379b2378edc420ca56706","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":80.617,"warning":null},{"id":"panel-753a406f6be326ef24fd319e","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-summarize","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":83.4,"warning":null},{"id":"panel-8a5c3b62302b72bf46b1755f","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-summarize","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-summarize; muse-wonder-1.0-max; checkpoint=muse-wonder","score":76.833,"warning":null},{"id":"panel-29667ef98b3f5bca5f681c85","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":56.135,"warning":null},{"id":"panel-0deb807684d82cf80a5af7d2","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":56.154,"warning":null},{"id":"panel-9d1be95f697a9fcab4a2f936","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":46.673,"warning":null},{"id":"panel-0e2707775f1b1b2efb1595a6","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":49.173,"warning":null},{"id":"panel-6e02eab7a3e609f6dd618035","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":45.923,"warning":null},{"id":"panel-83f1eb7d5def019910aba669","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":48.173,"warning":null},{"id":"panel-f2174ccbca480f517cbd3ca4","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":47.231,"warning":null},{"id":"panel-e90fea5b53cba5c187c6980e","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":51.077,"warning":null},{"id":"panel-3f2a9bc3bdb1688bb3f3d825","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":51.962,"warning":null},{"id":"panel-2159cc97a8071f41f95b2bac","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-tablejoin","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":45.077,"warning":null},{"id":"panel-393d1ac80d2e548ca5be81d3","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":42.442,"warning":null},{"id":"panel-01a0f8eaba8435ceda425f94","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":39.808,"warning":null},{"id":"panel-485fe406ee1458cf1cb6d9ab","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":46.885,"warning":null},{"id":"panel-35844166e61bc0d5a71876d2","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":52.346,"warning":null},{"id":"panel-d341295ae8ea744fc27e72c3","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":50.5,"warning":null},{"id":"panel-6f800c82438e0951dfa9fdfc","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":40.615,"warning":null},{"id":"panel-40d49ac5b8305a87cd60dc70","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":47.135,"warning":null},{"id":"panel-e7c14e2adb74037187f0ec1c","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":45.481,"warning":null},{"id":"panel-850a0242aec8243d437d32ec","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":45.327,"warning":null},{"id":"panel-ef9a0693d34f00cba25d1252","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":45.577,"warning":null},{"id":"panel-eca935ff02296892416ebe55","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":51.75,"warning":null},{"id":"panel-5b692cff35fdaff8c4dbf68d","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":49.385,"warning":null},{"id":"panel-96e3931bb0118e7f37553bc7","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":50.588,"warning":null},{"id":"panel-4f0671e99c8eb258b957ac14","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":55.904,"warning":null},{"id":"panel-a08e802836cc62463a64dce4","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":47.596,"warning":null},{"id":"panel-b68a09f2b5c46d2ae1d93b4d","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":49.308,"warning":null},{"id":"panel-15c08139fc4d22bca2d29df2","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":50.25,"warning":null},{"id":"panel-f50eede8d326b80321f28443","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-6-astra-max; checkpoint=gpt-6-astra","score":58.365,"warning":null},{"id":"panel-9beac39a82688739c9ddf636","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-6-luna-max; checkpoint=gpt-6-luna","score":49.615,"warning":null},{"id":"panel-74edf4942452c5c1b6fbab3b","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-6-sol-max; checkpoint=gpt-6-sol","score":54.019,"warning":null},{"id":"panel-af3833d1121c3a66f1fdf8de","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":57.423,"warning":null},{"id":"panel-d7420082932f9713a55ef9ea","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":55.25,"warning":null},{"id":"panel-6446ae5edd6721dc3adaea29","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; grok-4.7-xhigh; checkpoint=grok-4.7","score":44.923,"warning":null},{"id":"panel-0790444be4f573a3aef2e87f","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-tablejoin","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; mistral-large-4-high; checkpoint=mistral-large-4","score":49.096,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-378ce00ff970314e97d7c872","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":46.558,"warning":null},{"id":"panel-a4fc6c6ff3adc5d9f63e5f13","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":46.827,"warning":null},{"id":"panel-a024960d51ddeaa1a98abbb1","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-tablejoin","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":48.192,"warning":null},{"id":"panel-399e44bafc57b20172daf0ad","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-tablejoin","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablejoin; muse-wonder-1.0-max; checkpoint=muse-wonder","score":46.327,"warning":null},{"id":"panel-1b4c3991513f3ba7d226a43b","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":94.118,"warning":null},{"id":"panel-c99273c2737ca494de2a88bb","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":94.118,"warning":null},{"id":"panel-4ca467be66261e18a35fb6ea","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":98.039,"warning":null},{"id":"panel-24a6439acb54efac37d869bd","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":98.039,"warning":null},{"id":"panel-4e535ae7a8b06ef2cbfa9172","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":98.039,"warning":null},{"id":"panel-4d1ffd4d663cda024c844433","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":98.039,"warning":null},{"id":"panel-656470dfae4f4b91de1cf550","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":98.039,"warning":null},{"id":"panel-502f63677ac7031d9e46c35a","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":94.118,"warning":null},{"id":"panel-af88304c0bd2108ce908a459","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":94.118,"warning":null},{"id":"panel-a6622acee6854a8492d2d048","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-tablereformat","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":96.078,"warning":null},{"id":"panel-357d096335d24889fd34958c","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":98.039,"warning":null},{"id":"panel-11353ac09e4e1e58406e2979","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":98.039,"warning":null},{"id":"panel-ccb434288bd324d259c666b5","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":98.039,"warning":null},{"id":"panel-7d205d81c52a64897f3ddf2c","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":98.039,"warning":null},{"id":"panel-3e5264e1650bf46bf9444836","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":96.078,"warning":null},{"id":"panel-e77cd1796c179d78dc2ff4d7","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":98.039,"warning":null},{"id":"panel-e651b88f8758433c31ceeef5","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":98.039,"warning":null},{"id":"panel-61db091bb2c1f4799d64c0bc","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":98.039,"warning":null},{"id":"panel-968500870cb002c7a5b28066","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":96.078,"warning":null},{"id":"panel-31682b190232b4280ccd06b0","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":100,"warning":null},{"id":"panel-dd7582c72728fb5b3755e281","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":100,"warning":null},{"id":"panel-068a45e1f7235cccd90ff94f","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":100,"warning":null},{"id":"panel-6db09ec2337dab314cc9e815","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":98.039,"warning":null},{"id":"panel-61761890b0724129ed2367e2","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":100,"warning":null},{"id":"panel-372faf44f31cc0197fc03175","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":100,"warning":null},{"id":"panel-3b4dca0a810d317164c9b966","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":100,"warning":null},{"id":"panel-eb688c9af7ea6d23fe75bb6f","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":100,"warning":null},{"id":"panel-1fc821825e1232dce70ea7f7","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-6-astra-max; checkpoint=gpt-6-astra","score":100,"warning":null},{"id":"panel-a13b21030b066f91d50782e5","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-6-luna-max; checkpoint=gpt-6-luna","score":100,"warning":null},{"id":"panel-8c4b556d21f3678659ce21b1","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-6-sol-max; checkpoint=gpt-6-sol","score":100,"warning":null},{"id":"panel-2a91759316e40eda7915432e","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":100,"warning":null},{"id":"panel-d4467380405a91780cdede94","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":100,"warning":null},{"id":"panel-56e7d01f9fa933b0101d1680","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; grok-4.7-xhigh; checkpoint=grok-4.7","score":100,"warning":null},{"id":"panel-59b1300c9612863d90094bd2","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-tablereformat","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; mistral-large-4-high; checkpoint=mistral-large-4","score":100,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-5f71b89ede220ebd20a0c531","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":100,"warning":null},{"id":"panel-860ece53551bde0e2391fb57","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":100,"warning":null},{"id":"panel-4e85d1ceaaff86a48099f140","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-tablereformat","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":100,"warning":null},{"id":"panel-83bbb59d6ada1aae15c5feeb","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-tablereformat","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-tablereformat; muse-wonder-1.0-max; checkpoint=muse-wonder","score":100,"warning":null},{"id":"panel-3677adeb13582e5bebeb40d9","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":84.615,"warning":null},{"id":"panel-16ade649934f26164cea7aa7","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":80.769,"warning":null},{"id":"panel-1d67d70525fdb37b7393936c","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":86.538,"warning":null},{"id":"panel-f1021f786873558747419bfe","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":76.923,"warning":null},{"id":"panel-9eb761881c53f420662c77f0","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":78.846,"warning":null},{"id":"panel-ef6437efce9a4ee894b838eb","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":82.692,"warning":null},{"id":"panel-7e46697958c6097bfdd5bb2c","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":80.769,"warning":null},{"id":"panel-f91ef1733b24ae4b57da78dc","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":80.769,"warning":null},{"id":"panel-3884a3a33eb9bae7cf35d1c8","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":78.846,"warning":null},{"id":"panel-e5f6c6af7b7b6563cbd85d04","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":73.077,"warning":null},{"id":"panel-ddb61e1c223050b01fe6a24d","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":80.769,"warning":null},{"id":"panel-75eae8669d73a09ea4debaef","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":86.538,"warning":null},{"id":"panel-4d1460d97a4be3dfbcfa655f","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":69.231,"warning":null},{"id":"panel-6f887f9f34bf0ced09693be5","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":80.769,"warning":null},{"id":"panel-3130442bd69925da7c05ade7","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":80.769,"warning":null},{"id":"panel-0ee4308b1b8e689522786fba","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":50,"warning":null},{"id":"panel-6ca64d4b4bf1eb5d2a9614db","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":78.846,"warning":null},{"id":"panel-e9a18d4e2485815fbe9dcb90","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":82.692,"warning":null},{"id":"panel-7c589b3ec3b0735db386fb6c","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":76.923,"warning":null},{"id":"panel-46ee16e43dfcef55b9cdcc31","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":78.846,"warning":null},{"id":"panel-dd23f7bf3419e3011e2563f8","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":88.462,"warning":null},{"id":"panel-1be1c1a4d35577ac29967650","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":76.923,"warning":null},{"id":"panel-6dccf16089bfddf8511bc98e","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":80.769,"warning":null},{"id":"panel-c5a26905a6b81729ce07eee7","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":84.615,"warning":null},{"id":"panel-9f16b2319931415947e23f48","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":73.077,"warning":null},{"id":"panel-1b3605fdbbfa24a01c0f8ed7","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":84.615,"warning":null},{"id":"panel-0b801630aaf53bc41f3182dd","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":86.538,"warning":null},{"id":"panel-8fabb53ed27c9588a99748d3","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-6-astra-max; checkpoint=gpt-6-astra","score":84.615,"warning":null},{"id":"panel-2c85953b28c28fb1b4a58458","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-6-luna-max; checkpoint=gpt-6-luna","score":73.077,"warning":null},{"id":"panel-c3e7e6cf337436c3a20e1933","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-6-sol-max; checkpoint=gpt-6-sol","score":84.615,"warning":null},{"id":"panel-a928c737ed245c7c3e1de7dc","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":86.538,"warning":null},{"id":"panel-b484ed0ad68f3f16ba80df33","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":86.538,"warning":null},{"id":"panel-7e3267de1425c09c7573aeff","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; grok-4.7-xhigh; checkpoint=grok-4.7","score":84.615,"warning":null},{"id":"panel-c581cbeec9d119e11354fb42","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; mistral-large-4-high; checkpoint=mistral-large-4","score":80.769,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-a3b33640a7470089412d387c","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":76.923,"warning":null},{"id":"panel-2e9188c4762961d8b305cb9d","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":80.769,"warning":null},{"id":"panel-edd291968902d4d1849cd16d","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":84.615,"warning":null},{"id":"panel-fef023790048111cf5d004c6","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-theory_of_mind","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-theory_of_mind; muse-wonder-1.0-max; checkpoint=muse-wonder","score":86.538,"warning":null},{"id":"panel-802d1135b4384df772d4deb2","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":53.333,"warning":null},{"id":"panel-6edd971d60d57fb9f81575e2","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":60,"warning":null},{"id":"panel-62cf378fd2f06d44503c1f73","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":43.333,"warning":null},{"id":"panel-8599319f26c3e7491080d45d","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":40,"warning":null},{"id":"panel-4556b91816ae6e8c2fc6246c","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":20,"warning":null},{"id":"panel-0a0b1cc3b474cddeab9176c2","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":33.333,"warning":null},{"id":"panel-a9f78e146b99566e8c8961cb","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":43.333,"warning":null},{"id":"panel-56a4d57c73bfc7aaac0a1a33","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":33.333,"warning":null},{"id":"panel-ece90ea288dc3cb037efb07e","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":43.333,"warning":null},{"id":"panel-f97c66befa16467a34a3d8f2","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-typescript","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":23.333,"warning":null},{"id":"panel-85beb6c46f4ef1f3cb15eca8","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":50,"warning":null},{"id":"panel-b305f6586277db0044cd4ebe","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":56.667,"warning":null},{"id":"panel-647f0703b37c7b8f9a55c1d3","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":46.667,"warning":null},{"id":"panel-da04aa901faef5abb22913ce","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":23.333,"warning":null},{"id":"panel-66a7edad755fbb5ac65ce9ed","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":33.333,"warning":null},{"id":"panel-430f81ce32b4f74a36ac8608","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":26.667,"warning":null},{"id":"panel-89ba0bd0c133f90bcefa0bbf","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":26.667,"warning":null},{"id":"panel-7e1e202586c5c977d8ba4553","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":46.667,"warning":null},{"id":"panel-72a945c6670ce4413e2c9775","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":40,"warning":null},{"id":"panel-53a731259937b873fe9b238a","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":36.667,"warning":null},{"id":"panel-1a73cedae57299fa448f67d8","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":43.333,"warning":null},{"id":"panel-228b8932fc82372013ddad69","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":20,"warning":null},{"id":"panel-7b17f5d1ffbaaf27acddab5b","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":26.667,"warning":null},{"id":"panel-d531c1e72e94cbf4adfbe973","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":43.333,"warning":null},{"id":"panel-d75beeb6ecb2a4aae441bdd4","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":36.667,"warning":null},{"id":"panel-c9c58ab541711b7aacf86522","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":50,"warning":null},{"id":"panel-5997db067bc91ad16df75ea2","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":46.667,"warning":null},{"id":"panel-c494b83960b7daa2e65bbace","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-6-astra-max; checkpoint=gpt-6-astra","score":43.333,"warning":null},{"id":"panel-05cb3f3d0fdc8582eb15db40","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-6-luna-max; checkpoint=gpt-6-luna","score":40,"warning":null},{"id":"panel-61b881db817e51b6acc37991","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-6-sol-max; checkpoint=gpt-6-sol","score":40,"warning":null},{"id":"panel-eae2abe5e90f2e31edb06880","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":40,"warning":null},{"id":"panel-ec40628986de43d1c20d5106","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":46.667,"warning":null},{"id":"panel-55d2a85e79aa182c3e6d7d96","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; grok-4.7-xhigh; checkpoint=grok-4.7","score":43.333,"warning":null},{"id":"panel-b12381047fc8f7dc3d32a8f2","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-typescript","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; mistral-large-4-high; checkpoint=mistral-large-4","score":43.333,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-ed41d716690c2b0b08f4b261","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":43.333,"warning":null},{"id":"panel-7c980646387d7ffe0fbb8f07","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":50,"warning":null},{"id":"panel-026b7d686de16b86eee55e61","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-typescript","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":60,"warning":null},{"id":"panel-2644d92decaf8c527130b880","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-typescript","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench Mini-SWE-Agent, 250-step limit; livebench-2026-06-25-typescript; muse-wonder-1.0-max; checkpoint=muse-wonder","score":60,"warning":null},{"id":"panel-1d9c88711562ca4bc4793bf0","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":94,"warning":null},{"id":"panel-77e6cef54a0db4bdbe46d00b","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":94,"warning":null},{"id":"panel-0b8f1ee9107451c56f673a80","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":64,"warning":null},{"id":"panel-e1877ad981b8943c76d8a59c","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":54,"warning":null},{"id":"panel-49a1a9a410cf8daaf0cf8f4e","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":78,"warning":null},{"id":"panel-025bcfd2e9a4b3b37fa9de55","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":84,"warning":null},{"id":"panel-4a63a7797adf849e01ab6269","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":80,"warning":null},{"id":"panel-5317018cf4f7cd8ae532c388","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":78,"warning":null},{"id":"panel-2e71239f081d90f01235b7b9","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":92,"warning":null},{"id":"panel-21d30b7156f9abf0cc4dad65","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-typos","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":72,"warning":null},{"id":"panel-d7fb1cbe942c47c63bd72721","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":76,"warning":null},{"id":"panel-c6d6f00354d84902d81115f0","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":86,"warning":null},{"id":"panel-3d6b475cc89eceef6d184f40","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":84,"warning":null},{"id":"panel-ecfd0cf5293ccba5bfd04b78","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":82,"warning":null},{"id":"panel-b0514842d51bae1d0371113f","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":86,"warning":null},{"id":"panel-aa1c9bf038808a60c6fff323","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":74,"warning":null},{"id":"panel-5e5c864b2d0837ecf9a3a2cc","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":88,"warning":null},{"id":"panel-7d73d21d7873e5d779a4c384","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":90,"warning":null},{"id":"panel-38523bc85154f445a6cde6ac","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":90,"warning":null},{"id":"panel-cf1e57406dae14dbc01470ea","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":82,"warning":null},{"id":"panel-699ecb5439422ac056781a58","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":82,"warning":null},{"id":"panel-6be68058cd3af8da1eca505d","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":70,"warning":null},{"id":"panel-8a84c38764db6738f2389ed4","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":54,"warning":null},{"id":"panel-9aa01fc84eae16b96abca291","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":88,"warning":null},{"id":"panel-d01a35ef69f7df80895aaa00","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":70,"warning":null},{"id":"panel-ce3efb571062193428082f97","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":84,"warning":null},{"id":"panel-5cc0741a795c50acdfcdf9c2","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":86,"warning":null},{"id":"panel-0fef39ca040d37164c362ec0","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-6-astra-max; checkpoint=gpt-6-astra","score":82,"warning":null},{"id":"panel-46a4ea2215cfd141be5cc823","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-6-luna-max; checkpoint=gpt-6-luna","score":64,"warning":null},{"id":"panel-67d393c12786bf763fa6ecc0","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-6-sol-max; checkpoint=gpt-6-sol","score":78,"warning":null},{"id":"panel-47d84427c910e35000e25811","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":86,"warning":null},{"id":"panel-c1607342904340c98d7ad4da","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":84,"warning":null},{"id":"panel-52b987e4055d035ea7f9510e","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; grok-4.7-xhigh; checkpoint=grok-4.7","score":70,"warning":null},{"id":"panel-01811e82cd9297d95f6995b3","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-typos","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; mistral-large-4-high; checkpoint=mistral-large-4","score":60,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-607ac2b71c6861f6dfcf5271","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":62,"warning":null},{"id":"panel-ca5efb697bbdbfd817a0dc54","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":74,"warning":null},{"id":"panel-9a0cadd9cf939c95da4dccd3","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-typos","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":86,"warning":null},{"id":"panel-0b01edf83568515503d6ecd0","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-typos","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-typos; muse-wonder-1.0-max; checkpoint=muse-wonder","score":78,"warning":null},{"id":"panel-8da717a8690707003f08a090","modelId":"claude-fable-5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"claude-fable-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-fable-5-max-effort; checkpoint=claude-fable-5","score":100,"warning":null},{"id":"panel-a7ae39b18ff37882f52d61dd","modelId":"claude-fable-5-1","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"claude-fable-5-1-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-fable-5-1-max-effort; checkpoint=claude-fable-5-1","score":100,"warning":null},{"id":"panel-005b4e3e01f0017cc01be455","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"claude-haiku-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-haiku-5-5-max-effort; checkpoint=claude-haiku-5-5","score":94,"warning":null},{"id":"panel-6d24509140b27f7055d9c6fa","modelId":"claude-haiku-5-5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"claude-haiku-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-haiku-5-5-xhigh-effort; checkpoint=claude-haiku-5-5","score":98,"warning":null},{"id":"panel-e723dff4aa878aa3b1f555c0","modelId":"claude-opus-4-5-20251101","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"claude-opus-4-5-20251101-thinking-64k-high-effort (thinking 64k; high effort)","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-opus-4-5-20251101-thinking-64k-high-effort; checkpoint=claude-opus-4-5-20251101","score":77.5,"warning":null},{"id":"panel-f1903bf3ae7784fac180c5c6","modelId":"claude-opus-4-6","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"claude-opus-4-6-thinking-auto-high-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-opus-4-6-thinking-auto-high-effort; checkpoint=claude-opus-4-6","score":94,"warning":null},{"id":"panel-a325f695a71ef6705d5a5597","modelId":"claude-opus-4-7","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"claude-opus-4-7-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-opus-4-7-xhigh-effort; checkpoint=claude-opus-4-7","score":98,"warning":null},{"id":"panel-7aab7aefea402a837eb331c7","modelId":"claude-opus-4-8","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"claude-opus-4-8-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-opus-4-8-max-effort; checkpoint=claude-opus-4-8","score":100,"warning":null},{"id":"panel-b76aa763373c89156231a6da","modelId":"claude-opus-5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"claude-opus-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-opus-5-max-effort; checkpoint=claude-opus-5","score":100,"warning":null},{"id":"panel-9d65818b1f976a7825472a2f","modelId":"claude-sonnet-4-6","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"medium","sourceLabel":"claude-sonnet-4-6-thinking-auto-medium-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-sonnet-4-6-thinking-auto-medium-effort; checkpoint=claude-sonnet-4-6","score":96,"warning":null},{"id":"panel-fc47f8546d6a90f124e667fd","modelId":"claude-sonnet-5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"claude-sonnet-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-sonnet-5-xhigh-effort; checkpoint=claude-sonnet-5","score":88,"warning":null},{"id":"panel-bd3dd825b37c662e198ce2cb","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"claude-sonnet-5-5-max-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-sonnet-5-5-max-effort; checkpoint=claude-sonnet-5-5","score":100,"warning":null},{"id":"panel-236c99dd384165ec1c072c28","modelId":"claude-sonnet-5-5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"claude-sonnet-5-5-xhigh-effort","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; claude-sonnet-5-5-xhigh-effort; checkpoint=claude-sonnet-5-5","score":100,"warning":null},{"id":"panel-67097dc7f9543964f1eb66f1","modelId":"gemini-3.1-pro","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"gemini-3.1-pro-preview-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gemini-3.1-pro-preview-high; checkpoint=gemini-3.1-pro","score":85.25,"warning":null},{"id":"panel-99a523bc66fc8f9af24e40c7","modelId":"gemini-3.5-flash","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"gemini-3.5-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gemini-3.5-flash-high; checkpoint=gemini-3.5-flash","score":77.25,"warning":null},{"id":"panel-bf109486b4ee154dd5bf538c","modelId":"gemini-3.5-flash-lite","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"gemini-3.5-flash-lite-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gemini-3.5-flash-lite-high; checkpoint=gemini-3.5-flash-lite","score":32.75,"warning":null},{"id":"panel-2fadbe78a18174dfa36798ad","modelId":"gemini-3.6-flash","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"gemini-3.6-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gemini-3.6-flash-high; checkpoint=gemini-3.6-flash","score":85.75,"warning":null},{"id":"panel-d40d3cee5878b364f2758bf1","modelId":"gemini-3.7-flash","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"gemini-3.7-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gemini-3.7-flash-high; checkpoint=gemini-3.7-flash","score":94.5,"warning":null},{"id":"panel-1705e80b84b9b5b17c4be595","modelId":"gemini-3.8-flash","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"gemini-3.8-flash-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gemini-3.8-flash-high; checkpoint=gemini-3.8-flash","score":98.25,"warning":null},{"id":"panel-cdd9ee7d07449e206cc9e3dc","modelId":"gpt-5.2","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"gpt-5.2-2025-12-11-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.2-2025-12-11-high; checkpoint=gpt-5.2","score":84,"warning":null},{"id":"panel-52b264b698e93d9c6004f43d","modelId":"gpt-5.4","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"gpt-5.4-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.4-xhigh; checkpoint=gpt-5.4","score":98,"warning":null},{"id":"panel-cf093414bb1c79faa8013d1c","modelId":"gpt-5.4-mini","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"gpt-5.4-mini-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.4-mini-xhigh; checkpoint=gpt-5.4-mini","score":60.372,"warning":null},{"id":"panel-31bd5c440bbe0254845c74b9","modelId":"gpt-5.4-nano","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"gpt-5.4-nano-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.4-nano-xhigh; checkpoint=gpt-5.4-nano","score":97.619,"warning":null},{"id":"panel-b6b4a2a1c5b5a3d4aa7ecec7","modelId":"gpt-5.5","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"gpt-5.5-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.5-xhigh; checkpoint=gpt-5.5","score":100,"warning":null},{"id":"panel-41dd83b36bf6c44b1f57ac1d","modelId":"gpt-5.6-luna","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"gpt-5.6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.6-luna-max; checkpoint=gpt-5.6-luna","score":93.5,"warning":null},{"id":"panel-2f9395d08746fc6829802981","modelId":"gpt-5.6-sol","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"gpt-5.6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.6-sol-max; checkpoint=gpt-5.6-sol","score":100,"warning":null},{"id":"panel-c3767e7e55780617f0ade91d","modelId":"gpt-5.6-terra","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"gpt-5.6-terra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-5.6-terra-max; checkpoint=gpt-5.6-terra","score":100,"warning":null},{"id":"panel-1ffa9572fdf497d938c73d9d","modelId":"gpt-6-astra","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"gpt-6-astra-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-6-astra-max; checkpoint=gpt-6-astra","score":100,"warning":null},{"id":"panel-8a633eac0cf729f5cb0f32e4","modelId":"gpt-6-luna","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"gpt-6-luna-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-6-luna-max; checkpoint=gpt-6-luna","score":96,"warning":null},{"id":"panel-cb46ec39cb329fdaf0f24235","modelId":"gpt-6-sol","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"gpt-6-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-6-sol-max; checkpoint=gpt-6-sol","score":100,"warning":null},{"id":"panel-29bede6d1ff0f87cfdda8626","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"gpt-6.1-sol-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-6.1-sol-max; checkpoint=gpt-6.1-sol","score":100,"warning":null},{"id":"panel-0fda127c4d60ebb429cf5124","modelId":"gpt-6.1-sol","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"gpt-6.1-sol-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; gpt-6.1-sol-xhigh; checkpoint=gpt-6.1-sol","score":100,"warning":null},{"id":"panel-851f97e929141acd934cc57a","modelId":"grok-4.7","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"grok-4.7-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; grok-4.7-xhigh; checkpoint=grok-4.7","score":86,"warning":null},{"id":"panel-1b5121c6aff4d3d27981001a","modelId":"mistral-large-4","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"high","sourceLabel":"mistral-large-4-high","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; mistral-large-4-high; checkpoint=mistral-large-4","score":93,"warning":"LiveBench reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"panel-3ca3b50afff257eb11c4cb04","modelId":"muse-spark-1.1","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"muse-spark-1.1-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; muse-spark-1.1-xhigh; checkpoint=muse-spark-1.1","score":96,"warning":null},{"id":"panel-5924987a533b33a47f789e7e","modelId":"muse-spark-1.2","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"muse-spark-1.2-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; muse-spark-1.2-xhigh; checkpoint=muse-spark-1.2","score":95.25,"warning":null},{"id":"panel-1a10629747e118f97cfc0f67","modelId":"muse-spark-1.3","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"xhigh","sourceLabel":"muse-spark-1.3-xhigh","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; muse-spark-1.3-xhigh; checkpoint=muse-spark-1.3","score":98,"warning":null},{"id":"panel-53adcf37a1fd967c5d7edbdb","modelId":"muse-wonder","benchmarkId":"livebench-2026-06-25-zebra_puzzle","level":"max","sourceLabel":"muse-wonder-1.0-max","configuration":"LiveBench official task evaluation; livebench-2026-06-25-zebra_puzzle; muse-wonder-1.0-max; checkpoint=muse-wonder","score":100,"warning":null},{"id":"panel-eeaabbca375641a41f9d6033","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-08-tools","level":"high","sourceLabel":"gpt-5-2025-08-07-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-08-tools; gpt-5-2025-08-07-high; checkpoint=gpt-5","score":46.53846153846154,"warning":null},{"id":"panel-df1f433349e12b715f87b8ca","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-08-tools","level":"medium","sourceLabel":"gpt-5-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-08-tools; gpt-5-2025-08-07-medium; checkpoint=gpt-5","score":45.38461538461539,"warning":null},{"id":"panel-7aa0c89e21a5b4ba53c846fe","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-08-tools","level":"minimal","sourceLabel":"gpt-5-2025-08-07-minimal","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-08-tools; gpt-5-2025-08-07-minimal; checkpoint=gpt-5","score":33.46153846153846,"warning":null},{"id":"panel-8f5f1ae06f1007eabaadad12","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2025-08-tools","level":"medium","sourceLabel":"gpt-5-mini-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-08-tools; gpt-5-mini-2025-08-07-medium; checkpoint=gpt-5-mini","score":43.07692307692307,"warning":null},{"id":"panel-462303d94f9258ff82feec79","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-09-tools","level":"high","sourceLabel":"gpt-5-2025-08-07-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-09-tools; gpt-5-2025-08-07-high; checkpoint=gpt-5","score":36.3265306122449,"warning":null},{"id":"panel-6a5dad63f041041541673c1a","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-09-tools","level":"medium","sourceLabel":"gpt-5-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-09-tools; gpt-5-2025-08-07-medium; checkpoint=gpt-5","score":38.775510204081634,"warning":null},{"id":"panel-8613396169af318add6485b7","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-09-tools","level":"minimal","sourceLabel":"gpt-5-2025-08-07-minimal","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-09-tools; gpt-5-2025-08-07-minimal; checkpoint=gpt-5","score":30.612244897959183,"warning":null},{"id":"panel-4b776aced4fe63c3161c7bb8","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2025-09-tools","level":"medium","sourceLabel":"gpt-5-mini-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-09-tools; gpt-5-mini-2025-08-07-medium; checkpoint=gpt-5-mini","score":37.14285714285714,"warning":null},{"id":"panel-eeadcd73222785f2a270b0be","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-10-tools","level":"high","sourceLabel":"gpt-5-2025-08-07-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-10-tools; gpt-5-2025-08-07-high; checkpoint=gpt-5","score":38.82352941176471,"warning":null},{"id":"panel-128f7a0af9b7fa3967594c52","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-10-tools","level":"medium","sourceLabel":"gpt-5-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-10-tools; gpt-5-2025-08-07-medium; checkpoint=gpt-5","score":42.35294117647059,"warning":null},{"id":"panel-3e55e76525c8e98cf682c126","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-10-tools","level":"minimal","sourceLabel":"gpt-5-2025-08-07-minimal","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-10-tools; gpt-5-2025-08-07-minimal; checkpoint=gpt-5","score":25.968627450980396,"warning":null},{"id":"panel-7d9e8814bc198c5ebfeecf3a","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2025-10-tools","level":"medium","sourceLabel":"gpt-5-mini-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-10-tools; gpt-5-mini-2025-08-07-medium; checkpoint=gpt-5-mini","score":32.549019607843135,"warning":null},{"id":"panel-105cf9d397ba788c03ce8531","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-11-tools","level":"high","sourceLabel":"gpt-5-2025-08-07-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-11-tools; gpt-5-2025-08-07-high; checkpoint=gpt-5","score":55.74468085106383,"warning":null},{"id":"panel-5b22352026768409d27902f3","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-11-tools","level":"medium","sourceLabel":"gpt-5-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-11-tools; gpt-5-2025-08-07-medium; checkpoint=gpt-5","score":58.723404255319146,"warning":null},{"id":"panel-12e0ccd4a964c83e4b36650b","modelId":"gpt-5","benchmarkId":"swe-rebench-2025-11-tools","level":"minimal","sourceLabel":"gpt-5-2025-08-07-minimal","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-11-tools; gpt-5-2025-08-07-minimal; checkpoint=gpt-5","score":34.8936170212766,"warning":null},{"id":"panel-c89e195031cf4c7684a9a9b0","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2025-11-tools","level":"medium","sourceLabel":"gpt-5-mini-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-11-tools; gpt-5-mini-2025-08-07-medium; checkpoint=gpt-5-mini","score":44.255319148936174,"warning":null},{"id":"panel-b21a69a8ceeb7196e4fbdfc9","modelId":"gpt-5.2","benchmarkId":"swe-rebench-2025-11-tools","level":"medium","sourceLabel":"gpt-5.2-2025-12-11-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-11-tools; gpt-5.2-2025-12-11-medium; checkpoint=gpt-5.2","score":61.27659574468085,"warning":null},{"id":"panel-9d3595bc411ad8842b5f1eef","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2025-12-tools","level":"high","sourceLabel":"gpt-5-mini-2025-08-07-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-12-tools; gpt-5-mini-2025-08-07-high; checkpoint=gpt-5-mini","score":44.30851063829787,"warning":null},{"id":"panel-9f5fbd93e0f5873f7f2e34ad","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2025-12-tools","level":"medium","sourceLabel":"gpt-5-mini-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-12-tools; gpt-5-mini-2025-08-07-medium; checkpoint=gpt-5-mini","score":40.66489361702128,"warning":null},{"id":"panel-3158ebf36b887b27663bbccd","modelId":"gpt-5.2","benchmarkId":"swe-rebench-2025-12-tools","level":"medium","sourceLabel":"gpt-5.2-2025-12-11-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-12-tools; gpt-5.2-2025-12-11-medium; checkpoint=gpt-5.2","score":59.41489361702128,"warning":null},{"id":"panel-709e5e2b9100c29ff4ccf9e6","modelId":"gpt-5.2","benchmarkId":"swe-rebench-2025-12-tools","level":"xhigh","sourceLabel":"gpt-5.2-2025-12-11-xhigh","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-12-tools; gpt-5.2-2025-12-11-xhigh; checkpoint=gpt-5.2","score":61.48936170212765,"warning":null},{"id":"panel-46441b51f4e16452b66da4d8","modelId":"gpt-oss-120b","benchmarkId":"swe-rebench-2025-12-tools","level":"high","sourceLabel":"gpt-oss-120b-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2025-12-tools; gpt-oss-120b-high; checkpoint=gpt-oss-120b","score":36.959219858156025,"warning":null},{"id":"panel-439f39520f3958e5a7901ccc","modelId":"claude-opus-4-6","benchmarkId":"swe-rebench-2026-01-tools","level":"high","sourceLabel":"Claude Opus 4.6-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-01-tools; Claude Opus 4.6-high; checkpoint=claude-opus-4-6","score":51.666666666666686,"warning":null},{"id":"panel-882148776d6b6bf196c8e72f","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2026-01-tools","level":"high","sourceLabel":"gpt-5-mini-2025-08-07-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-01-tools; gpt-5-mini-2025-08-07-high; checkpoint=gpt-5-mini","score":35,"warning":null},{"id":"panel-c8b39cb9c7921b483f979a9c","modelId":"gpt-5-mini","benchmarkId":"swe-rebench-2026-01-tools","level":"medium","sourceLabel":"gpt-5-mini-2025-08-07-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-01-tools; gpt-5-mini-2025-08-07-medium; checkpoint=gpt-5-mini","score":30.83333333333333,"warning":null},{"id":"panel-8380ef3f8a50b6f89b396276","modelId":"gpt-5.2","benchmarkId":"swe-rebench-2026-01-tools","level":"medium","sourceLabel":"gpt-5.2-2025-12-11-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-01-tools; gpt-5.2-2025-12-11-medium; checkpoint=gpt-5.2","score":51.046099290780134,"warning":null},{"id":"panel-2a5f25442a3b5295bfa02259","modelId":"gpt-5.2","benchmarkId":"swe-rebench-2026-01-tools","level":"xhigh","sourceLabel":"gpt-5.2-2025-12-11-xhigh","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-01-tools; gpt-5.2-2025-12-11-xhigh; checkpoint=gpt-5.2","score":51.66666666666666,"warning":null},{"id":"panel-9cfedefbaaad4dfb46e7ab3e","modelId":"claude-opus-4-6","benchmarkId":"swe-rebench-2026-02-tools","level":"high","sourceLabel":"Claude Opus 4.6-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-02-tools; Claude Opus 4.6-high; checkpoint=claude-opus-4-6","score":65.26315789473685,"warning":null},{"id":"panel-31ec105465e7a26d35da4265","modelId":"gpt-5.2","benchmarkId":"swe-rebench-2026-02-tools","level":"medium","sourceLabel":"gpt-5.2-2025-12-11-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-02-tools; gpt-5.2-2025-12-11-medium; checkpoint=gpt-5.2","score":64.4172932330827,"warning":null},{"id":"panel-ca4f23c7585afe314456c6ee","modelId":"gpt-5.3-codex","benchmarkId":"swe-rebench-2026-02-tools","level":"xhigh","sourceLabel":"gpt-5.3-codex-xhigh","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-02-tools; gpt-5.3-codex-xhigh; checkpoint=gpt-5.3-codex","score":58.59649122807018,"warning":null},{"id":"panel-ef0926b5f5d004f8a5348da5","modelId":"gpt-5.4","benchmarkId":"swe-rebench-2026-02-tools","level":"medium","sourceLabel":"gpt-5.4-2026-03-05-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-02-tools; gpt-5.4-2026-03-05-medium; checkpoint=gpt-5.4","score":62.807017543859644,"warning":null},{"id":"panel-af6affc17d37d1ebceed5651","modelId":"claude-opus-4-6","benchmarkId":"swe-rebench-2026-03-tools","level":"high","sourceLabel":"Claude Opus 4.6-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; Claude Opus 4.6-high; checkpoint=claude-opus-4-6","score":51.304347826086946,"warning":null},{"id":"panel-582529e031a763a72cb0c962","modelId":"claude-opus-4-7","benchmarkId":"swe-rebench-2026-03-tools","level":"high","sourceLabel":"Claude Opus 4.7-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; Claude Opus 4.7-high; checkpoint=claude-opus-4-7","score":53.47826086956522,"warning":null},{"id":"panel-aa3ee5d91a696499570b9479","modelId":"claude-opus-4-8","benchmarkId":"swe-rebench-2026-03-tools","level":"xhigh","sourceLabel":"Claude Opus 4.8-xhigh","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; Claude Opus 4.8-xhigh; checkpoint=claude-opus-4-8","score":56.52173913043479,"warning":null},{"id":"panel-84d19d866a5f84d2e5157771","modelId":"deepseek-v4-pro-0424","benchmarkId":"swe-rebench-2026-03-tools","level":"high","sourceLabel":"DeepSeek-V4 Pro [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; DeepSeek-V4 Pro [high]; checkpoint=deepseek-v4-pro-0424","score":46.72463768115942,"warning":null},{"id":"panel-4e74e790b8c2ab2987cd3979","modelId":"glm-5.2","benchmarkId":"swe-rebench-2026-03-tools","level":"high","sourceLabel":"GLM-5.2 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; GLM-5.2 [high]; checkpoint=glm-5.2","score":53.47826086956522,"warning":null},{"id":"panel-91308a65440d721514f94949","modelId":"gpt-5.4","benchmarkId":"swe-rebench-2026-03-tools","level":"medium","sourceLabel":"gpt-5.4-2026-03-05-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; gpt-5.4-2026-03-05-medium; checkpoint=gpt-5.4","score":57.826086956521735,"warning":null},{"id":"panel-082ec7a4e998253830c59d5e","modelId":"gpt-5.5","benchmarkId":"swe-rebench-2026-03-tools","level":"medium","sourceLabel":"gpt-5.5-2026-04-23-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; gpt-5.5-2026-04-23-medium; checkpoint=gpt-5.5","score":60.434782608695656,"warning":null},{"id":"panel-03a4267243f5da43f4f0f916","modelId":"gpt-5.5","benchmarkId":"swe-rebench-2026-03-tools","level":"xhigh","sourceLabel":"gpt-5.5-2026-04-23-xhigh","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-03-tools; gpt-5.5-2026-04-23-xhigh; checkpoint=gpt-5.5","score":63.04347826086957,"warning":null},{"id":"panel-c4a47cfea3565c4484c1c8b8","modelId":"claude-opus-4-6","benchmarkId":"swe-rebench-2026-04-tools","level":"high","sourceLabel":"Claude Opus 4.6-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; Claude Opus 4.6-high; checkpoint=claude-opus-4-6","score":43.72093023255814,"warning":null},{"id":"panel-d4309f168dc4babc6bc9d587","modelId":"claude-opus-4-7","benchmarkId":"swe-rebench-2026-04-tools","level":"high","sourceLabel":"Claude Opus 4.7-high","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; Claude Opus 4.7-high; checkpoint=claude-opus-4-7","score":57.67441860465116,"warning":null},{"id":"panel-59f0399257b1743a621e4448","modelId":"claude-opus-4-8","benchmarkId":"swe-rebench-2026-04-tools","level":"xhigh","sourceLabel":"Claude Opus 4.8-xhigh","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; Claude Opus 4.8-xhigh; checkpoint=claude-opus-4-8","score":57.674418604651166,"warning":null},{"id":"panel-2e61455480fb766893d98afe","modelId":"deepseek-v4-pro-0424","benchmarkId":"swe-rebench-2026-04-tools","level":"high","sourceLabel":"DeepSeek-V4 Pro [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; DeepSeek-V4 Pro [high]; checkpoint=deepseek-v4-pro-0424","score":41.58361018826135,"warning":null},{"id":"panel-7073ced98bba7f5b5d8c68cd","modelId":"glm-5.2","benchmarkId":"swe-rebench-2026-04-tools","level":"high","sourceLabel":"GLM-5.2 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; GLM-5.2 [high]; checkpoint=glm-5.2","score":51.162790697674424,"warning":null},{"id":"panel-f0febcfcf5b14d5f6f8f1b5e","modelId":"gpt-5.4","benchmarkId":"swe-rebench-2026-04-tools","level":"medium","sourceLabel":"gpt-5.4-2026-03-05-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; gpt-5.4-2026-03-05-medium; checkpoint=gpt-5.4","score":49.76744186046512,"warning":null},{"id":"panel-1e3fa2d67e3e69ad184eb7a4","modelId":"gpt-5.5","benchmarkId":"swe-rebench-2026-04-tools","level":"medium","sourceLabel":"gpt-5.5-2026-04-23-medium","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; gpt-5.5-2026-04-23-medium; checkpoint=gpt-5.5","score":57.20930232558139,"warning":null},{"id":"panel-3f17e0eb43dae6ec089f2c0b","modelId":"gpt-5.5","benchmarkId":"swe-rebench-2026-04-tools","level":"xhigh","sourceLabel":"gpt-5.5-2026-04-23-xhigh","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-04-tools; gpt-5.5-2026-04-23-xhigh; checkpoint=gpt-5.5","score":60.46511627906976,"warning":null},{"id":"panel-5d62ab6fe9c2e7d1f8fda50c","modelId":"deepseek-v4-pro-0424","benchmarkId":"swe-rebench-2026-05-tools","level":"high","sourceLabel":"DeepSeek-V4 Pro [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-05-tools; DeepSeek-V4 Pro [high]; checkpoint=deepseek-v4-pro-0424","score":41.55555555555556,"warning":null},{"id":"panel-9b20aa7050ddddd133712208","modelId":"glm-5.2","benchmarkId":"swe-rebench-2026-05-tools","level":"high","sourceLabel":"GLM-5.2 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-05-tools; GLM-5.2 [high]; checkpoint=glm-5.2","score":61.111111111111114,"warning":null},{"id":"panel-7227c50a2326d0f76441e35b","modelId":"claude-fable-5","benchmarkId":"swe-rebench-2026-06-tools","level":"high","sourceLabel":"Fable 5 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; Fable 5 [high]; checkpoint=claude-fable-5","score":57.61904761904761,"warning":null},{"id":"panel-79497e6cca80d008679179a1","modelId":"claude-opus-5","benchmarkId":"swe-rebench-2026-06-tools","level":"high","sourceLabel":"Opus 5 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; Opus 5 [high]; checkpoint=claude-opus-5","score":60,"warning":null},{"id":"panel-8ea63df7cd359367585d2968","modelId":"claude-sonnet-5","benchmarkId":"swe-rebench-2026-06-tools","level":"high","sourceLabel":"Sonnet 5 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; Sonnet 5 [high]; checkpoint=claude-sonnet-5","score":52.85714285714285,"warning":null},{"id":"panel-17c093bae14aabe9e7a0ba93","modelId":"deepseek-v4-pro-0424","benchmarkId":"swe-rebench-2026-06-tools","level":"high","sourceLabel":"DeepSeek-V4 Pro [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; DeepSeek-V4 Pro [high]; checkpoint=deepseek-v4-pro-0424","score":35.238095238095234,"warning":null},{"id":"panel-9633b6cae80f96f07bdab378","modelId":"glm-5.2","benchmarkId":"swe-rebench-2026-06-tools","level":"high","sourceLabel":"GLM-5.2 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; GLM-5.2 [high]; checkpoint=glm-5.2","score":58.0952380952381,"warning":null},{"id":"panel-af745c954411352e8ff6dfe1","modelId":"gpt-5.6-luna","benchmarkId":"swe-rebench-2026-06-tools","level":"medium","sourceLabel":"GPT-5.6 Luna [medium]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; GPT-5.6 Luna [medium]; checkpoint=gpt-5.6-luna","score":30.95238095238095,"warning":null},{"id":"panel-ed7429eba1da33d52494d3cd","modelId":"gpt-5.6-sol","benchmarkId":"swe-rebench-2026-06-tools","level":"medium","sourceLabel":"GPT-5.6 Sol [medium]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; GPT-5.6 Sol [medium]; checkpoint=gpt-5.6-sol","score":55.23809523809524,"warning":null},{"id":"panel-c9c71176cdf68d6328da3f8c","modelId":"grok-4.5","benchmarkId":"swe-rebench-2026-06-tools","level":"high","sourceLabel":"Grok 4.5 [high]","configuration":"SWE-rebench standardized ReAct tools; swe-rebench-2026-06-tools; Grok 4.5 [high]; checkpoint=grok-4.5","score":55.714285714285715,"warning":null},{"id":"epoch-run-2c4rvmf5N7QXTugrFLMZ83","modelId":"claude-opus-4-7","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"xhigh","sourceLabel":"claude-opus-4-7_xhigh","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-opus-4-7_xhigh; run=2c4rvmf5N7QXTugrFLMZ83; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":51.7,"warning":null},{"id":"epoch-run-2CdomqtBVz2it7a2Zx2G2t","modelId":"gpt-5.5","benchmarkId":"epoch-gpqa-diamond-1-0-8-choice","level":"low","sourceLabel":"gpt-5.5_low","configuration":"Epoch AI Inspect; task 1.0.8; scorer choice; model=gpt-5.5_low; run=2CdomqtBVz2it7a2Zx2G2t; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.65656565656566,"warning":null},{"id":"epoch-run-2f8Vn5QBdJnc9i5y9jzXS7","modelId":"qwen-3.8-max-0902","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"xhigh","sourceLabel":"qwen3.8-max-0902_xhigh","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.8-max-0902_xhigh; run=2f8Vn5QBdJnc9i5y9jzXS7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":100,"warning":null},{"id":"epoch-run-2fQJP3brRApfBdQq7Zy3bm","modelId":"llama-3.1-405b-instruct","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"Llama-3.1-405B-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=Llama-3.1-405B-Instruct; run=2fQJP3brRApfBdQq7Zy3bm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":9.722222222222223,"warning":null},{"id":"epoch-run-2gzNF2d53h2oCKDnojLtSP","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3-30b-a3b-thinking-2507","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-30b-a3b-thinking-2507; run=2gzNF2d53h2oCKDnojLtSP; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70.07575757575758,"warning":null},{"id":"epoch-run-2hbbp8z6PhB7c2T33os9q5","modelId":"claude-opus-4-6","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-4-6_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-6_max; run=2hbbp8z6PhB7c2T33os9q5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26.82926829268293,"warning":null},{"id":"epoch-run-2jKf4iEfVQnHkgAwEtD9Nn","modelId":"claude-fable-5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"claude-fable-5_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-fable-5_low; run=2jKf4iEfVQnHkgAwEtD9Nn; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":31,"warning":null},{"id":"epoch-run-2xfFf6KZAp2gUWYhHGkFgs","modelId":"deepseek-v4-pro","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"deepseek-v4-pro-0813_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=deepseek-v4-pro-0813_max; run=2xfFf6KZAp2gUWYhHGkFgs; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":52.90581162324649,"warning":null},{"id":"epoch-run-2y5jXmHGjyEyZvGjNDtRZe","modelId":"qwen3.6-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"qwen3.6-flash","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.6-flash; run=2y5jXmHGjyEyZvGjNDtRZe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12,"warning":null},{"id":"epoch-run-2z7kL9DpfvY44T9Wo99FYx","modelId":"gemini-2.5-pro","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"gemini-2.5-pro","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-2.5-pro; run=2z7kL9DpfvY44T9Wo99FYx; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":0,"warning":null},{"id":"epoch-run-37pJqPpXTE6ia9ikmmGPJ4","modelId":"claude-opus-4-8","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-5-model-extracted-exact-match","level":"max","sourceLabel":"claude-opus-4-8_max","configuration":"Epoch AI Inspect; task 1.1.5; scorer model_extracted_exact_match; model=claude-opus-4-8_max; run=37pJqPpXTE6ia9ikmmGPJ4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":98.33333333333333,"warning":null},{"id":"epoch-run-3CGChL5vNDM6Df5C8Z8owN","modelId":"gemma-3-1b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"gemma-3-1b-it","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemma-3-1b-it; run=3CGChL5vNDM6Df5C8Z8owN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":1.1111111111111112,"warning":null},{"id":"epoch-run-3h8STJpG2VzY9LiBUzwotv","modelId":"gpt-6-astra","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-6-astra_max; run=3h8STJpG2VzY9LiBUzwotv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.77020202020202,"warning":null},{"id":"epoch-run-3iRQkf5v7Wu7YuZhxa76Gg","modelId":"grok-4.3","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"grok-4.3_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.3_high; run=3iRQkf5v7Wu7YuZhxa76Gg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":14.634146341463413,"warning":null},{"id":"epoch-run-3L22tfAid5ZM42p92tqC73","modelId":"gemini-3.6-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"minimal","sourceLabel":"gemini-3.6-flash_minimal","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3.6-flash_minimal; run=3L22tfAid5ZM42p92tqC73; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25,"warning":null},{"id":"epoch-run-3n5KDcBaKqJh3WTg7qU2rs","modelId":"qwen3.6-35b-a3b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.6-35b-a3b","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.6-35b-a3b; run=3n5KDcBaKqJh3WTg7qU2rs; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.66666666666667,"warning":null},{"id":"epoch-run-3QtkFWzSLzd4hgYGsNd5cL","modelId":"qwen3.6-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.6-flash_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.6-flash_none; run=3QtkFWzSLzd4hgYGsNd5cL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-3UbYVQELDHAcQgu7sTTnRf","modelId":"magistral-small-2509","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"magistral-small-2509","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=magistral-small-2509; run=3UbYVQELDHAcQgu7sTTnRf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47.6010101010101,"warning":null},{"id":"epoch-run-3ZnPtGRJkvqHk8wvWWRgCe","modelId":"claude-opus-4-8","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"xhigh","sourceLabel":"claude-opus-4-8_xhigh","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-opus-4-8_xhigh; run=3ZnPtGRJkvqHk8wvWWRgCe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":31,"warning":null},{"id":"epoch-run-452pB5w5ea6uy92e2q2NMv","modelId":"qwen3.6-27b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.6-27b","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.6-27b; run=452pB5w5ea6uy92e2q2NMv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.11111111111111,"warning":null},{"id":"epoch-run-4a3M8rRWUoPphCSr2dguXH","modelId":"gemini-3.1-pro","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"low","sourceLabel":"gemini-3.1-pro-preview_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3.1-pro-preview_low; run=4a3M8rRWUoPphCSr2dguXH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.999999999999996,"warning":null},{"id":"epoch-run-4ACEC4x26HWRGiAepHGw43","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"Qwen3-235B-A22B-Thinking-2507","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=Qwen3-235B-A22B-Thinking-2507; run=4ACEC4x26HWRGiAepHGw43; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":40.44044044044044,"warning":null},{"id":"epoch-run-4afZ2hjLiJnPy5pD7TfE3K","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"claude-opus-4-5-20251101","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-5-20251101; run=4afZ2hjLiJnPy5pD7TfE3K; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":5,"warning":null},{"id":"epoch-run-4bLcFNu5zF6YuZf3qVKoHE","modelId":"gpt-5.6-terra","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"gpt-5.6-terra_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-terra_max; run=4bLcFNu5zF6YuZf3qVKoHE; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.3080808080808,"warning":null},{"id":"epoch-run-4E5xGsuRMwMBhAfC4F2G4N","modelId":"kimi-k3","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"kimi-k3_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=kimi-k3_high; run=4E5xGsuRMwMBhAfC4F2G4N; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.33333333333333,"warning":null},{"id":"epoch-run-4EayGByBV2xodNBkrZMcFb","modelId":"gpt-5.5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"gpt-5.5_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.5_low; run=4EayGByBV2xodNBkrZMcFb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":30,"warning":null},{"id":"epoch-run-4gZy2dAAT2EPZDGz5puD3H","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"minimal","sourceLabel":"gemini-3.1-flash-lite_minimal","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.1-flash-lite_minimal; run=4gZy2dAAT2EPZDGz5puD3H; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21.403508771929825,"warning":null},{"id":"epoch-run-4hmmfwG9tgZJvKhL78nKzL","modelId":"glm-5.1","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"glm-5.1_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.1_none; run=4hmmfwG9tgZJvKhL78nKzL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":24.912280701754387,"warning":null},{"id":"epoch-run-4HVvM9ZWKKPLQPcC8WYYMT","modelId":"deepseek-v4-pro","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"deepseek-v4-pro-0813_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=deepseek-v4-pro-0813_max; run=4HVvM9ZWKKPLQPcC8WYYMT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":98.61111111111111,"warning":null},{"id":"epoch-run-4LWNorP8Fy8V8doxirNFcH","modelId":"qwen3.6-plus","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"qwen3.6-plus","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=qwen3.6-plus; run=4LWNorP8Fy8V8doxirNFcH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":44.14414414414414,"warning":null},{"id":"epoch-run-4oSbA2RKFsfuAfVayAMAww","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_48K (thinking 48k)","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-sonnet-4-5-20250929_48K; run=4oSbA2RKFsfuAfVayAMAww; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17,"warning":null},{"id":"epoch-run-4S3ZrJDkwh4gmGrmJQdvzR","modelId":"glm-5.3","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"glm-5.3_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.3_max; run=4S3ZrJDkwh4gmGrmJQdvzR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":29.268292682926827,"warning":null},{"id":"epoch-run-4S4GXbkRhLz8wWP97jYG57","modelId":"deepseek-v4-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"deepseek-v4-flash-0731_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=deepseek-v4-flash-0731_max; run=4S4GXbkRhLz8wWP97jYG57; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34,"warning":null},{"id":"epoch-run-4uDz9b6SosxDSZf7swMD9D","modelId":"gpt-6-luna","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-6-luna_max; run=4uDz9b6SosxDSZf7swMD9D; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":98.88888888888889,"warning":null},{"id":"epoch-run-4uxDyWmq5f5azVYfWBGUHh","modelId":"gpt-5.6-sol","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"none","sourceLabel":"gpt-5.6-sol_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-sol_none; run=4uxDyWmq5f5azVYfWBGUHh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33,"warning":null},{"id":"epoch-run-58eQmyCfPa3FufhXJFvAL2","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-gpqa-diamond-1-0-6-choice","level":"thinking","sourceLabel":"claude-sonnet-4-6_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.6; scorer choice; model=claude-sonnet-4-6_32K; run=58eQmyCfPa3FufhXJFvAL2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.37373737373737,"warning":null},{"id":"epoch-run-58MQQiJ7vvgvP9e8LcG6eV","modelId":"gpt-oss-120b","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"medium","sourceLabel":"gpt-oss-120b_medium","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-oss-120b_medium; run=58MQQiJ7vvgvP9e8LcG6eV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":2,"warning":null},{"id":"epoch-run-5AijYw5bFJtSuahnbzNo6Z","modelId":"kimi-k2.7-code","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"kimi-k2.7-code","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=kimi-k2.7-code; run=5AijYw5bFJtSuahnbzNo6Z; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36.5,"warning":null},{"id":"epoch-run-5e4Jp9dxtNisWx2wbRRZxM","modelId":"qwen3.7-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"qwen3.7-flash","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.7-flash; run=5e4Jp9dxtNisWx2wbRRZxM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":15,"warning":null},{"id":"epoch-run-5FGXKvXQwGPtcCX2PNrLR5","modelId":"gpt-5.5","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"gpt-5.5_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.5_xhigh; run=5FGXKvXQwGPtcCX2PNrLR5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.26315789473684,"warning":null},{"id":"epoch-run-5FVmQUSS4uRd23f6nNgeRe","modelId":"qwen3.5-9b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.5-9B","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-9B; run=5FVmQUSS4uRd23f6nNgeRe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":42.77777777777778,"warning":null},{"id":"epoch-run-5kMRXeasRvtghXpDotCdgV","modelId":"gpt-5.6-sol","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-sol_max; run=5kMRXeasRvtghXpDotCdgV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.99999999999999,"warning":null},{"id":"epoch-run-5mmrBSQb7hVSJ9VYtNm46i","modelId":"phi-4","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"phi-4","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=phi-4; run=5mmrBSQb7hVSJ9VYtNm46i; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":13.750000000000002,"warning":null},{"id":"epoch-run-5Q7zWYmoRik8mhABL4wpVR","modelId":"gpt-5.6-luna","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"none","sourceLabel":"gpt-5.6-luna_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-luna_none; run=5Q7zWYmoRik8mhABL4wpVR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":3,"warning":null},{"id":"epoch-run-5sNRWemuVwb4adBaSwrLFf","modelId":"gpt-6-astra","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"medium","sourceLabel":"gpt-6-astra_medium","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-astra_medium; run=5sNRWemuVwb4adBaSwrLFf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.5609756097561,"warning":null},{"id":"epoch-run-5u4XrDnQSRnuXSK6oXKAth","modelId":"gpt-5.5-pro","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"gpt-5.5-pro_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.5-pro_xhigh; run=5u4XrDnQSRnuXSK6oXKAth; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":62,"warning":null},{"id":"epoch-run-5yyuCCi2Dpfh6xLBMxqFyB","modelId":"gemini-3.8-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"high","sourceLabel":"gemini-3.8-flash_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gemini-3.8-flash_high; run=5yyuCCi2Dpfh6xLBMxqFyB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47,"warning":null},{"id":"epoch-run-64rokFyL2L8Vd4SZVdayCy","modelId":"claude-opus-4-6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"claude-opus-4-6_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-opus-4-6_max; run=64rokFyL2L8Vd4SZVdayCy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25,"warning":null},{"id":"epoch-run-68KxyCJbfMSq8JBL3LRJfr","modelId":"gemini-3.5-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"gemini-3.5-flash_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.5-flash_low; run=68KxyCJbfMSq8JBL3LRJfr; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":42,"warning":null},{"id":"epoch-run-6cmgkNAQZekBr6e6rVdLe9","modelId":"gpt-5.6-sol","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gpt-5.6-sol_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-sol_low; run=6cmgkNAQZekBr6e6rVdLe9; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":89.8989898989899,"warning":null},{"id":"epoch-run-6E989Xxeuw925T7cCV3XGB","modelId":"claude-fable-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"claude-fable-5_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-fable-5_low; run=6E989Xxeuw925T7cCV3XGB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.77777777777777,"warning":null},{"id":"epoch-run-6gYpimjqJm3CL6mvKEN9qs","modelId":"glm-5.2","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"glm-5.2_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=glm-5.2_low; run=6gYpimjqJm3CL6mvKEN9qs; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-6iEkMa88YjLDgcde4bLm8B","modelId":"gemini-3.6-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gemini-3.6-flash_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.6-flash_low; run=6iEkMa88YjLDgcde4bLm8B; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.22222222222221,"warning":null},{"id":"epoch-run-6P9AJQmBzaPmE7ossbqLjz","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-sonnet-4-6_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-4-6_max; run=6P9AJQmBzaPmE7ossbqLjz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":23,"warning":null},{"id":"epoch-run-6phXzd4aCbnBUSQmBsaywa","modelId":"glm-4.7","benchmarkId":"epoch-gpqa-diamond-1-0-4-choice","level":"unspecified","sourceLabel":"glm-4.7","configuration":"Epoch AI Inspect; task 1.0.4; scorer choice; model=glm-4.7; run=6phXzd4aCbnBUSQmBsaywa; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.33333333333334,"warning":null},{"id":"epoch-run-6pTYmQCw3EMbErvAqpivH5","modelId":"qwen3.7-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.7-flash","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.7-flash; run=6pTYmQCw3EMbErvAqpivH5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19.298245614035086,"warning":null},{"id":"epoch-run-6rZFxNR3GjfhWey6DzEiYz","modelId":"qwen3-30b-a3b-thinking-2507","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3-30b-a3b-thinking-2507","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-30b-a3b-thinking-2507; run=6rZFxNR3GjfhWey6DzEiYz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70.27777777777777,"warning":null},{"id":"epoch-run-6XZAenohp9pCgQ3gVeUAi8","modelId":"qwen3.5-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.5-flash","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.5-flash; run=6XZAenohp9pCgQ3gVeUAi8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":9.473684210526317,"warning":null},{"id":"epoch-run-6YAzFq89fwWiJ3uJdptCMr","modelId":"llama-3.1-8b-instruct","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"Llama-3.1-8B-Instruct","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=Llama-3.1-8B-Instruct; run=6YAzFq89fwWiJ3uJdptCMr; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":1.6666666666666667,"warning":null},{"id":"epoch-run-72DQVwbYiQW5pjhshjBaGU","modelId":"gemini-3.5-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.5-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.5-flash_high; run=72DQVwbYiQW5pjhshjBaGU; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":62.807017543859644,"warning":null},{"id":"epoch-run-79ZP6iwoBaF5t9a5VRqPnw","modelId":"minimax-m3","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"MiniMax-M3","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=MiniMax-M3; run=79ZP6iwoBaF5t9a5VRqPnw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.11111111111111,"warning":null},{"id":"epoch-run-7aAKcnsCXoSumfAu8mr2un","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"minimal","sourceLabel":"gemini-3.5-flash-lite_minimal","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.5-flash-lite_minimal; run=7aAKcnsCXoSumfAu8mr2un; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":74.24242424242425,"warning":null},{"id":"epoch-run-7cc6cDCXjsPjtKt36bbeMR","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"gemini-3-flash-preview_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gemini-3-flash-preview_high; run=7cc6cDCXjsPjtKt36bbeMR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.8,"warning":null},{"id":"epoch-run-7crLvnfQc987dC6mr3iGbf","modelId":"grok-4.6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"xhigh","sourceLabel":"grok-4.6_xhigh","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=grok-4.6_xhigh; run=7crLvnfQc987dC6mr3iGbf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":99.16666666666667,"warning":null},{"id":"epoch-run-7DQNxGG5q6i7VtnzF7vNxA","modelId":"glm-5.2","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"low","sourceLabel":"glm-5.2_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=glm-5.2_low; run=7DQNxGG5q6i7VtnzF7vNxA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19,"warning":null},{"id":"epoch-run-7F3QFMAkwnCV62HcvxP96q","modelId":"gpt-6-luna","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-6-luna_max; run=7F3QFMAkwnCV62HcvxP96q; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.46717171717171,"warning":null},{"id":"epoch-run-7GiYWM7UagUsHpNsdymd4z","modelId":"qwen3.6-35b-a3b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.6-35b-a3b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.6-35b-a3b_none; run=7GiYWM7UagUsHpNsdymd4z; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":68.88888888888889,"warning":null},{"id":"epoch-run-7HWW3sjRstKzb9JpcnvLaV","modelId":"grok-4.5","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"grok-4.5_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=grok-4.5_high; run=7HWW3sjRstKzb9JpcnvLaV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":48.3,"warning":null},{"id":"epoch-run-7n3HBUK684GmdT6Tkycu6m","modelId":"gpt-6-luna","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-luna_max; run=7n3HBUK684GmdT6Tkycu6m; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.09756097560976,"warning":null},{"id":"epoch-run-7oUkeLfwMK3MtkSNHX6QPu","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-sonnet-4-5-20250929_32K; run=7oUkeLfwMK3MtkSNHX6QPu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":23.859649122807017,"warning":null},{"id":"epoch-run-7QkPRbRPpRx8nJT2ZcqSx6","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-extracted-exact-match","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_16K (thinking 16k)","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_extracted_exact_match; model=claude-sonnet-4-5-20250929_16K; run=7QkPRbRPpRx8nJT2ZcqSx6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.11111111111111,"warning":null},{"id":"epoch-run-7RjsANKwwsePSstcZeSMhX","modelId":"gpt-5.6-sol","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-sol_max; run=7RjsANKwwsePSstcZeSMhX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":100,"warning":null},{"id":"epoch-run-7vThbrtbczScg2UcsAYXtF","modelId":"grok-4.5","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"grok-4.5_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.5_high; run=7vThbrtbczScg2UcsAYXtF; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":24.390243902439025,"warning":null},{"id":"epoch-run-84ARuGyhQsjgh8umRmBvas","modelId":"gemini-3.7-flash","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"gemini-3.7-flash_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gemini-3.7-flash_high; run=84ARuGyhQsjgh8umRmBvas; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":69.19999999999999,"warning":null},{"id":"epoch-run-85fhMxLaebWtLoi6oLpTbi","modelId":"claude-opus-4-6","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"claude-opus-4-6_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-opus-4-6_max; run=85fhMxLaebWtLoi6oLpTbi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47,"warning":null},{"id":"epoch-run-86SY5JbPB49xXhHFQW6tHu","modelId":"gemini-3.6-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gemini-3.6-flash_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.6-flash_high; run=86SY5JbPB49xXhHFQW6tHu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.12878787878788,"warning":null},{"id":"epoch-run-89zcekqEYnE8zhwG4pR5Nu","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Llama-70B","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=DeepSeek-R1-Distill-Llama-70B; run=89zcekqEYnE8zhwG4pR5Nu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":51.388888888888886,"warning":null},{"id":"epoch-run-8grv57UmR89LXkEYhyssr9","modelId":"gpt-5.6-terra","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"none","sourceLabel":"gpt-5.6-terra_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-terra_none; run=8grv57UmR89LXkEYhyssr9; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":14.000000000000002,"warning":null},{"id":"epoch-run-8iiRQiAoCepDELQfUv3VDq","modelId":"kimi-k2.6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"kimi-k2.6_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=kimi-k2.6_none; run=8iiRQiAoCepDELQfUv3VDq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12,"warning":null},{"id":"epoch-run-8o2J9uQLKDCTiCFwYdkJRK","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.5-flash-lite_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.5-flash-lite_high; run=8o2J9uQLKDCTiCFwYdkJRK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25.964912280701753,"warning":null},{"id":"epoch-run-8qwT2CF9U5zV99PFirnVsp","modelId":"llama-3.2-1b-instruct","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"Llama-3.2-1B-Instruct","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=Llama-3.2-1B-Instruct; run=8qwT2CF9U5zV99PFirnVsp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":0.5555555555555556,"warning":null},{"id":"epoch-run-8UqaLobhmNJLuPGTTpeaCw","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Qwen-14B","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=DeepSeek-R1-Distill-Qwen-14B; run=8UqaLobhmNJLuPGTTpeaCw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":44.696969696969695,"warning":null},{"id":"epoch-run-97t77hRrNae6Va8fBqBhPy","modelId":"minimax-m3","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"MiniMax-M3_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=MiniMax-M3_none; run=97t77hRrNae6Va8fBqBhPy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26.666666666666668,"warning":null},{"id":"epoch-run-99ys2jgQiPQ5UwrqnazfqA","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gemini-3.1-flash-lite_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.1-flash-lite_low; run=99ys2jgQiPQ5UwrqnazfqA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":74.24242424242425,"warning":null},{"id":"epoch-run-9cXtChGrowgGDwa9AcAecB","modelId":"grok-4.6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"xhigh","sourceLabel":"grok-4.6_xhigh","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=grok-4.6_xhigh; run=9cXtChGrowgGDwa9AcAecB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34,"warning":null},{"id":"epoch-run-9ekki5tJhyjmA7ZMLoUxzo","modelId":"glm-4.7-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"glm-4.7-flash","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-4.7-flash; run=9ekki5tJhyjmA7ZMLoUxzo; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":58.333333333333336,"warning":null},{"id":"epoch-run-9kErpmgQ42wTvgJaFn478v","modelId":"qwen3.6-35b-a3b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3.6-35b-a3b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.6-35b-a3b_none; run=9kErpmgQ42wTvgJaFn478v; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.84848484848484,"warning":null},{"id":"epoch-run-9kSgmUd9vh4CBXz3tJhctN","modelId":"grok-4.5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"grok-4.5_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=grok-4.5_high; run=9kSgmUd9vh4CBXz3tJhctN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.43434343434343,"warning":null},{"id":"epoch-run-9LWdzVN5w4ihNvncsaCWMu","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-5-model-extracted-exact-match","level":"thinking","sourceLabel":"claude-sonnet-4-6_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.5; scorer model_extracted_exact_match; model=claude-sonnet-4-6_32K; run=9LWdzVN5w4ihNvncsaCWMu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.7936507936508,"warning":null},{"id":"epoch-run-9rDLegdh3zmBfFzTj3fkCc","modelId":"glm-4.7","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"glm-4.7","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=glm-4.7; run=9rDLegdh3zmBfFzTj3fkCc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21,"warning":null},{"id":"epoch-run-9SSHEP95FYVZvHDdoHSRy3","modelId":"llama-3.2-90b-vision-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"Llama-3.2-90B-Vision-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=Llama-3.2-90B-Vision-Instruct; run=9SSHEP95FYVZvHDdoHSRy3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":41.035353535353536,"warning":null},{"id":"epoch-run-9uKmMPn6eiRPsfcLKod7M7","modelId":"glm-5.2","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"glm-5.2_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-5.2_low; run=9uKmMPn6eiRPsfcLKod7M7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":75.55555555555556,"warning":null},{"id":"epoch-run-9WS6kHXT8XFtJWtE2wWLAo","modelId":"qwen3.5-35b-a3b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.5-35b-a3b","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-35b-a3b; run=9WS6kHXT8XFtJWtE2wWLAo; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":54.44444444444444,"warning":null},{"id":"epoch-run-9YJf2nviwjM6WVhw8qeJta","modelId":"grok-4.6","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"xhigh","sourceLabel":"grok-4.6_xhigh","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=grok-4.6_xhigh; run=9YJf2nviwjM6WVhw8qeJta; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.18181818181817,"warning":null},{"id":"epoch-run-9ZFhnyMNCKZ8XdR58WGD3D","modelId":"claude-fable-5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"claude-fable-5_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-fable-5_high; run=9ZFhnyMNCKZ8XdR58WGD3D; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45,"warning":null},{"id":"epoch-run-9zx2EXPTvPmxsL9qFCjP3x","modelId":"qwen3.6-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.6-flash","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.6-flash; run=9zx2EXPTvPmxsL9qFCjP3x; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.44444444444444,"warning":null},{"id":"epoch-run-A33om9MTYMC7xYeLv9X7Ha","modelId":"kimi-k3","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"kimi-k3_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=kimi-k3_max; run=A33om9MTYMC7xYeLv9X7Ha; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.22222222222221,"warning":null},{"id":"epoch-run-a4uE6KgMhGSUjcnemRnbkA","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3-flash-preview_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3-flash-preview_high; run=a4uE6KgMhGSUjcnemRnbkA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.55555555555556,"warning":null},{"id":"epoch-run-a4uV8HMMBAcHXLVSmZfBgX","modelId":"deepseek-v4-pro","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"deepseek-v4-pro-0813_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=deepseek-v4-pro-0813_max; run=a4uV8HMMBAcHXLVSmZfBgX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26.82926829268293,"warning":null},{"id":"epoch-run-A4yUA9PfzRr92JAqDDo9ro","modelId":"qwen3.5-397b-a17b","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.5-397b-a17b_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.5-397b-a17b_none; run=A4yUA9PfzRr92JAqDDo9ro; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":31.2280701754386,"warning":null},{"id":"epoch-run-a5qR5FNnftfR8pk9HJ3JUz","modelId":"qwen3.7-plus","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.7-plus_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.7-plus_none; run=a5qR5FNnftfR8pk9HJ3JUz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17,"warning":null},{"id":"epoch-run-ABT89ieSWGnGj4RYMduHiG","modelId":"qwen-3.8-max","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"qwen3.8-max_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.8-max_xhigh; run=ABT89ieSWGnGj4RYMduHiG; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":74.73684210526315,"warning":null},{"id":"epoch-run-AczFCvkWR5SmdbV7eXruqj","modelId":"qwen3.5-9b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3.5-9B_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-9B_none; run=AczFCvkWR5SmdbV7eXruqj; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":78.91414141414141,"warning":null},{"id":"epoch-run-adGE8rXdNuUFs96ioppbc5","modelId":"claude-opus-4-8","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"claude-opus-4-8_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-opus-4-8_none; run=adGE8rXdNuUFs96ioppbc5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.44444444444444,"warning":null},{"id":"epoch-run-aekViWbdqVVeXpEG5hrSbQ","modelId":"glm-5.2","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"minimal","sourceLabel":"glm-5.2_minimal","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=glm-5.2_minimal; run=aekViWbdqVVeXpEG5hrSbQ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-AfKJ2DGMhkjZmeFZxGoY3w","modelId":"gpt-6-astra","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"gpt-6-astra_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-astra_xhigh; run=AfKJ2DGMhkjZmeFZxGoY3w; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.6,"warning":null},{"id":"epoch-run-afxoicJc4C6caVDwUpsiME","modelId":"qwen3.5-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.5-flash_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.5-flash_none; run=afxoicJc4C6caVDwUpsiME; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18.24561403508772,"warning":null},{"id":"epoch-run-agebmxGseTvmvN4AHN8i4m","modelId":"qwen3-235b-a22b","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"qwen3-235b-a22b","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=qwen3-235b-a22b; run=agebmxGseTvmvN4AHN8i4m; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70.70707070707071,"warning":null},{"id":"epoch-run-aGRKHaq9uSawfYVpC565zU","modelId":"qwen-3.8-max","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"xhigh","sourceLabel":"qwen3.8-max_xhigh","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.8-max_xhigh; run=aGRKHaq9uSawfYVpC565zU; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":99.44444444444444,"warning":null},{"id":"epoch-run-AhtQ5ggs4pTfLDqHFjueGy","modelId":"gpt-5.6-luna","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"gpt-5.6-luna_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-luna_max; run=AhtQ5ggs4pTfLDqHFjueGy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":98.33333333333333,"warning":null},{"id":"epoch-run-aJD7ZfFftu2cZoToJhhzF7","modelId":"gpt-6-sol","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-6-sol_max; run=aJD7ZfFftu2cZoToJhhzF7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.31818181818183,"warning":null},{"id":"epoch-run-ajQFZGMjvyMTH2ULDZhwtz","modelId":"gemini-3.6-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3.6-flash_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.6-flash_high; run=ajQFZGMjvyMTH2ULDZhwtz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.16666666666667,"warning":null},{"id":"epoch-run-aKc2rzsA6YPYwCkE7nhvvB","modelId":"claude-opus-4-6","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-opus-4-6_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-6_max; run=aKc2rzsA6YPYwCkE7nhvvB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21,"warning":null},{"id":"epoch-run-aLrHd8RCiNo3Yb4GqSawM7","modelId":"claude-fable-5-1","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-fable-5-1_max; run=aLrHd8RCiNo3Yb4GqSawM7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":48,"warning":null},{"id":"epoch-run-AoAD5uVtgVDnXmwA7STuno","modelId":"kimi-k3","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"kimi-k3_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=kimi-k3_high; run=AoAD5uVtgVDnXmwA7STuno; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.91919191919192,"warning":null},{"id":"epoch-run-aPmR8fV49N98a6NVKpASAf","modelId":"gemini-3.6-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"minimal","sourceLabel":"gemini-3.6-flash_minimal","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.6-flash_minimal; run=aPmR8fV49N98a6NVKpASAf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.85858585858585,"warning":null},{"id":"epoch-run-ASLdzByzxxMxFeLc7dwTPi","modelId":"qwen3-8b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3-8b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-8b_none; run=ASLdzByzxxMxFeLc7dwTPi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":22.22222222222222,"warning":null},{"id":"epoch-run-Atc4F5iuMZu3vNBWVGCBnv","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"claude-sonnet-4-6_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-4-6_high; run=Atc4F5iuMZu3vNBWVGCBnv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19,"warning":null},{"id":"epoch-run-awDKyKbPXRzBVZzfdQucef","modelId":"qwen3.5-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.5-flash","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-flash; run=awDKyKbPXRzBVZzfdQucef; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.44444444444444,"warning":null},{"id":"epoch-run-axMdgNDC8UYYLnZYLHZpF5","modelId":"gemini-3.7-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.7-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.7-flash_high; run=axMdgNDC8UYYLnZYLHZpF5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.57894736842105,"warning":null},{"id":"epoch-run-aYau6EG7CCaWVSyVq3vvQc","modelId":"glm-4.7","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"glm-4.7","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=glm-4.7; run=aYau6EG7CCaWVSyVq3vvQc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32.2,"warning":null},{"id":"epoch-run-AZHoMLTNQEcKLwtv7QzBLZ","modelId":"claude-sonnet-5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"unspecified","sourceLabel":"claude-sonnet-5","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=claude-sonnet-5; run=AZHoMLTNQEcKLwtv7QzBLZ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":16,"warning":null},{"id":"epoch-run-azk4theXxAsizmoi2HDD49","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-5-20251101_32K; run=azk4theXxAsizmoi2HDD49; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":4.878048780487805,"warning":null},{"id":"epoch-run-Azz52eFkXwVJiRCGpaUBwB","modelId":"deepseek-v4-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"deepseek-v4-flash-0731_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=deepseek-v4-flash-0731_max; run=Azz52eFkXwVJiRCGpaUBwB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32,"warning":null},{"id":"epoch-run-B3Kno3buYWRYbZxU33STiX","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gemini-3.5-flash-lite_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.5-flash-lite_low; run=B3Kno3buYWRYbZxU33STiX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":60,"warning":null},{"id":"epoch-run-b6yWmeFgsWeEiG9Qzs3NSf","modelId":"deepseek-v4-pro","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"deepseek-v4-pro-0813_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=deepseek-v4-pro-0813_max; run=b6yWmeFgsWeEiG9Qzs3NSf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":64.56140350877193,"warning":null},{"id":"epoch-run-b7Lw9ExxAhVY3hdExoh4MP","modelId":"glm-5.1","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"glm-5.1","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-5.1; run=b7Lw9ExxAhVY3hdExoh4MP; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.33333333333333,"warning":null},{"id":"epoch-run-B8MaPUe3EYwUZASsNqE7C9","modelId":"kimi-k3","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"kimi-k3_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=kimi-k3_max; run=B8MaPUe3EYwUZASsNqE7C9; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.11868686868688,"warning":null},{"id":"epoch-run-b9csPZ2c9CAsRXmzzfVoEW","modelId":"qwen3.5-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"qwen3.5-flash","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.5-flash; run=b9csPZ2c9CAsRXmzzfVoEW; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20,"warning":null},{"id":"epoch-run-bA5j9K2shzZVFGTMQ5g6EU","modelId":"gpt-5.6-terra","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"gpt-5.6-terra_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-terra_low; run=bA5j9K2shzZVFGTMQ5g6EU; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27,"warning":null},{"id":"epoch-run-bAERV8wdmoHMLz2K4bZCSK","modelId":"gpt-6-luna","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-6-luna_max; run=bAERV8wdmoHMLz2K4bZCSK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.000000000000001,"warning":null},{"id":"epoch-run-BAi4jidcT3F6reBkSXDJKv","modelId":"qwen3-32b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"Qwen3-32B","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=Qwen3-32B; run=BAi4jidcT3F6reBkSXDJKv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.94444444444444,"warning":null},{"id":"epoch-run-BbM4xbJcS5sjrzZABGg8K7","modelId":"claude-opus-4-6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"unspecified","sourceLabel":"claude-opus-4-6","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=claude-opus-4-6; run=BbM4xbJcS5sjrzZABGg8K7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":15,"warning":null},{"id":"epoch-run-bbqokudairnKtbyjBjuWCx","modelId":"qwen3-1.7b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3-1.7b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-1.7b_none; run=bbqokudairnKtbyjBjuWCx; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":8.055555555555555,"warning":null},{"id":"epoch-run-bBV5wCrrZtihrgxEsg9jPD","modelId":"deepseek-r1-0528","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"DeepSeek-R1-0528","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=DeepSeek-R1-0528; run=bBV5wCrrZtihrgxEsg9jPD; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.38888888888889,"warning":null},{"id":"epoch-run-bCdwpRXgZtvQLYC4wA9JXr","modelId":"muse-spark-1.1","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"muse-spark-1.1","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=muse-spark-1.1; run=bCdwpRXgZtvQLYC4wA9JXr; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.78894472361809,"warning":null},{"id":"epoch-run-bcntamvGsSH6QgETdoaSph","modelId":"kimi-k3","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"kimi-k3_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=kimi-k3_max; run=bcntamvGsSH6QgETdoaSph; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":72.1830985915493,"warning":null},{"id":"epoch-run-bDKKx7zTdW5WwMCfwqsLi4","modelId":"gpt-5.6-luna","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"none","sourceLabel":"gpt-5.6-luna_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-luna_none; run=bDKKx7zTdW5WwMCfwqsLi4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":63.63636363636363,"warning":null},{"id":"epoch-run-bDS2AGEf3AywbVcdDyAKZK","modelId":"phi-3-medium-128k-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"Phi-3-medium-128k-instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=Phi-3-medium-128k-instruct; run=bDS2AGEf3AywbVcdDyAKZK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27.58838383838384,"warning":null},{"id":"epoch-run-bdsZjrG4ayzP8NAJg5q586","modelId":"grok-4.3","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"grok-4.3_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.3_high; run=bdsZjrG4ayzP8NAJg5q586; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":42.80701754385965,"warning":null},{"id":"epoch-run-beRGgJXTJggTi8PPEZNoZ5","modelId":"mistral-large-2411","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"mistral-large-2411","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=mistral-large-2411; run=beRGgJXTJggTi8PPEZNoZ5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.777777777777778,"warning":null},{"id":"epoch-run-beSbPHiiWcwJBSzYRh7Kh4","modelId":"gemini-3.6-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"minimal","sourceLabel":"gemini-3.6-flash_minimal","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.6-flash_minimal; run=beSbPHiiWcwJBSzYRh7Kh4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.999999999999996,"warning":null},{"id":"epoch-run-bfftd5EzUK9yAekafEofvK","modelId":"qwen-3.8-max","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"xhigh","sourceLabel":"qwen3.8-max_xhigh","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=qwen3.8-max_xhigh; run=bfftd5EzUK9yAekafEofvK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45.800000000000004,"warning":null},{"id":"epoch-run-bg3JRdd33iWfvRoGSvyXXE","modelId":"grok-4.6","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"grok-4.6_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.6_xhigh; run=bg3JRdd33iWfvRoGSvyXXE; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":31.70731707317073,"warning":null},{"id":"epoch-run-BgmrBFsbsaD8rNpebv49Da","modelId":"gemini-3.1-pro","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gemini-3.1-pro-preview_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.1-pro-preview_high; run=BgmrBFsbsaD8rNpebv49Da; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.44444444444444,"warning":null},{"id":"epoch-run-bJV4D8SEo9zMJPEYTTXJNZ","modelId":"ministral-3b-2410","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"ministral-3b-2410","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=ministral-3b-2410; run=bJV4D8SEo9zMJPEYTTXJNZ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25.252525252525253,"warning":null},{"id":"epoch-run-BLgwhLGGofDBzxoc34YFHK","modelId":"claude-haiku-4-5-20251001","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"thinking","sourceLabel":"claude-haiku-4-5-20251001_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-haiku-4-5-20251001_32K; run=BLgwhLGGofDBzxoc34YFHK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":4,"warning":null},{"id":"epoch-run-BLjXudxuEyLbZiKBSZz326","modelId":"glm-5.2","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"glm-5.2_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=glm-5.2_none; run=BLjXudxuEyLbZiKBSZz326; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-BnX67AoZzgVbHgaM8F8Brz","modelId":"qwen3-30b-a3b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"Qwen3-30B-A3B","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=Qwen3-30B-A3B; run=BnX67AoZzgVbHgaM8F8Brz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":61.67929292929293,"warning":null},{"id":"epoch-run-BP3KYD8rYyFg3LviNGtjLL","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"Qwen3-235B-A22B-Thinking-2507","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=Qwen3-235B-A22B-Thinking-2507; run=BP3KYD8rYyFg3LviNGtjLL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":9,"warning":null},{"id":"epoch-run-bpZy8tAbh7UV6A6U7XZtBg","modelId":"nemotron-3-ultra","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"nemotron-3-ultra","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=nemotron-3-ultra; run=bpZy8tAbh7UV6A6U7XZtBg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.35353535353535,"warning":null},{"id":"epoch-run-bq5y6Tp83KLtxC58UDfpyN","modelId":"qwen-3.8-max","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"qwen3.8-max_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=qwen3.8-max_xhigh; run=bq5y6Tp83KLtxC58UDfpyN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":46,"warning":null},{"id":"epoch-run-bqiVinHXFcgyFGPstJwXJN","modelId":"claude-sonnet-5","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-sonnet-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-sonnet-5_max; run=bqiVinHXFcgyFGPstJwXJN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":29.268292682926827,"warning":null},{"id":"epoch-run-Br3mjs2GrYcCwX9AMwcihU","modelId":"gemini-3.8-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gemini-3.8-flash_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.8-flash_high; run=Br3mjs2GrYcCwX9AMwcihU; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.39141414141415,"warning":null},{"id":"epoch-run-bSyTSKD8RiRNZCGNu6MnFp","modelId":"kimi-k2.7-code","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"kimi-k2.7-code","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=kimi-k2.7-code; run=bSyTSKD8RiRNZCGNu6MnFp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":14.000000000000002,"warning":null},{"id":"epoch-run-BTeRdhgt6TFBj6QkGX2j6P","modelId":"kimi-k2.6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-3-model-extracted-exact-match","level":"unspecified","sourceLabel":"kimi-k2.6","configuration":"Epoch AI Inspect; task 1.1.3; scorer model_extracted_exact_match; model=kimi-k2.6; run=BTeRdhgt6TFBj6QkGX2j6P; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":96.11111111111111,"warning":null},{"id":"epoch-run-buLf8Ne29xmzMZBRQkvrQG","modelId":"qwen3.6-27b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.6-27b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.6-27b; run=buLf8Ne29xmzMZBRQkvrQG; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.85858585858585,"warning":null},{"id":"epoch-run-BVat526nF6NUGoVaLCp7DT","modelId":"seed-oss-36b-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"seed-oss-36b-instruct","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=seed-oss-36b-instruct; run=BVat526nF6NUGoVaLCp7DT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.52777777777779,"warning":null},{"id":"epoch-run-BvShPsULJMkwxxf8vBSNfu","modelId":"gpt-5.6-luna","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"medium","sourceLabel":"gpt-5.6-luna_medium","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-luna_medium; run=BvShPsULJMkwxxf8vBSNfu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12,"warning":null},{"id":"epoch-run-bWnggCv6to2QKSjngrtuos","modelId":"claude-opus-5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"unspecified","sourceLabel":"claude-opus-5","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=claude-opus-5; run=bWnggCv6to2QKSjngrtuos; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":37,"warning":null},{"id":"epoch-run-BWz8uig4QyqPtke59pM88z","modelId":"qwen3.7-plus","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3.7-plus_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.7-plus_none; run=BWz8uig4QyqPtke59pM88z; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":81.81818181818183,"warning":null},{"id":"epoch-run-bXgZZgzRE3wXoSRUgrbsVf","modelId":"gemini-3.1-pro","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-5-model-extracted-exact-match","level":"unspecified","sourceLabel":"gemini-3.1-pro-preview","configuration":"Epoch AI Inspect; task 1.0.5; scorer model_extracted_exact_match; model=gemini-3.1-pro-preview; run=bXgZZgzRE3wXoSRUgrbsVf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.6,"warning":"Run-specific qualification requires review: In the initial run, no epochs completed for 2/45 samples. Despite several re-runs of these two samples, no epochs completed. All epochs for the other 43/45 samples were correct. The model's score was edited to be 43/45."},{"id":"epoch-run-bXrmQqzF9sXbuZfRrSc7NP","modelId":"qwen3.5-397b-a17b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3.5-397b-a17b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-397b-a17b_none; run=bXrmQqzF9sXbuZfRrSc7NP; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.36363636363636,"warning":null},{"id":"epoch-run-BxrNfyMffkVUq4n7U5eK3Q","modelId":"kimi-k2.6","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"kimi-k2.6","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=kimi-k2.6; run=BxrNfyMffkVUq4n7U5eK3Q; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.19298245614035,"warning":null},{"id":"epoch-run-bxS76CJuVws8iwWU2Wr2PM","modelId":"claude-fable-5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-fable-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-fable-5_max; run=bxS76CJuVws8iwWU2Wr2PM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47,"warning":null},{"id":"epoch-run-c2KVkMCWnpo6BbnMWgdaDH","modelId":"deepseek-v4-pro","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"deepseek-v4-pro-0813_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=deepseek-v4-pro-0813_max; run=c2KVkMCWnpo6BbnMWgdaDH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":43,"warning":null},{"id":"epoch-run-c2YLifDU4zYCX5Jbi9HLL7","modelId":"qwen-3.8-max","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"xhigh","sourceLabel":"qwen3.8-max_xhigh","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.8-max_xhigh; run=c2YLifDU4zYCX5Jbi9HLL7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":38,"warning":null},{"id":"epoch-run-C56VHYLMeww2wXebeJh3gg","modelId":"deepseek-v4-pro","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"deepseek-v4-pro-0813_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=deepseek-v4-pro-0813_max; run=C56VHYLMeww2wXebeJh3gg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":46,"warning":null},{"id":"epoch-run-c5kpB4axzmygRywWqBKFmv","modelId":"glm-5.2","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"glm-5.2_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-5.2_max; run=c5kpB4axzmygRywWqBKFmv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.85606060606061,"warning":null},{"id":"epoch-run-CA5LW4phRnHSsZUGXEm8H6","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-opus-4-5-20251101_32K; run=CA5LW4phRnHSsZUGXEm8H6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45.7,"warning":null},{"id":"epoch-run-CAoLSNkdR27AmHb84fnr8B","modelId":"kimi-k3","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"low","sourceLabel":"kimi-k3_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=kimi-k3_low; run=CAoLSNkdR27AmHb84fnr8B; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12,"warning":null},{"id":"epoch-run-CAyLyhDHZiMQBZDB3qd8tG","modelId":"gemma-3-12b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"gemma-3-12b-it","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemma-3-12b-it; run=CAyLyhDHZiMQBZDB3qd8tG; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":16.666666666666664,"warning":null},{"id":"epoch-run-CBbeMKRRSeRsXnvSfn7Png","modelId":"glm-5.1","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"glm-5.1","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-5.1; run=CBbeMKRRSeRsXnvSfn7Png; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":89.8989898989899,"warning":null},{"id":"epoch-run-cbQnfsMSA3aQaBtjm3aFx2","modelId":"qwen3.7-max","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"qwen3.7-max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=qwen3.7-max; run=cbQnfsMSA3aQaBtjm3aFx2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26,"warning":null},{"id":"epoch-run-ccbx7KQY9bTD563vzBXwSg","modelId":"glm-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-5-model-extracted-exact-match","level":"unspecified","sourceLabel":"glm-5","configuration":"Epoch AI Inspect; task 1.0.5; scorer model_extracted_exact_match; model=glm-5; run=ccbx7KQY9bTD563vzBXwSg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80,"warning":"Run-specific qualification requires review: Two samples did not complete, which is greater than our 2% error tolerance setting. As such, the score has been manually edited to grade them as incorrect."},{"id":"epoch-run-CdmtwZ8NR8YeTXWiLVTRDJ","modelId":"gpt-5.5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gpt-5.5_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.5_low; run=CdmtwZ8NR8YeTXWiLVTRDJ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.44444444444444,"warning":null},{"id":"epoch-run-CeDghdBk8CqtiFSbUzxNH5","modelId":"gpt-5.6-luna","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"max","sourceLabel":"gpt-5.6-luna_max","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=gpt-5.6-luna_max; run=CeDghdBk8CqtiFSbUzxNH5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":41,"warning":null},{"id":"epoch-run-cF97JqHbFxxp7ePjYVK9HV","modelId":"gemini-3.6-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"minimal","sourceLabel":"gemini-3.6-flash_minimal","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.6-flash_minimal; run=cF97JqHbFxxp7ePjYVK9HV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80,"warning":null},{"id":"epoch-run-CFBCC2QrkCAFBAvFj5pqPF","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gemini-3.1-flash-lite_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.1-flash-lite_low; run=CFBCC2QrkCAFBAvFj5pqPF; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":44.44444444444444,"warning":null},{"id":"epoch-run-cFbntXuxUtfreULvt7vEt7","modelId":"qwen3.5-plus","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.5-plus","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-plus; run=cFbntXuxUtfreULvt7vEt7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.66666666666667,"warning":null},{"id":"epoch-run-CfmvbqzjpF6cCJQfzqdD4w","modelId":"qwen3-32b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"Qwen3-32B_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=Qwen3-32B_none; run=CfmvbqzjpF6cCJQfzqdD4w; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":23.055555555555557,"warning":null},{"id":"epoch-run-cgnopvpB6mzsfb9yGnYK2z","modelId":"claude-haiku-4-5-20251001","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"claude-haiku-4-5-20251001","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=claude-haiku-4-5-20251001; run=cgnopvpB6mzsfb9yGnYK2z; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":60.47979797979798,"warning":null},{"id":"epoch-run-chbh6XstiqtUa5W9zuyfjS","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gemini-3.5-flash-lite_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.5-flash-lite_high; run=chbh6XstiqtUa5W9zuyfjS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.33333333333334,"warning":null},{"id":"epoch-run-cjUVqF2qQJ4qhFKBU5BQpW","modelId":"gpt-5.6-terra","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"none","sourceLabel":"gpt-5.6-terra_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-terra_none; run=cjUVqF2qQJ4qhFKBU5BQpW; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":53.333333333333336,"warning":null},{"id":"epoch-run-ckbZdkrXPwbFdjn6bmUQsi","modelId":"claude-opus-4-8","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"claude-opus-4-8_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-opus-4-8_none; run=ckbZdkrXPwbFdjn6bmUQsi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.35353535353535,"warning":null},{"id":"epoch-run-CmHy8cVECGkvQTwakfjRo2","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"minimal","sourceLabel":"gemini-3.1-flash-lite_minimal","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.1-flash-lite_minimal; run=CmHy8cVECGkvQTwakfjRo2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":73.73737373737373,"warning":null},{"id":"epoch-run-cMwETYmC6b4yKupyJfEHv2","modelId":"gpt-oss-20b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gpt-oss-20b_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-oss-20b_low; run=cMwETYmC6b4yKupyJfEHv2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":40.27777777777778,"warning":null},{"id":"epoch-run-coxYk5vK5cBhQ7iZohMtA5","modelId":"claude-opus-4-6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"claude-opus-4-6_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-opus-4-6_max; run=coxYk5vK5cBhQ7iZohMtA5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.11111111111111,"warning":null},{"id":"epoch-run-cqUZ6gcq2HBnShFRWvKjGr","modelId":"gemini-2.5-pro","benchmarkId":"epoch-gpqa-diamond-1-0-1-choice","level":"unspecified","sourceLabel":"gemini-2.5-pro","configuration":"Epoch AI Inspect; task 1.0.1; scorer choice; model=gemini-2.5-pro; run=cqUZ6gcq2HBnShFRWvKjGr; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.29040404040404,"warning":null},{"id":"epoch-run-csK2fe5T8p2R7WgoJUu6bZ","modelId":"gpt-6-astra","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-astra_max; run=csK2fe5T8p2R7WgoJUu6bZ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.6842105263158,"warning":null},{"id":"epoch-run-cSTYZ3FKp5Mbs3xopn68NK","modelId":"magistral-small-2509","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"magistral-small-2509","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=magistral-small-2509; run=cSTYZ3FKp5Mbs3xopn68NK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.055555555555557,"warning":null},{"id":"epoch-run-CtmxkTPbFRofQWTkEw5aSe","modelId":"minimax-m3","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"MiniMax-M3","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=MiniMax-M3; run=CtmxkTPbFRofQWTkEw5aSe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":13,"warning":null},{"id":"epoch-run-cXkPywkhy3NnW992ri73yG","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-gpqa-diamond-1-0-2-choice","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_16K (thinking 16k)","configuration":"Epoch AI Inspect; task 1.0.2; scorer choice; model=claude-opus-4-5-20251101_16K; run=cXkPywkhy3NnW992ri73yG; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.47979797979798,"warning":null},{"id":"epoch-run-cXn5zzns6VFwbTgotDwZjV","modelId":"gemini-3.5-flash","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.5-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.5-flash_high; run=cXn5zzns6VFwbTgotDwZjV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26.82926829268293,"warning":null},{"id":"epoch-run-d23hYvSNGFo7rniLFyBWiu","modelId":"claude-fable-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"claude-fable-5_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-fable-5_high; run=d23hYvSNGFo7rniLFyBWiu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.33333333333334,"warning":null},{"id":"epoch-run-d3m9Xip8cof4s77z6LKkEh","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-sonnet-4-5-20250929_32K; run=d3m9Xip8cof4s77z6LKkEh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":2.4390243902439024,"warning":null},{"id":"epoch-run-D3mtasTumfvjR4fjmnHWRy","modelId":"claude-opus-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"claude-opus-5_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-opus-5_max; run=D3mtasTumfvjR4fjmnHWRy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.87626262626263,"warning":null},{"id":"epoch-run-D3YDUZSfEUDsS4frLbEKDu","modelId":"qwen3.7-max","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.7-max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.7-max; run=D3YDUZSfEUDsS4frLbEKDu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.55555555555556,"warning":null},{"id":"epoch-run-D5cbXacPnJ8akq2jAtifEQ","modelId":"gemma-4-31b-it","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"gemma-4-31b-it","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gemma-4-31b-it; run=D5cbXacPnJ8akq2jAtifEQ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":10.4,"warning":null},{"id":"epoch-run-d5QCJNq9uLtxjhiMhqXLmg","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"gemini-3.5-flash-lite_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.5-flash-lite_low; run=d5QCJNq9uLtxjhiMhqXLmg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21,"warning":null},{"id":"epoch-run-D7WDyCr24AVdzr4eKAreJu","modelId":"qwen3.5-397b-a17b","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.5-397b-a17b_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.5-397b-a17b_none; run=D7WDyCr24AVdzr4eKAreJu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-DB8ja6pEZWGocQbdciyisL","modelId":"gpt-oss-20b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"medium","sourceLabel":"gpt-oss-20b_medium","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-oss-20b_medium; run=DB8ja6pEZWGocQbdciyisL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":65.27777777777779,"warning":null},{"id":"epoch-run-dBHxnnTzLGz3nHis5LET6t","modelId":"claude-fable-5-1","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-fable-5-1_max; run=dBHxnnTzLGz3nHis5LET6t; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.99999999999999,"warning":null},{"id":"epoch-run-DCJQMqipndGxwMXe8BDGeg","modelId":"gpt-oss-20b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gpt-oss-20b_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-oss-20b_low; run=DCJQMqipndGxwMXe8BDGeg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":53.15656565656566,"warning":null},{"id":"epoch-run-DDfYPyKQW7vgDtGgkwcKSi","modelId":"gpt-5.6-sol","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"none","sourceLabel":"gpt-5.6-sol_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-sol_none; run=DDfYPyKQW7vgDtGgkwcKSi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.82828282828282,"warning":null},{"id":"epoch-run-DFWQcrbMegFAcEWZKWoJxC","modelId":"gpt-5.5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"gpt-5.5_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.5_xhigh; run=DFWQcrbMegFAcEWZKWoJxC; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":63,"warning":null},{"id":"epoch-run-DFzqUCfxnQPUdK8b8D7aYu","modelId":"claude-sonnet-5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"claude-sonnet-5_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-5_xhigh; run=DFzqUCfxnQPUdK8b8D7aYu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34,"warning":null},{"id":"epoch-run-DGEzZA6nvTne4MEGSuo3Tm","modelId":"ministral-8b-2410","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"ministral-8b-2410","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=ministral-8b-2410; run=DGEzZA6nvTne4MEGSuo3Tm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27.146464646464647,"warning":null},{"id":"epoch-run-dgSJTFQ4UtuQKEhB9ss6K7","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gemini-3.5-flash-lite_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.5-flash-lite_low; run=dgSJTFQ4UtuQKEhB9ss6K7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":75.75757575757575,"warning":null},{"id":"epoch-run-DGSmhVxWsi7chsuKS64VLg","modelId":"qwen3-30b-a3b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"Qwen3-30B-A3B_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=Qwen3-30B-A3B_none; run=DGSmhVxWsi7chsuKS64VLg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":50.378787878787875,"warning":null},{"id":"epoch-run-DhJ248BX3ZPBHXvkRG4Qsz","modelId":"gemma-3-27b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"gemma-3-27b-it","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemma-3-27b-it; run=DhJ248BX3ZPBHXvkRG4Qsz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":22.5,"warning":null},{"id":"epoch-run-DienatPbgfgn3EzM7Gvzxr","modelId":"gpt-5.6-terra","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"max","sourceLabel":"gpt-5.6-terra_max","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=gpt-5.6-terra_max; run=DienatPbgfgn3EzM7Gvzxr; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":43.2,"warning":null},{"id":"epoch-run-DiXCMhdfhcjaCUSzYWgACy","modelId":"claude-opus-4-6","benchmarkId":"epoch-gpqa-diamond-1-0-4-choice","level":"thinking","sourceLabel":"claude-opus-4-6_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.4; scorer choice; model=claude-opus-4-6_32K; run=DiXCMhdfhcjaCUSzYWgACy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.53030303030303,"warning":null},{"id":"epoch-run-djbbEqvBW9qux3pZwcZN2i","modelId":"grok-4.5","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"grok-4.5_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.5_high; run=djbbEqvBW9qux3pZwcZN2i; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.19298245614035,"warning":null},{"id":"epoch-run-DjFkVt62QwqQ4xXwzYR5Ng","modelId":"qwen3.5-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.5-flash_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.5-flash_none; run=DjFkVt62QwqQ4xXwzYR5Ng; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":16,"warning":null},{"id":"epoch-run-dKCWFaoGCWJTnD5CPWQGru","modelId":"grok-4.6","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"xhigh","sourceLabel":"grok-4.6_xhigh","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=grok-4.6_xhigh; run=dKCWFaoGCWJTnD5CPWQGru; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":48.9,"warning":null},{"id":"epoch-run-dKZDn279PKuc2Kxm68D4r2","modelId":"qwen3.6-27b","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.6-27b_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.6-27b_none; run=dKZDn279PKuc2Kxm68D4r2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.000000000000001,"warning":null},{"id":"epoch-run-dLDBahYBTGRNjbsXA5pyr3","modelId":"glm-5.3","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"glm-5.3_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-5.3_max; run=dLDBahYBTGRNjbsXA5pyr3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.9090909090909,"warning":null},{"id":"epoch-run-DmdKPD9Z9RXwYyATAoG4Uw","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"claude-sonnet-4-6_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-sonnet-4-6_high; run=DmdKPD9Z9RXwYyATAoG4Uw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.33333333333334,"warning":null},{"id":"epoch-run-DMjUrG6eZEqwcQ4SqUoViu","modelId":"qwen3.7-max","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"qwen3.7-max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=qwen3.7-max; run=DMjUrG6eZEqwcQ4SqUoViu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":55.76730190571715,"warning":null},{"id":"epoch-run-DmptTSPXx5uM5TYZoV59JT","modelId":"qwen3.7-plus","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"qwen3.7-plus","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.7-plus; run=DmptTSPXx5uM5TYZoV59JT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17,"warning":null},{"id":"epoch-run-DnFTaE9qNsmBdYNqohhynY","modelId":"gpt-6-sol","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-6-sol_max; run=DnFTaE9qNsmBdYNqohhynY; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.00000000000001,"warning":null},{"id":"epoch-run-donMeEKBZZdqpQsojKWJEm","modelId":"gpt-6-sol","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gpt-6-sol_max; run=donMeEKBZZdqpQsojKWJEm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":60.699999999999996,"warning":null},{"id":"epoch-run-doNrgGQY5SytQV22WRqw5L","modelId":"qwen3.7-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.7-flash_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.7-flash_none; run=doNrgGQY5SytQV22WRqw5L; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19.298245614035086,"warning":null},{"id":"epoch-run-dqgqP2uEPswfECCVj2Kbr9","modelId":"claude-opus-4-8","benchmarkId":"epoch-gpqa-diamond-1-0-10-choice","level":"max","sourceLabel":"claude-opus-4-8_max","configuration":"Epoch AI Inspect; task 1.0.10; scorer choice; model=claude-opus-4-8_max; run=dqgqP2uEPswfECCVj2Kbr9; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.03535353535354,"warning":null},{"id":"epoch-run-dszRh7ELyCEukWnwYnUxMp","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"deepseek-r1-0528-qwen3-8b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=deepseek-r1-0528-qwen3-8b; run=dszRh7ELyCEukWnwYnUxMp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":9.280303030303031,"warning":null},{"id":"epoch-run-dwyb2yzFewGT6w6oFdVMAc","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Qwen-32B","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=DeepSeek-R1-Distill-Qwen-32B; run=dwyb2yzFewGT6w6oFdVMAc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":64.14141414141415,"warning":null},{"id":"epoch-run-dYYtcwENFRtm9jZTDA2p9C","modelId":"gemini-3.5-flash","benchmarkId":"epoch-gpqa-diamond-1-0-9-choice","level":"high","sourceLabel":"gemini-3.5-flash_high","configuration":"Epoch AI Inspect; task 1.0.9; scorer choice; model=gemini-3.5-flash_high; run=dYYtcwENFRtm9jZTDA2p9C; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":92.8030303030303,"warning":null},{"id":"epoch-run-E28uZiJvrPgKtawbJpoZQ8","modelId":"claude-fable-5","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-fable-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-fable-5_max; run=E28uZiJvrPgKtawbJpoZQ8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.01754385964912,"warning":null},{"id":"epoch-run-e8MXnAM8HMF8eEzYtYA7xv","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"medium","sourceLabel":"claude-sonnet-4-6_medium","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-sonnet-4-6_medium; run=e8MXnAM8HMF8eEzYtYA7xv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.33333333333334,"warning":null},{"id":"epoch-run-EatRQnSaUpUwKzdtjYnQHo","modelId":"gpt-5.5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"xhigh","sourceLabel":"gpt-5.5_xhigh","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.5_xhigh; run=EatRQnSaUpUwKzdtjYnQHo; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.00000000000001,"warning":null},{"id":"epoch-run-eB9e2mtd5nRwqdFyLEsz6k","modelId":"claude-opus-4-6","benchmarkId":"epoch-gpqa-diamond-1-0-4-choice","level":"thinking","sourceLabel":"claude-opus-4-6_64K (thinking 64k)","configuration":"Epoch AI Inspect; task 1.0.4; scorer choice; model=claude-opus-4-6_64K; run=eB9e2mtd5nRwqdFyLEsz6k; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.76262626262627,"warning":null},{"id":"epoch-run-ECdG3m8n5ABfWVZ55rPFmT","modelId":"glm-5.2","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"glm-5.2_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.2_none; run=ECdG3m8n5ABfWVZ55rPFmT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":42.45614035087719,"warning":null},{"id":"epoch-run-EceR3NeHzSxacdmAmfRxgT","modelId":"deepseek-r1-distill-qwen-14b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Qwen-14B","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=DeepSeek-R1-Distill-Qwen-14B; run=EceR3NeHzSxacdmAmfRxgT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":50.55555555555556,"warning":null},{"id":"epoch-run-eCq63CuekohAswYftWXJVj","modelId":"kimi-k3","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"kimi-k3_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=kimi-k3_max; run=eCq63CuekohAswYftWXJVj; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26,"warning":null},{"id":"epoch-run-EDFusDaaFkh5n2MqKNxMBD","modelId":"deepseek-v4-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"deepseek-v4-flash-0731_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=deepseek-v4-flash-0731_max; run=EDFusDaaFkh5n2MqKNxMBD; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.03535353535354,"warning":null},{"id":"epoch-run-ehqiuEDETZTqJ8zt6Mc7Ja","modelId":"kimi-k2.6","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"kimi-k2.6","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=kimi-k2.6; run=ehqiuEDETZTqJ8zt6Mc7Ja; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20,"warning":null},{"id":"epoch-run-eiLEpRUvK48wUKDh4sYkzV","modelId":"claude-fable-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-5-model-extracted-exact-match","level":"max","sourceLabel":"claude-fable-5_max","configuration":"Epoch AI Inspect; task 1.1.5; scorer model_extracted_exact_match; model=claude-fable-5_max; run=eiLEpRUvK48wUKDh4sYkzV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":99.72222222222223,"warning":null},{"id":"epoch-run-eiTYduCDNq3HgwTg4ZhEw6","modelId":"gemini-3.5-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gemini-3.5-flash_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.5-flash_low; run=eiTYduCDNq3HgwTg4ZhEw6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.88888888888889,"warning":null},{"id":"epoch-run-eJvV6RezECZwh3b4ekg4pm","modelId":"seed-oss-36b-instruct","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"seed-oss-36b-instruct","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=seed-oss-36b-instruct; run=eJvV6RezECZwh3b4ekg4pm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":67.5,"warning":null},{"id":"epoch-run-ENc5ZM5TpBYNKkTWN4xyQM","modelId":"kimi-k2.7-code","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"kimi-k2.7-code","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=kimi-k2.7-code; run=ENc5ZM5TpBYNKkTWN4xyQM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.55555555555556,"warning":null},{"id":"epoch-run-ENh7ZPQj7ihBePjsC8a7HV","modelId":"qwen3.7-max","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.7-max_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.7-max_none; run=ENh7ZPQj7ihBePjsC8a7HV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26,"warning":null},{"id":"epoch-run-enhRcQmfN6RFB2zu7R3Uxr","modelId":"gpt-3.5-turbo-1106","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"gpt-3.5-turbo-1106","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=gpt-3.5-turbo-1106; run=enhRcQmfN6RFB2zu7R3Uxr; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.030303030303028,"warning":null},{"id":"epoch-run-Eo5cAQJK28CHpKDfCSDYG4","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3-30b-a3b-instruct-2507","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-30b-a3b-instruct-2507; run=Eo5cAQJK28CHpKDfCSDYG4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":62.22222222222222,"warning":null},{"id":"epoch-run-EoSFpgFhJB7iDPLcMNuLH4","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"minimal","sourceLabel":"gemini-3.1-flash-lite_minimal","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.1-flash-lite_minimal; run=EoSFpgFhJB7iDPLcMNuLH4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":37.77777777777778,"warning":null},{"id":"epoch-run-Epg7eNLwwGPXtX3sSAwPf8","modelId":"gpt-5.6-luna","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"gpt-5.6-luna_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-luna_max; run=Epg7eNLwwGPXtX3sSAwPf8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21,"warning":null},{"id":"epoch-run-ESnXA8LqMqhjcsoWQaMLVQ","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Qwen-1.5B","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=DeepSeek-R1-Distill-Qwen-1.5B; run=ESnXA8LqMqhjcsoWQaMLVQ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33.58585858585859,"warning":null},{"id":"epoch-run-EtgrDq6ei6kZe8K7QbXUe3","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"gemini-3-flash-preview","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3-flash-preview; run=EtgrDq6ei6kZe8K7QbXUe3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":51.2280701754386,"warning":null},{"id":"epoch-run-ETNzxNbPKwj4qxkdo3VVgT","modelId":"grok-4.20-0309-reasoning","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"grok-4.20-0309-reasoning","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=grok-4.20-0309-reasoning; run=ETNzxNbPKwj4qxkdo3VVgT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.999999999999996,"warning":null},{"id":"epoch-run-eVKA2eavdRAbSN6AnzyBxA","modelId":"kimi-k2.6","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"kimi-k2.6","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=kimi-k2.6; run=eVKA2eavdRAbSN6AnzyBxA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25.64102564102564,"warning":null},{"id":"epoch-run-EWAyxFsRUvuu4NpmmLqWkn","modelId":"qwen3.6-35b-a3b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.6-35b-a3b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.6-35b-a3b; run=EWAyxFsRUvuu4NpmmLqWkn; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.83838383838383,"warning":null},{"id":"epoch-run-EXpHB6kFqDjRfG9A5LsDf6","modelId":"qwen3-32b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"Qwen3-32B_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=Qwen3-32B_none; run=EXpHB6kFqDjRfG9A5LsDf6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":54.10353535353535,"warning":null},{"id":"epoch-run-eXVWr3RBeSsWy5DAU8ed46","modelId":"gemini-3.1-pro","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"gemini-3.1-pro-preview","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.1-pro-preview; run=eXVWr3RBeSsWy5DAU8ed46; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26.82926829268293,"warning":null},{"id":"epoch-run-ExyL3GN8s34tizrXV3VmS3","modelId":"claude-fable-5-1","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-fable-5-1_max; run=ExyL3GN8s34tizrXV3VmS3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70.8,"warning":null},{"id":"epoch-run-eYmgjYJfM7Utm46nfoRfZ6","modelId":"gemini-3.5-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-4-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3.5-flash_high","configuration":"Epoch AI Inspect; task 1.1.4; scorer model_extracted_exact_match; model=gemini-3.5-flash_high; run=eYmgjYJfM7Utm46nfoRfZ6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.55555555555556,"warning":null},{"id":"epoch-run-EzMaDdXs4YQTMLgppBGxAb","modelId":"qwen3.5-397b-a17b","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.5-397b-a17b","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.5-397b-a17b; run=EzMaDdXs4YQTMLgppBGxAb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":29.47368421052631,"warning":null},{"id":"epoch-run-F3CtjUceBrMi9iYmjULx7J","modelId":"qwen3-1.7b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3-1.7b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-1.7b; run=F3CtjUceBrMi9iYmjULx7J; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":38.005050505050505,"warning":null},{"id":"epoch-run-f46NQY9jSxyKqo9qzrmk6F","modelId":"gpt-6-astra","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"gpt-6-astra_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-astra_high; run=f46NQY9jSxyKqo9qzrmk6F; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.6,"warning":null},{"id":"epoch-run-f6pnAATjxmrFJS353EjQKi","modelId":"deepseek-r1-distill-llama-70b","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Llama-70B","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=DeepSeek-R1-Distill-Llama-70B; run=f6pnAATjxmrFJS353EjQKi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":55.744949494949495,"warning":null},{"id":"epoch-run-f6ZdMtRpdfskGbrBCzrKSn","modelId":"qwen3.6-plus","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.6-plus","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.6-plus; run=f6ZdMtRpdfskGbrBCzrKSn; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.38383838383838,"warning":null},{"id":"epoch-run-f7BCT27LzcAtaLDP2tcrWF","modelId":"gpt-6-astra","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-astra_max; run=f7BCT27LzcAtaLDP2tcrWF; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.6,"warning":null},{"id":"epoch-run-f7dotY9NheKNKK2zz6bfKZ","modelId":"qwen3.7-max","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"qwen3.7-max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.7-max; run=f7dotY9NheKNKK2zz6bfKZ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32,"warning":null},{"id":"epoch-run-F7v3JtKFb27j2YVmsndLM5","modelId":"gpt-5.6-terra","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"none","sourceLabel":"gpt-5.6-terra_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-terra_none; run=F7v3JtKFb27j2YVmsndLM5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":77.27272727272727,"warning":null},{"id":"epoch-run-F83pa7Gfsb4UNVttTnS5Hk","modelId":"gemini-3.6-flash","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"gemini-3.6-flash_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gemini-3.6-flash_high; run=F83pa7Gfsb4UNVttTnS5Hk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.2,"warning":null},{"id":"epoch-run-f8Pac6hqib4t7x2s4HCRoT","modelId":"grok-4.5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"grok-4.5_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=grok-4.5_high; run=f8Pac6hqib4t7x2s4HCRoT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.77777777777777,"warning":null},{"id":"epoch-run-FA7pM9hKdkgTXmGHptcno7","modelId":"gpt-6-astra","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-6-astra_max; run=FA7pM9hKdkgTXmGHptcno7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84,"warning":null},{"id":"epoch-run-FagAUZ7m5PJHseDATvp2vf","modelId":"gpt-5.6-luna","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"gpt-5.6-luna_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-luna_low; run=FagAUZ7m5PJHseDATvp2vf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17,"warning":null},{"id":"epoch-run-fEgRyzyFYZraNnkkbRVSSg","modelId":"qwen3.5-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"qwen3.5-flash","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=qwen3.5-flash; run=fEgRyzyFYZraNnkkbRVSSg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":14.000000000000002,"warning":null},{"id":"epoch-run-ff7GeS7rdperjxSBZ7MHZY","modelId":"claude-opus-5","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-5_max; run=ff7GeS7rdperjxSBZ7MHZY; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.6140350877193,"warning":null},{"id":"epoch-run-FfAvNS5H9bMDSBF4JNNiDz","modelId":"claude-opus-5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"claude-opus-5_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-opus-5_max; run=FfAvNS5H9bMDSBF4JNNiDz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":59,"warning":null},{"id":"epoch-run-Ffoxqm7YsZr3FimCGsGAJv","modelId":"qwen3-32b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"Qwen3-32B","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=Qwen3-32B; run=Ffoxqm7YsZr3FimCGsGAJv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":65.71969696969697,"warning":null},{"id":"epoch-run-FGEsLTxzZ7fE9aq7DhdqZb","modelId":"llama-3.2-1b-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"Llama-3.2-1B-Instruct","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=Llama-3.2-1B-Instruct; run=FGEsLTxzZ7fE9aq7DhdqZb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":23.92676767676768,"warning":null},{"id":"epoch-run-FHZCKbtfvpKGXU99vZFnXN","modelId":"qwen3.7-max","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.7-max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.7-max; run=FHZCKbtfvpKGXU99vZFnXN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34.146341463414636,"warning":null},{"id":"epoch-run-fJKhFRm5dn9wzbXojee7uj","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"minimal","sourceLabel":"gemini-3.5-flash-lite_minimal","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3.5-flash-lite_minimal; run=fJKhFRm5dn9wzbXojee7uj; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12,"warning":null},{"id":"epoch-run-FkywQ2KS4AkLPzjSPmhHcS","modelId":"gemini-2.5-pro","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"gemini-2.5-pro","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-2.5-pro; run=FkywQ2KS4AkLPzjSPmhHcS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":24.561403508771928,"warning":null},{"id":"epoch-run-FLDt53waLTeshC7PK9uVy2","modelId":"grok-4.20-0309-reasoning","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"grok-4.20-0309-reasoning","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.20-0309-reasoning; run=FLDt53waLTeshC7PK9uVy2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":44.91228070175438,"warning":null},{"id":"epoch-run-FNMbfvEEm9PRnrJnrqk2PH","modelId":"mistral-medium-2505","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"mistral-medium-2505","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=mistral-medium-2505; run=FNMbfvEEm9PRnrJnrqk2PH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":59.53282828282829,"warning":null},{"id":"epoch-run-FNXAt76ZUS3sWhQDFMga7U","modelId":"claude-sonnet-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"claude-sonnet-5_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-sonnet-5_max; run=FNXAt76ZUS3sWhQDFMga7U; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80.3030303030303,"warning":null},{"id":"epoch-run-FoMSS4FmH3VxpsvZiUJPCv","modelId":"qwen3.6-flash","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"qwen3.6-flash","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=qwen3.6-flash; run=FoMSS4FmH3VxpsvZiUJPCv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":15.915915915915916,"warning":null},{"id":"epoch-run-fPE4Zajhq5WfccSxEYXXR7","modelId":"glm-4.7-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"glm-4.7-flash_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-4.7-flash_none; run=fPE4Zajhq5WfccSxEYXXR7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25,"warning":null},{"id":"epoch-run-FpQf5xBwJPb4ZEo6SgdKWq","modelId":"qwen-3.8-max-0902","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"qwen3.8-max-0902_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.8-max-0902_xhigh; run=FpQf5xBwJPb4ZEo6SgdKWq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":65.6140350877193,"warning":null},{"id":"epoch-run-FQ3Suh2vKacz7UNqJybTWk","modelId":"deepseek-v4-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"deepseek-v4-flash-0731_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=deepseek-v4-flash-0731_max; run=FQ3Suh2vKacz7UNqJybTWk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.44444444444444,"warning":null},{"id":"epoch-run-fQjoHKyUEPAuwdrBGfL8SG","modelId":"gemma-2-9b-it","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"gemma-2-9b-it","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=gemma-2-9b-it; run=fQjoHKyUEPAuwdrBGfL8SG; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27.46212121212121,"warning":null},{"id":"epoch-run-frszScNxpHXXjSHawCt9Zb","modelId":"kimi-k3","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"kimi-k3_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=kimi-k3_low; run=frszScNxpHXXjSHawCt9Zb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":68.88888888888889,"warning":null},{"id":"epoch-run-fsZmzw8jjyUgMQGwxodBwN","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"claude-sonnet-4-5-20250929","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=claude-sonnet-4-5-20250929; run=fsZmzw8jjyUgMQGwxodBwN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":73.73737373737373,"warning":null},{"id":"epoch-run-FTDFCpXzDTgfgWBNdHLHGc","modelId":"gemini-3.1-pro","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"gemini-3.1-pro-preview_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.1-pro-preview_high; run=FTDFCpXzDTgfgWBNdHLHGc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":51,"warning":null},{"id":"epoch-run-FueK5bFAzGbkUao9JAibms","modelId":"muse-spark-1.2","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"muse-spark-1.2_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=muse-spark-1.2_xhigh; run=FueK5bFAzGbkUao9JAibms; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27,"warning":null},{"id":"epoch-run-fWNxmiSUC7BX8whqvzssY8","modelId":"glm-5.3-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"glm-5.3-flash_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=glm-5.3-flash_max; run=fWNxmiSUC7BX8whqvzssY8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21,"warning":null},{"id":"epoch-run-fxwRTtt8bLV4DNutwtMEox","modelId":"llama-3.1-8b-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"Llama-3.1-8B-Instruct","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=Llama-3.1-8B-Instruct; run=fxwRTtt8bLV4DNutwtMEox; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26.957070707070706,"warning":null},{"id":"epoch-run-fzoUFmy77NpEDtXyA6w6ap","modelId":"grok-4.3","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"grok-4.3_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=grok-4.3_high; run=fzoUFmy77NpEDtXyA6w6ap; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.33333333333333,"warning":null},{"id":"epoch-run-G3dinPvwrci2Nu3m9pby2i","modelId":"qwen3-4b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3-4b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-4b; run=G3dinPvwrci2Nu3m9pby2i; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":52.33585858585859,"warning":null},{"id":"epoch-run-g4bbCaiFDqw2YYNDAcmn2a","modelId":"gemini-3.1-pro","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"high","sourceLabel":"gemini-3.1-pro-preview_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gemini-3.1-pro-preview_high; run=g4bbCaiFDqw2YYNDAcmn2a; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34,"warning":null},{"id":"epoch-run-g7czyLHudHegzP9yRD939X","modelId":"gpt-5.6-sol","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"none","sourceLabel":"gpt-5.6-sol_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-sol_none; run=g7czyLHudHegzP9yRD939X; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":68.88888888888889,"warning":null},{"id":"epoch-run-g7dXwZH2tWvqLg3X5idGs2","modelId":"claude-fable-5-1","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-fable-5-1_max; run=g7dXwZH2tWvqLg3X5idGs2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.17543859649123,"warning":null},{"id":"epoch-run-GbQrTQyihBUsb9sxbqT8pT","modelId":"gpt-oss-120b","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"high","sourceLabel":"gpt-oss-120b_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-oss-120b_high; run=GbQrTQyihBUsb9sxbqT8pT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":0,"warning":null},{"id":"epoch-run-GdfzaHmiZht2DKPZDtQAET","modelId":"qwen3-1.7b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3-1.7b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-1.7b_none; run=GdfzaHmiZht2DKPZDtQAET; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":30.61868686868687,"warning":null},{"id":"epoch-run-GEaAdb9UmoUFbEtBx7WzAR","modelId":"claude-haiku-4-5-20251001","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"thinking","sourceLabel":"claude-haiku-4-5-20251001_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=claude-haiku-4-5-20251001_32K; run=GEaAdb9UmoUFbEtBx7WzAR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.21212121212122,"warning":null},{"id":"epoch-run-GEwyLqZT93Hp55fdtHrJhm","modelId":"gpt-oss-20b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"medium","sourceLabel":"gpt-oss-20b_medium","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-oss-20b_medium; run=GEwyLqZT93Hp55fdtHrJhm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":60.79545454545454,"warning":null},{"id":"epoch-run-gG2B6fLUDBy5jRraxL2pz8","modelId":"qwen3-8b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3-8b","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-8b; run=gG2B6fLUDBy5jRraxL2pz8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.111111111111114,"warning":null},{"id":"epoch-run-ggG3n6WVPZn7Gu2RNyfQbK","modelId":"qwen3.6-35b-a3b","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.6-35b-a3b","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-35b-a3b; run=ggG3n6WVPZn7Gu2RNyfQbK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17.543859649122805,"warning":null},{"id":"epoch-run-GGpgwGRFVATVJdQr8TgRc9","modelId":"minimax-m3","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"MiniMax-M3","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=MiniMax-M3; run=GGpgwGRFVATVJdQr8TgRc9; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.000000000000001,"warning":null},{"id":"epoch-run-GJ6YsSuzyRD9YXe6SBwvsu","modelId":"gemini-3.5-flash","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"gemini-3.5-flash_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gemini-3.5-flash_high; run=GJ6YsSuzyRD9YXe6SBwvsu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.2,"warning":null},{"id":"epoch-run-gJizntrYwhnbpCviv9mNBN","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-4-5-20250929_32K; run=gJizntrYwhnbpCviv9mNBN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":10,"warning":null},{"id":"epoch-run-GL7htq7XRnbzqpAVK5f862","modelId":"glm-4.7-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"glm-4.7-flash","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-4.7-flash; run=GL7htq7XRnbzqpAVK5f862; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":60.542929292929294,"warning":null},{"id":"epoch-run-GMngscbSmqoPbGZWNv2Jvw","modelId":"gpt-oss-20b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gpt-oss-20b_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-oss-20b_high; run=GMngscbSmqoPbGZWNv2Jvw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":50.83333333333333,"warning":null},{"id":"epoch-run-goZR5yGxZic9TZ23ezCL3w","modelId":"glm-5.3","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"glm-5.3_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-5.3_max; run=goZR5yGxZic9TZ23ezCL3w; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.11111111111111,"warning":null},{"id":"epoch-run-GQKsxFbdBWB546zE8ci49f","modelId":"gemini-3.5-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"high","sourceLabel":"gemini-3.5-flash_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gemini-3.5-flash_high; run=GQKsxFbdBWB546zE8ci49f; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32,"warning":null},{"id":"epoch-run-GRQV9Q8gHoHnSEPWofiQuq","modelId":"qwen3-14b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3-14b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-14b_none; run=GRQV9Q8gHoHnSEPWofiQuq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":53.40909090909091,"warning":null},{"id":"epoch-run-gRRQHECkYhz5UaYsRKDh2F","modelId":"qwen3-14b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3-14b","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-14b; run=gRRQHECkYhz5UaYsRKDh2F; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.38888888888889,"warning":null},{"id":"epoch-run-Gs234P3jzGiLm68KjyzYZL","modelId":"deepseek-r1-distill-qwen-1.5b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Qwen-1.5B","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=DeepSeek-R1-Distill-Qwen-1.5B; run=Gs234P3jzGiLm68KjyzYZL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21.38888888888889,"warning":null},{"id":"epoch-run-gsBidnvEcmCfLnn5sxePtk","modelId":"claude-opus-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"claude-opus-5","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-opus-5; run=gsBidnvEcmCfLnn5sxePtk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.77777777777777,"warning":null},{"id":"epoch-run-GSde7u2Mb6GZGpubJjjPam","modelId":"glm-4.7-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"glm-4.7-flash_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-4.7-flash_none; run=GSde7u2Mb6GZGpubJjjPam; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45.13888888888889,"warning":null},{"id":"epoch-run-GThtce6GAmRAS8srQyGQoA","modelId":"claude-opus-4-8","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"claude-opus-4-8_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-opus-4-8_low; run=GThtce6GAmRAS8srQyGQoA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.38383838383838,"warning":null},{"id":"epoch-run-gu39ndv9QxWP3NRxSUJRaX","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"epoch-gpqa-diamond-1-0-3-choice","level":"unspecified","sourceLabel":"Qwen3-235B-A22B-Thinking-2507","configuration":"Epoch AI Inspect; task 1.0.3; scorer choice; model=Qwen3-235B-A22B-Thinking-2507; run=gu39ndv9QxWP3NRxSUJRaX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80.05050505050505,"warning":null},{"id":"epoch-run-gVdraDT3pXmaExP9VbqME4","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gemini-3-flash-preview_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3-flash-preview_high; run=gVdraDT3pXmaExP9VbqME4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":89.39393939393939,"warning":null},{"id":"epoch-run-GVm7khGvtdnQKGfkgAHD8q","modelId":"grok-4.20-0309-reasoning","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"grok-4.20-0309-reasoning","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=grok-4.20-0309-reasoning; run=GVm7khGvtdnQKGfkgAHD8q; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":30.2,"warning":null},{"id":"epoch-run-GVmgtgKMwEqc54f3jA7auP","modelId":"claude-opus-4-8","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-4-8_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-8_max; run=GVmgtgKMwEqc54f3jA7auP; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80,"warning":null},{"id":"epoch-run-GYzmrzMfeSjNjm2J6L9PeT","modelId":"glm-4.7","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-4-model-extracted-exact-match","level":"unspecified","sourceLabel":"glm-4.7","configuration":"Epoch AI Inspect; task 1.0.4; scorer model_extracted_exact_match; model=glm-4.7; run=GYzmrzMfeSjNjm2J6L9PeT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.33333333333334,"warning":null},{"id":"epoch-run-gZ4bwozTKtpmZTqnqX3a33","modelId":"gemini-3.8-flash","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"gemini-3.8-flash_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gemini-3.8-flash_high; run=gZ4bwozTKtpmZTqnqX3a33; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":69.69999999999999,"warning":null},{"id":"epoch-run-gZERSAQwxBAevQzBsWfUie","modelId":"kimi-k3","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"kimi-k3_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=kimi-k3_max; run=gZERSAQwxBAevQzBsWfUie; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":39.02439024390244,"warning":null},{"id":"epoch-run-gZMFRbVKnBmcnZU4YqqKMN","modelId":"qwen3.5-397b-a17b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.5-397b-a17b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-397b-a17b; run=gZMFRbVKnBmcnZU4YqqKMN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.85858585858585,"warning":null},{"id":"epoch-run-gzQH7StUXaxAtXgxkTgyxJ","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"minimal","sourceLabel":"gemini-3.5-flash-lite_minimal","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.5-flash-lite_minimal; run=gzQH7StUXaxAtXgxkTgyxJ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.000000000000001,"warning":null},{"id":"epoch-run-H2FvRXagQRQmiNa7WrNMU2","modelId":"qwen3.6-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.6-flash","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.6-flash; run=H2FvRXagQRQmiNa7WrNMU2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.33333333333334,"warning":null},{"id":"epoch-run-H3E7EjjXfGCPnm9ooLdmAo","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-3-model-extracted-exact-match","level":"unspecified","sourceLabel":"gemini-3-flash-preview","configuration":"Epoch AI Inspect; task 1.0.3; scorer model_extracted_exact_match; model=gemini-3-flash-preview; run=H3E7EjjXfGCPnm9ooLdmAo; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":92.77777777777779,"warning":null},{"id":"epoch-run-H8CebpQyV2qEE8SQzjBrWe","modelId":"gemini-3.5-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"minimal","sourceLabel":"gemini-3.5-flash_minimal","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.5-flash_minimal; run=H8CebpQyV2qEE8SQzjBrWe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27,"warning":null},{"id":"epoch-run-h9aoDxyWEhybvARHpyuoco","modelId":"qwen3.5-35b-a3b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.5-35b-a3b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-35b-a3b_none; run=h9aoDxyWEhybvARHpyuoco; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70,"warning":null},{"id":"epoch-run-hDESmyGFrbZDomjff3rqzD","modelId":"minimax-m3","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"MiniMax-M3","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=MiniMax-M3; run=hDESmyGFrbZDomjff3rqzD; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.9090909090909,"warning":null},{"id":"epoch-run-HhiYxEx3Bf8YDPUh34K4DC","modelId":"claude-sonnet-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"xhigh","sourceLabel":"claude-sonnet-5_xhigh","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-sonnet-5_xhigh; run=HhiYxEx3Bf8YDPUh34K4DC; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.53030303030303,"warning":null},{"id":"epoch-run-hHnQVatVTbQjM4sfrjw7GH","modelId":"kimi-k2.6","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"unspecified","sourceLabel":"kimi-k2.6","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=kimi-k2.6; run=hHnQVatVTbQjM4sfrjw7GH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34.9,"warning":null},{"id":"epoch-run-hJ6XXLBm9yfXtSbbhWuwpw","modelId":"gemini-3.6-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"gemini-3.6-flash_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.6-flash_low; run=hJ6XXLBm9yfXtSbbhWuwpw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":42,"warning":null},{"id":"epoch-run-hnP6Z88x8bPEBRZ2PW6kps","modelId":"grok-4.6","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"grok-4.6_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=grok-4.6_high; run=hnP6Z88x8bPEBRZ2PW6kps; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.00252525252525,"warning":null},{"id":"epoch-run-Hpyy6CpmXmd5q7Zbdq63CR","modelId":"gemini-3.8-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"gemini-3.8-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.8-flash_high; run=Hpyy6CpmXmd5q7Zbdq63CR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.99999999999999,"warning":null},{"id":"epoch-run-htvM9QKQJ8TfBxTzDxfZfp","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"medium","sourceLabel":"claude-sonnet-4-6_medium","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-4-6_medium; run=htvM9QKQJ8TfBxTzDxfZfp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":16,"warning":null},{"id":"epoch-run-Hui8iZgjwKPxz7DkRU8sJX","modelId":"glm-5.2","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"glm-5.2_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=glm-5.2_max; run=Hui8iZgjwKPxz7DkRU8sJX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":22,"warning":null},{"id":"epoch-run-hVifTUdQZrFEYabgCRu8Qq","modelId":"gemini-3.1-pro","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"gemini-3.1-pro-preview","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.1-pro-preview; run=hVifTUdQZrFEYabgCRu8Qq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":59.64912280701754,"warning":null},{"id":"epoch-run-hwrnvtEbDAocFtjcsHqsQh","modelId":"claude-fable-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"claude-fable-5_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-fable-5_high; run=hwrnvtEbDAocFtjcsHqsQh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":100,"warning":null},{"id":"epoch-run-HX9sRVWJtZiNvneALnVzs8","modelId":"gpt-6-luna","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gpt-6-luna_max; run=HX9sRVWJtZiNvneALnVzs8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":41.4,"warning":null},{"id":"epoch-run-hYATFPeUnMq7Std2nLUVAJ","modelId":"gpt-oss-20b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gpt-oss-20b_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-oss-20b_high; run=hYATFPeUnMq7Std2nLUVAJ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45.95959595959596,"warning":null},{"id":"epoch-run-HzNnNaRmyNExSVaMjmFCHe","modelId":"gpt-5.6-terra","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"none","sourceLabel":"gpt-5.6-terra_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-terra_none; run=HzNnNaRmyNExSVaMjmFCHe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":5,"warning":null},{"id":"epoch-run-i2ZUgMREKiFAnE2M54w3Py","modelId":"qwen3.5-397b-a17b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.5-397b-a17b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-397b-a17b_none; run=i2ZUgMREKiFAnE2M54w3Py; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.22222222222221,"warning":null},{"id":"epoch-run-i6i3gQf3ng2aAipvowPpws","modelId":"qwen3.7-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.7-flash","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.7-flash; run=i6i3gQf3ng2aAipvowPpws; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.32323232323232,"warning":null},{"id":"epoch-run-i7MVHsH5pikQKjYn6LBqHS","modelId":"qwen3.7-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.7-flash_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.7-flash_none; run=i7MVHsH5pikQKjYn6LBqHS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":14.000000000000002,"warning":null},{"id":"epoch-run-iBQxysYR7zPxB6hxvJyTLc","modelId":"gpt-6-luna","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-luna_max; run=iBQxysYR7zPxB6hxvJyTLc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":78.94736842105263,"warning":null},{"id":"epoch-run-icTHLT2S2egxYtgbjQHUng","modelId":"claude-opus-4-8","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"claude-opus-4-8_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-8_low; run=icTHLT2S2egxYtgbjQHUng; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.999999999999996,"warning":null},{"id":"epoch-run-iCysWvFJevNyw2dbYosU6J","modelId":"glm-5.3-flash","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"glm-5.3-flash_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.3-flash_max; run=iCysWvFJevNyw2dbYosU6J; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17.073170731707318,"warning":null},{"id":"epoch-run-iE9NNTJWbGpRia2fUVGkPf","modelId":"gpt-5.6-terra","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"low","sourceLabel":"gpt-5.6-terra_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-terra_low; run=iE9NNTJWbGpRia2fUVGkPf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19,"warning":null},{"id":"epoch-run-ig7gQ23WRms9SQLLNPtbTi","modelId":"gemini-3.5-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gemini-3.5-flash_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.5-flash_low; run=ig7gQ23WRms9SQLLNPtbTi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.88888888888889,"warning":null},{"id":"epoch-run-ii6JE9j57eSbuswEsJArTa","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=claude-sonnet-4-5-20250929_32K; run=ii6JE9j57eSbuswEsJArTa; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":81.7258883248731,"warning":null},{"id":"epoch-run-ikUWdu6HJ2qRgZynfzDCFe","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-2-model-extracted-exact-match","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.2; scorer model_extracted_exact_match; model=claude-opus-4-5-20251101_32K; run=ikUWdu6HJ2qRgZynfzDCFe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.11111111111111,"warning":null},{"id":"epoch-run-iTY94tKhVEcZsUZivMwLJa","modelId":"gpt-5.6-luna","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"low","sourceLabel":"gpt-5.6-luna_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-luna_low; run=iTY94tKhVEcZsUZivMwLJa; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17,"warning":null},{"id":"epoch-run-ivr2LfmdZoCMfLME7jh2L8","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"minimal","sourceLabel":"gemini-3.5-flash-lite_minimal","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.5-flash-lite_minimal; run=ivr2LfmdZoCMfLME7jh2L8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":51.11111111111111,"warning":null},{"id":"epoch-run-iXvqUzwxAiA3Rg3dwnutFs","modelId":"glm-5.3","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"glm-5.3_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.3_max; run=iXvqUzwxAiA3Rg3dwnutFs; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":68.77192982456141,"warning":null},{"id":"epoch-run-J2kUEjR9r6Spd594qDNTTf","modelId":"claude-haiku-4-5-20251001","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"unspecified","sourceLabel":"claude-haiku-4-5-20251001","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=claude-haiku-4-5-20251001; run=J2kUEjR9r6Spd594qDNTTf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":13.200000000000001,"warning":null},{"id":"epoch-run-J2ZgtxyaQZfyDMPTPxS8nm","modelId":"gpt-5.6-terra","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"gpt-5.6-terra_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-terra_max; run=J2ZgtxyaQZfyDMPTPxS8nm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47,"warning":null},{"id":"epoch-run-J6evjTMgHkrun8txnEvzdN","modelId":"gpt-5.6-terra","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-5.6-terra_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-terra_max; run=J6evjTMgHkrun8txnEvzdN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.96491228070175,"warning":null},{"id":"epoch-run-J8hYKbXiGhetQRwu5bUSp6","modelId":"glm-5.2","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"glm-5.2_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-5.2_none; run=J8hYKbXiGhetQRwu5bUSp6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.888888888888886,"warning":null},{"id":"epoch-run-j99TDv5LAYUSvTUcR47T75","modelId":"qwen3.6-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"qwen3.6-flash","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=qwen3.6-flash; run=j99TDv5LAYUSvTUcR47T75; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12,"warning":null},{"id":"epoch-run-JarZHkwUMZ6HuA4ArMNT45","modelId":"claude-opus-4-6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"low","sourceLabel":"claude-opus-4-6_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=claude-opus-4-6_low; run=JarZHkwUMZ6HuA4ArMNT45; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.000000000000001,"warning":null},{"id":"epoch-run-jbyVWnEvvFVNNc8AEF6WcQ","modelId":"qwen3.6-plus","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.6-plus","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-plus; run=jbyVWnEvvFVNNc8AEF6WcQ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":38.24561403508772,"warning":null},{"id":"epoch-run-JCMjXU6FRiiLV3xvs9R2Xv","modelId":"gemini-3.6-flash","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.6-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.6-flash_high; run=JCMjXU6FRiiLV3xvs9R2Xv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21.951219512195124,"warning":null},{"id":"epoch-run-JcQMgXdZKU54u52B8TGQKp","modelId":"claude-opus-4-6","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"claude-opus-4-6_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-opus-4-6_max; run=JcQMgXdZKU54u52B8TGQKp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.38383838383838,"warning":null},{"id":"epoch-run-JcuoAWhcnngPEZBPmiUyLF","modelId":"gpt-5.6-terra","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gpt-5.6-terra_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-terra_low; run=JcuoAWhcnngPEZBPmiUyLF; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.37373737373737,"warning":null},{"id":"epoch-run-jD6rJdvfNPErKEz9qkpssj","modelId":"gemma-2-27b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"gemma-2-27b-it","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=gemma-2-27b-it; run=jD6rJdvfNPErKEz9qkpssj; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":1.3888888888888888,"warning":null},{"id":"epoch-run-jdFH2zZdssYKDVJQPKXvNB","modelId":"gemma-3-27b-it","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"gemma-3-27b-it","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemma-3-27b-it; run=jdFH2zZdssYKDVJQPKXvNB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47.72727272727273,"warning":null},{"id":"epoch-run-jDWEviMweCdzGxLfz8sdBm","modelId":"claude-opus-4-8","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"non-reasoning","sourceLabel":"claude-opus-4-8_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-8_none; run=jDWEviMweCdzGxLfz8sdBm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19,"warning":null},{"id":"epoch-run-JebA8dYLVov6H49ZQSLXh5","modelId":"glm-5.2","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"glm-5.2_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.2_max; run=JebA8dYLVov6H49ZQSLXh5; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":29.268292682926827,"warning":null},{"id":"epoch-run-Jevnk9hi9oRBbGcFu2JxBC","modelId":"gpt-6-luna","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-6-luna_max; run=Jevnk9hi9oRBbGcFu2JxBC; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36,"warning":null},{"id":"epoch-run-jFJucVunfLgoQqoFvPVQWP","modelId":"gemma-4-31b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"minimal","sourceLabel":"gemma-4-31b-it_minimal","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemma-4-31b-it_minimal; run=jFJucVunfLgoQqoFvPVQWP; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":73.33333333333333,"warning":null},{"id":"epoch-run-jGgkDRHbodcaTzQ3Mp6SxS","modelId":"qwen3.6-plus","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.6-plus_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.6-plus_none; run=jGgkDRHbodcaTzQ3Mp6SxS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12,"warning":null},{"id":"epoch-run-ji3kMn49Jcr9t4whxcpjj4","modelId":"claude-opus-5","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-5_max; run=ji3kMn49Jcr9t4whxcpjj4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":73.17073170731707,"warning":null},{"id":"epoch-run-JiqgqBTfkGx797uowZgJbk","modelId":"claude-opus-4-8","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-4-8_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-8_max; run=JiqgqBTfkGx797uowZgJbk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.09756097560976,"warning":null},{"id":"epoch-run-JJyzHWKJUA3iTZDU8iYefz","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"gemini-3.5-flash-lite_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.5-flash-lite_high; run=JJyzHWKJUA3iTZDU8iYefz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":16,"warning":null},{"id":"epoch-run-JoraLb9y55QXpMg4Rrmr5x","modelId":"gemini-3.6-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gemini-3.6-flash_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.6-flash_low; run=JoraLb9y55QXpMg4Rrmr5x; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.36363636363636,"warning":null},{"id":"epoch-run-jrawnM2wHFbVfJFJsap4kS","modelId":"qwen3.5-plus","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"qwen3.5-plus","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=qwen3.5-plus; run=jrawnM2wHFbVfJFJsap4kS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25.35070140280561,"warning":null},{"id":"epoch-run-jRiWmBxtBAyiXiRBRMY8wX","modelId":"qwen3-8b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3-8b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-8b; run=jRiWmBxtBAyiXiRBRMY8wX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.755050505050505,"warning":null},{"id":"epoch-run-jtfdCURnwfZzak4BSiBKzx","modelId":"grok-4.6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"grok-4.6_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=grok-4.6_high; run=jtfdCURnwfZzak4BSiBKzx; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.77777777777777,"warning":null},{"id":"epoch-run-jTuYE8VJTos73Kw77UFevq","modelId":"grok-4.5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"grok-4.5_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=grok-4.5_high; run=jTuYE8VJTos73Kw77UFevq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":35,"warning":null},{"id":"epoch-run-Ju7pUkkwYVj6ugi9Ri5Jjp","modelId":"claude-opus-4-7","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"claude-opus-4-7_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-opus-4-7_max; run=Ju7pUkkwYVj6ugi9Ri5Jjp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.36363636363636,"warning":null},{"id":"epoch-run-JuM7FArkmhLqqKBqudySsC","modelId":"gemini-3.5-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"minimal","sourceLabel":"gemini-3.5-flash_minimal","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.5-flash_minimal; run=JuM7FArkmhLqqKBqudySsC; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80,"warning":null},{"id":"epoch-run-jUmmCNezoW7jWiK72SVCaP","modelId":"gpt-5.6-terra","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"gpt-5.6-terra_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-terra_max; run=jUmmCNezoW7jWiK72SVCaP; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":99.72222222222223,"warning":null},{"id":"epoch-run-juNVfKBcYTwaKnk9GRA2zL","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.5-flash-lite_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.5-flash-lite_high; run=juNVfKBcYTwaKnk9GRA2zL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":0,"warning":null},{"id":"epoch-run-Jvdw6GAXiS6T79bkpbk6Ra","modelId":"qwen3.7-plus","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.7-plus","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.7-plus; run=Jvdw6GAXiS6T79bkpbk6Ra; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.33333333333333,"warning":null},{"id":"epoch-run-Jvt5BNr4ERbE9QGgMcEbmT","modelId":"glm-5.3-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"glm-5.3-flash_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-5.3-flash_max; run=Jvt5BNr4ERbE9QGgMcEbmT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.15151515151516,"warning":null},{"id":"epoch-run-jWHk7BTYHQRXk2rXCbnpcS","modelId":"minimax-m3","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"MiniMax-M3_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=MiniMax-M3_none; run=jWHk7BTYHQRXk2rXCbnpcS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":8,"warning":null},{"id":"epoch-run-jwS3LYLCPbZpHDzzkX6LyV","modelId":"gpt-6-sol","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-sol_max; run=jwS3LYLCPbZpHDzzkX6LyV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90,"warning":null},{"id":"epoch-run-jyCZnpXG2bh9G43X8pT75m","modelId":"claude-opus-4-7","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-2-model-extracted-exact-match","level":"xhigh","sourceLabel":"claude-opus-4-7_xhigh","configuration":"Epoch AI Inspect; task 1.1.2; scorer model_extracted_exact_match; model=claude-opus-4-7_xhigh; run=jyCZnpXG2bh9G43X8pT75m; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.8,"warning":"Run-specific qualification requires review: All incorrect answers were actually cases of the model failing to submit an answer. For each sample, we average the score of the epochs that did complete for that sample. If none completed, we assign a score of 0. Then we average these scores across all samples. The resulting score is 44/45."},{"id":"epoch-run-jYiGAFfUQKJuRStPzU3PX3","modelId":"qwen3.6-plus","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.6-plus","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.6-plus; run=jYiGAFfUQKJuRStPzU3PX3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.33333333333333,"warning":null},{"id":"epoch-run-JYuFdyZZSiMvmSZNzMTywb","modelId":"qwen3.5-397b-a17b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.5-397b-a17b","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-397b-a17b; run=JYuFdyZZSiMvmSZNzMTywb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.88888888888889,"warning":null},{"id":"epoch-run-k4zGHaXESxFRTRMkyDLZrz","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_48K (thinking 48k)","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-opus-4-5-20251101_48K; run=k4zGHaXESxFRTRMkyDLZrz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":22,"warning":null},{"id":"epoch-run-K8zTckSXLKLesoLBv24TWc","modelId":"qwen3.5-flash","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"qwen3.5-flash","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=qwen3.5-flash; run=K8zTckSXLKLesoLBv24TWc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20.32032032032032,"warning":null},{"id":"epoch-run-KaEir82ukWPm2Fq4FanY2M","modelId":"gpt-6-astra","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-6-astra_max; run=KaEir82ukWPm2Fq4FanY2M; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":74,"warning":null},{"id":"epoch-run-KAPGpH2CLMafAt8MnAQJT3","modelId":"gemma-3-4b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"gemma-3-4b-it","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemma-3-4b-it; run=KAPGpH2CLMafAt8MnAQJT3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.5,"warning":null},{"id":"epoch-run-KbUNQf8nnwPa8ixg6AnEwB","modelId":"deepseek-r1-distill-qwen-32b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"DeepSeek-R1-Distill-Qwen-32B","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=DeepSeek-R1-Distill-Qwen-32B; run=KbUNQf8nnwPa8ixg6AnEwB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":55.55555555555556,"warning":null},{"id":"epoch-run-kCQr6ARwJyUfkwWqKAfuDX","modelId":"qwen3.5-plus","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"unspecified","sourceLabel":"qwen3.5-plus","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=qwen3.5-plus; run=kCQr6ARwJyUfkwWqKAfuDX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":16,"warning":null},{"id":"epoch-run-kDNNbUYXhrAH5uDBGykFaf","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"medium","sourceLabel":"claude-sonnet-4-6_medium","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-sonnet-4-6_medium; run=kDNNbUYXhrAH5uDBGykFaf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.22222222222221,"warning":null},{"id":"epoch-run-KesKa5EXtFG7H4UMBFzgCh","modelId":"gpt-5.5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"none","sourceLabel":"gpt-5.5_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.5_none; run=KesKa5EXtFG7H4UMBFzgCh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":77.27272727272727,"warning":null},{"id":"epoch-run-KfQKr2ibCFMx5xBELoiM3Y","modelId":"gemma-3-12b-it","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"gemma-3-12b-it","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemma-3-12b-it; run=KfQKr2ibCFMx5xBELoiM3Y; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":39.457070707070706,"warning":null},{"id":"epoch-run-kG3hbd8qLEd6fLAJHyQTiN","modelId":"glm-5.3","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"glm-5.3_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=glm-5.3_max; run=kG3hbd8qLEd6fLAJHyQTiN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":37,"warning":null},{"id":"epoch-run-kgJ2d6XbKj3dZutDcBYSKU","modelId":"claude-opus-4-6","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-4-6_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-6_max; run=kgJ2d6XbKj3dZutDcBYSKU; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":65.96491228070175,"warning":null},{"id":"epoch-run-KH7KfZvFc9kZbqiJFdfYTc","modelId":"qwen3-4b-instruct-2507","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3-4b-instruct-2507","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-4b-instruct-2507; run=KH7KfZvFc9kZbqiJFdfYTc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":52.22222222222223,"warning":null},{"id":"epoch-run-KKm5DbPUgXo88eYr64g4SK","modelId":"qwen3.5-35b-a3b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3.5-35b-a3b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-35b-a3b_none; run=KKm5DbPUgXo88eYr64g4SK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":81.18686868686868,"warning":null},{"id":"epoch-run-KM4G49tQZB8ce67RxYGyMy","modelId":"minimax-m3","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"MiniMax-M3_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=MiniMax-M3_none; run=KM4G49tQZB8ce67RxYGyMy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":81.31313131313132,"warning":null},{"id":"epoch-run-Knij32Jp6bQJ2V9NHpg5VS","modelId":"gpt-5.5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"high","sourceLabel":"gpt-5.5_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.5_high; run=Knij32Jp6bQJ2V9NHpg5VS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":52,"warning":null},{"id":"epoch-run-KptKXK5nE2Ns2WbiaSSnbc","modelId":"gemini-3.7-flash","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.7-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.7-flash_high; run=KptKXK5nE2Ns2WbiaSSnbc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36.58536585365854,"warning":null},{"id":"epoch-run-kQjrYCaa7yYyMXofeR23Dd","modelId":"gemini-3.7-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3.7-flash_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.7-flash_high; run=kQjrYCaa7yYyMXofeR23Dd; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.22222222222221,"warning":null},{"id":"epoch-run-kR9F3oHBCnwQW4AiTPKwpd","modelId":"gemini-3.1-pro","benchmarkId":"epoch-gpqa-diamond-1-0-5-choice","level":"unspecified","sourceLabel":"gemini-3.1-pro-preview","configuration":"Epoch AI Inspect; task 1.0.5; scorer choice; model=gemini-3.1-pro-preview; run=kR9F3oHBCnwQW4AiTPKwpd; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.1,"warning":"Run-specific qualification requires review: In the initial run, 3% of samples failed to complete. Inspect defaults to not scoring these at all, but we override this behavior when >2% of epochs fail. For each sample, we average the score of the epochs that did complete for that sample. If none completed, we assign the random baseline of 25%. Then we average these scores across all samples. In this case, the resulting difference is very small: 94.1% modified vs. 94.4% original."},{"id":"epoch-run-kRPSnHnDdheajNdqmKsMjs","modelId":"gpt-6-astra","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"low","sourceLabel":"gpt-6-astra_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-astra_low; run=kRPSnHnDdheajNdqmKsMjs; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.8048780487805,"warning":null},{"id":"epoch-run-Kwwo3UFcPDk6iLAeSeeRaa","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-gpqa-diamond-1-0-2-choice","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.2; scorer choice; model=claude-opus-4-5-20251101_32K; run=Kwwo3UFcPDk6iLAeSeeRaa; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.0479797979798,"warning":null},{"id":"epoch-run-KxvHAT9SV2dTDn7jBnTk6f","modelId":"claude-opus-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"claude-opus-5_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-opus-5_max; run=KxvHAT9SV2dTDn7jBnTk6f; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":98.88888888888889,"warning":null},{"id":"epoch-run-kZ3VgrQJxfKQ5msjpmSaAz","modelId":"claude-opus-4-6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-4-model-extracted-exact-match","level":"thinking","sourceLabel":"claude-opus-4-6_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.4; scorer model_extracted_exact_match; model=claude-opus-4-6_32K; run=kZ3VgrQJxfKQ5msjpmSaAz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.05555555555556,"warning":null},{"id":"epoch-run-kZUdrqdGtCEJa3ip6L9YnH","modelId":"qwen-3.8-max-0902","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"qwen3.8-max-0902_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.8-max-0902_xhigh; run=kZUdrqdGtCEJa3ip6L9YnH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34.146341463414636,"warning":null},{"id":"epoch-run-L7kW8yxFKCjVXhxNnt3VTq","modelId":"gpt-6-astra","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-6-astra_max; run=L7kW8yxFKCjVXhxNnt3VTq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":100,"warning":null},{"id":"epoch-run-LcttjHHzTyAW6MqNE29xiK","modelId":"qwen3-8b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3-8b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-8b_none; run=LcttjHHzTyAW6MqNE29xiK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47.474747474747474,"warning":null},{"id":"epoch-run-Lgy52ZT4YpFrUqqKVu2qUs","modelId":"gemini-3.8-flash","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.8-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.8-flash_high; run=Lgy52ZT4YpFrUqqKVu2qUs; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":21.951219512195124,"warning":null},{"id":"epoch-run-LHRL6nQJZXGJmYppR7kbGw","modelId":"llama-3.2-90b-vision-instruct","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"Llama-3.2-90B-Vision-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=Llama-3.2-90B-Vision-Instruct; run=LHRL6nQJZXGJmYppR7kbGw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":2.638888888888889,"warning":null},{"id":"epoch-run-LiRiryJsGb3h5UuxX8bmJv","modelId":"deepseek-r1-0528","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"DeepSeek-R1-0528","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=DeepSeek-R1-0528; run=LiRiryJsGb3h5UuxX8bmJv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":76.32575757575758,"warning":null},{"id":"epoch-run-LjgJoDosTLkiGVT59smA7q","modelId":"kimi-k2.6","benchmarkId":"epoch-gpqa-diamond-1-0-8-choice","level":"unspecified","sourceLabel":"kimi-k2.6","configuration":"Epoch AI Inspect; task 1.0.8; scorer choice; model=kimi-k2.6; run=LjgJoDosTLkiGVT59smA7q; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.78282828282829,"warning":null},{"id":"epoch-run-LJyKf8Xy4dymxNHbGLwNdA","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"high","sourceLabel":"claude-sonnet-4-6_high","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=claude-sonnet-4-6_high; run=LJyKf8Xy4dymxNHbGLwNdA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":35.5,"warning":null},{"id":"epoch-run-LkVuUxvsHNzpcg3sUiH7f3","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_16K (thinking 16k)","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=claude-sonnet-4-5-20250929_16K; run=LkVuUxvsHNzpcg3sUiH7f3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":78.78787878787878,"warning":null},{"id":"epoch-run-LPo74T9L3CawnKX5LLUrSZ","modelId":"claude-sonnet-5","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"claude-sonnet-5_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-sonnet-5_max; run=LPo74T9L3CawnKX5LLUrSZ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33.7,"warning":null},{"id":"epoch-run-LSKDetgzyVojru45w7RJhf","modelId":"gpt-5.5-pro","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"gpt-5.5-pro_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.5-pro_xhigh; run=LSKDetgzyVojru45w7RJhf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.71929824561403,"warning":null},{"id":"epoch-run-LsnPrukDADRTP8fsn8zD9S","modelId":"claude-opus-4-7","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"unspecified","sourceLabel":"claude-opus-4-7","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=claude-opus-4-7; run=LsnPrukDADRTP8fsn8zD9S; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":13,"warning":null},{"id":"epoch-run-LUfGWqWQim7rdBPgSpbrXS","modelId":"gpt-5.6-luna","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"gpt-5.6-luna_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-luna_max; run=LUfGWqWQim7rdBPgSpbrXS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":43,"warning":null},{"id":"epoch-run-LWbCXNGMXtjQYxyM2uoFMV","modelId":"deepseek-r1-0528-qwen3-8b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"deepseek-r1-0528-qwen3-8b","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=deepseek-r1-0528-qwen3-8b; run=LWbCXNGMXtjQYxyM2uoFMV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":43.888888888888886,"warning":null},{"id":"epoch-run-LWkDtkmd3TNmLzMhfZD5HW","modelId":"gpt-5.6-luna","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"gpt-5.6-luna_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-luna_max; run=LWkDtkmd3TNmLzMhfZD5HW; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.60353535353535,"warning":null},{"id":"epoch-run-m4WQuAN88hKsqbT9zXbg7z","modelId":"grok-4.3","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"grok-4.3_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=grok-4.3_high; run=m4WQuAN88hKsqbT9zXbg7z; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-M8sY8vHnvSLxQUg3HTMTCp","modelId":"gemini-3.6-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"low","sourceLabel":"gemini-3.6-flash_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3.6-flash_low; run=M8sY8vHnvSLxQUg3HTMTCp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":23,"warning":null},{"id":"epoch-run-McJyXLHBzx5VjAqXjmrrsb","modelId":"gemma-3-4b-it","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"gemma-3-4b-it","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemma-3-4b-it; run=McJyXLHBzx5VjAqXjmrrsb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":23.232323232323232,"warning":null},{"id":"epoch-run-MEhV5sh6nbM8cKJGQ6HzcS","modelId":"claude-opus-4-8","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"claude-opus-4-8_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-opus-4-8_low; run=MEhV5sh6nbM8cKJGQ6HzcS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":97.77777777777777,"warning":null},{"id":"epoch-run-mEkZaYpiurQ6TEMAALK3su","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-gpqa-diamond-1-0-2-choice","level":"unspecified","sourceLabel":"claude-opus-4-5-20251101","configuration":"Epoch AI Inspect; task 1.0.2; scorer choice; model=claude-opus-4-5-20251101; run=mEkZaYpiurQ6TEMAALK3su; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80.68181818181817,"warning":null},{"id":"epoch-run-MEwMcmSV32JHrsNoAq4BFL","modelId":"llama-3.1-70b-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"Llama-3.1-70B-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=Llama-3.1-70B-Instruct; run=MEwMcmSV32JHrsNoAq4BFL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":44.19191919191919,"warning":null},{"id":"epoch-run-MFTd4BUbtzbfcCky6d79d4","modelId":"gpt-5.6-terra","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"gpt-5.6-terra_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-terra_max; run=MFTd4BUbtzbfcCky6d79d4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":35,"warning":null},{"id":"epoch-run-Mg2yb9Tp3yc2H3ApJX6JBK","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"low","sourceLabel":"gemini-3.5-flash-lite_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3.5-flash-lite_low; run=Mg2yb9Tp3yc2H3ApJX6JBK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19,"warning":null},{"id":"epoch-run-mGXYSUNCMisvaRzhQp6g2L","modelId":"qwen3.5-plus","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.5-plus","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-plus; run=mGXYSUNCMisvaRzhQp6g2L; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.84848484848484,"warning":null},{"id":"epoch-run-MHAMdn52f7wLMZ6MSCs5E2","modelId":"gpt-5.5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"none","sourceLabel":"gpt-5.5_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.5_none; run=MHAMdn52f7wLMZ6MSCs5E2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-MhqeuRVx4JcRfxPQDYPq7Z","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"unspecified","sourceLabel":"claude-sonnet-4-5-20250929","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=claude-sonnet-4-5-20250929; run=MhqeuRVx4JcRfxPQDYPq7Z; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":23.7,"warning":null},{"id":"epoch-run-Mhwme9odWJMzMjoYkW8r6B","modelId":"glm-5.1","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"glm-5.1","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.1; run=Mhwme9odWJMzMjoYkW8r6B; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36.84210526315789,"warning":null},{"id":"epoch-run-Mifz96Jm442bn43Qnxj5rT","modelId":"gpt-5.6-sol","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gpt-5.6-sol_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-sol_low; run=Mifz96Jm442bn43Qnxj5rT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.55555555555556,"warning":null},{"id":"epoch-run-mKhVMEicZeGjdvJdTRd7mK","modelId":"deepseek-v4-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"deepseek-v4-flash-0731_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=deepseek-v4-flash-0731_max; run=mKhVMEicZeGjdvJdTRd7mK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.54385964912281,"warning":null},{"id":"epoch-run-mLLmVVvo6t23MM5yxq7ryW","modelId":"muse-spark-1.2","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"xhigh","sourceLabel":"muse-spark-1.2_xhigh","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=muse-spark-1.2_xhigh; run=mLLmVVvo6t23MM5yxq7ryW; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":60.30150753768844,"warning":null},{"id":"epoch-run-mLmUnBuX6DYKtb6x46GteM","modelId":"claude-opus-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"claude-opus-5","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-opus-5; run=mLmUnBuX6DYKtb6x46GteM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":92.92929292929293,"warning":null},{"id":"epoch-run-mM6vdJpgjLY9prnetXzZJS","modelId":"grok-4.20-0309-reasoning","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"grok-4.20-0309-reasoning","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=grok-4.20-0309-reasoning; run=mM6vdJpgjLY9prnetXzZJS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":92.22222222222223,"warning":null},{"id":"epoch-run-mqsc4vFuKtAHXxbYyX9WwK","modelId":"gemini-3.5-flash-lite","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3.5-flash-lite_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.5-flash-lite_high; run=mqsc4vFuKtAHXxbYyX9WwK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.11111111111111,"warning":null},{"id":"epoch-run-mRAASXsUuoJmc2oR5G89zR","modelId":"claude-fable-5-1","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-fable-5-1_max; run=mRAASXsUuoJmc2oR5G89zR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":100,"warning":null},{"id":"epoch-run-mS5EwCRZVpvUvJEehPUDZu","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-5-20251101_32K; run=mS5EwCRZVpvUvJEehPUDZu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34.385964912280706,"warning":null},{"id":"epoch-run-mstwSH7noJfvhGcd6Lz5kd","modelId":"gemma-4-26b-a4b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"minimal","sourceLabel":"gemma-4-26b-a4b-it_minimal","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemma-4-26b-a4b-it_minimal; run=mstwSH7noJfvhGcd6Lz5kd; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.22222222222221,"warning":null},{"id":"epoch-run-MsXAXDEKP3xBKtiA7aTBBG","modelId":"llama-4-scout","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"Llama-4-Scout-17B-16E-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=Llama-4-Scout-17B-16E-Instruct; run=MsXAXDEKP3xBKtiA7aTBBG; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":51.830808080808076,"warning":null},{"id":"epoch-run-MvdunHkVURna3JuyDjhzEc","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"high","sourceLabel":"gemini-3-flash-preview_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3-flash-preview_high; run=MvdunHkVURna3JuyDjhzEc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20,"warning":null},{"id":"epoch-run-mvNb3hNgSZ2gS6jHMtvCdk","modelId":"kimi-k3","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"kimi-k3_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=kimi-k3_low; run=mvNb3hNgSZ2gS6jHMtvCdk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.84848484848484,"warning":null},{"id":"epoch-run-MVZf59jt9uEoWG9w5W9apB","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.1-flash-lite_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.1-flash-lite_high; run=MVZf59jt9uEoWG9w5W9apB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27.719298245614034,"warning":null},{"id":"epoch-run-mZLWt2JSYvJyGhMvji7CT8","modelId":"kimi-k3","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"kimi-k3_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=kimi-k3_max; run=mZLWt2JSYvJyGhMvji7CT8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":50.6,"warning":null},{"id":"epoch-run-n4hyz75SVpwPEisKVVjvNg","modelId":"glm-5","benchmarkId":"epoch-gpqa-diamond-1-0-5-choice","level":"unspecified","sourceLabel":"glm-5","configuration":"Epoch AI Inspect; task 1.0.5; scorer choice; model=glm-5; run=n4hyz75SVpwPEisKVVjvNg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.81725888324873,"warning":null},{"id":"epoch-run-n9gpXfUzo8SbWqvrG2q4xM","modelId":"mistral-large-2411","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"mistral-large-2411","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=mistral-large-2411; run=n9gpXfUzo8SbWqvrG2q4xM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":51.32575757575758,"warning":null},{"id":"epoch-run-N9uFn7EZFsvHNcHEwEy3J6","modelId":"gpt-6-sol","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-6-sol_max; run=N9uFn7EZFsvHNcHEwEy3J6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.00000000000001,"warning":null},{"id":"epoch-run-nb9LnDeeQyvRCqsbsTANGD","modelId":"gemini-3.6-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.6-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.6-flash_high; run=nb9LnDeeQyvRCqsbsTANGD; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":58.94736842105262,"warning":null},{"id":"epoch-run-nBcLJJomVoeg2zh5F8Mnsu","modelId":"qwen-3.8-max-0902","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"xhigh","sourceLabel":"qwen3.8-max-0902_xhigh","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=qwen3.8-max-0902_xhigh; run=nBcLJJomVoeg2zh5F8Mnsu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47.3,"warning":null},{"id":"epoch-run-NCikvVWc5SEyMEbCqvVNb2","modelId":"glm-5.3-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"glm-5.3-flash_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-5.3-flash_max; run=NCikvVWc5SEyMEbCqvVNb2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.88888888888889,"warning":null},{"id":"epoch-run-nDsNFM5ioGzyWcpeUKVotg","modelId":"qwen3.6-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.6-flash","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-flash; run=nDsNFM5ioGzyWcpeUKVotg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17.192982456140353,"warning":null},{"id":"epoch-run-nE4hmJo3QV6X5jeGWAeAze","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gemini-3.1-flash-lite_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.1-flash-lite_high; run=nE4hmJo3QV6X5jeGWAeAze; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":81.81818181818183,"warning":null},{"id":"epoch-run-nEJGtXGN588zynPkynZkxz","modelId":"deepseek-v4-pro","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"deepseek-v4-pro-0813_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=deepseek-v4-pro-0813_max; run=nEJGtXGN588zynPkynZkxz; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":91.66666666666666,"warning":null},{"id":"epoch-run-NeYuqgQyWtFWAHEjQYPm5Z","modelId":"gemini-3.8-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3.8-flash_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.8-flash_high; run=NeYuqgQyWtFWAHEjQYPm5Z; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":98.88888888888889,"warning":null},{"id":"epoch-run-neZvEzLi6yeL9z5KjGef7H","modelId":"claude-haiku-4-5-20251001","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-extracted-exact-match","level":"thinking","sourceLabel":"claude-haiku-4-5-20251001_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_extracted_exact_match; model=claude-haiku-4-5-20251001_32K; run=neZvEzLi6yeL9z5KjGef7H; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.66666666666666,"warning":null},{"id":"epoch-run-NfN7VwMPjSVqViqrBQPWk2","modelId":"glm-5.2","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"medium","sourceLabel":"glm-5.2_medium","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=glm-5.2_medium; run=NfN7VwMPjSVqViqrBQPWk2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":15,"warning":null},{"id":"epoch-run-nhim2QCRgpir5cZdqrmUz3","modelId":"gemini-3.1-pro","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3.1-pro-preview_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.1-pro-preview_high; run=nhim2QCRgpir5cZdqrmUz3; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":95.55555555555556,"warning":null},{"id":"epoch-run-NiuExVccWkWbWqi6ex6vvf","modelId":"grok-4.6","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"grok-4.6_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=grok-4.6_xhigh; run=NiuExVccWkWbWqi6ex6vvf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33,"warning":null},{"id":"epoch-run-njpkLAWZTvept6tnkraYrR","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"minimal","sourceLabel":"gemini-3-flash-preview_minimal","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3-flash-preview_minimal; run=njpkLAWZTvept6tnkraYrR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25,"warning":null},{"id":"epoch-run-NLiiJJSyA9wmg2FHENFpAq","modelId":"gpt-5.6-luna","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-5.6-luna_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-luna_max; run=NLiiJJSyA9wmg2FHENFpAq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":60.97560975609756,"warning":null},{"id":"epoch-run-NMdaAXyqqPRmWfuxZUfHFF","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"gemini-3.1-flash-lite_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gemini-3.1-flash-lite_high; run=NMdaAXyqqPRmWfuxZUfHFF; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80,"warning":null},{"id":"epoch-run-nrh3xa3Bn765Q6DXWoVfbS","modelId":"gemini-3.5-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"low","sourceLabel":"gemini-3.5-flash_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3.5-flash_low; run=nrh3xa3Bn765Q6DXWoVfbS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.000000000000004,"warning":null},{"id":"epoch-run-nSb2focDRMyPJXErk3MV94","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-2-model-extracted-exact-match","level":"unspecified","sourceLabel":"claude-opus-4-5-20251101","configuration":"Epoch AI Inspect; task 1.0.2; scorer model_extracted_exact_match; model=claude-opus-4-5-20251101; run=nSb2focDRMyPJXErk3MV94; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":48.05555555555556,"warning":null},{"id":"epoch-run-nuKCgZReCBsXKCkNj9bK37","modelId":"gpt-5.6-sol","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-sol_max; run=nuKCgZReCBsXKCkNj9bK37; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.49747474747475,"warning":null},{"id":"epoch-run-nW39ZzuDQ3xTAGVoEfFG95","modelId":"gpt-5.6-luna","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"gpt-5.6-luna_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gpt-5.6-luna_low; run=nW39ZzuDQ3xTAGVoEfFG95; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.32323232323232,"warning":null},{"id":"epoch-run-NyAvSf73nUe84HuY6Vpjkh","modelId":"claude-opus-4-7","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"claude-opus-4-7_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-7_low; run=NyAvSf73nUe84HuY6Vpjkh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26,"warning":null},{"id":"epoch-run-nzQV9YvictnDF2EwxqTJQs","modelId":"kimi-k3","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"kimi-k3_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=kimi-k3_max; run=nzQV9YvictnDF2EwxqTJQs; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36,"warning":null},{"id":"epoch-run-o5th8Rb9WiaDbv4oEMDoFa","modelId":"qwen3.7-plus","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.7-plus_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.7-plus_none; run=o5th8Rb9WiaDbv4oEMDoFa; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80,"warning":null},{"id":"epoch-run-o6migB5WP27uLr9nUWpCiM","modelId":"qwen3.5-9b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.5-9B","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-9B; run=o6migB5WP27uLr9nUWpCiM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":78.97727272727273,"warning":null},{"id":"epoch-run-o6VtA8qruLpoShhPy2JdLg","modelId":"qwen3.6-27b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.6-27b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.6-27b_none; run=o6VtA8qruLpoShhPy2JdLg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.66666666666666,"warning":null},{"id":"epoch-run-oAu9CZSNPZgwiwKnoNpM8Q","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-gpqa-diamond-1-0-3-choice","level":"unspecified","sourceLabel":"gemini-3-flash-preview","configuration":"Epoch AI Inspect; task 1.0.3; scorer choice; model=gemini-3-flash-preview; run=oAu9CZSNPZgwiwKnoNpM8Q; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.20707070707071,"warning":null},{"id":"epoch-run-oWqwMmwEd7LU3eU5Ftak64","modelId":"gpt-5.6-sol","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-sol_max; run=oWqwMmwEd7LU3eU5Ftak64; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":59,"warning":null},{"id":"epoch-run-oXZ7ZVZtADrwFPKAe6zTvL","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"high","sourceLabel":"claude-sonnet-4-6_high","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-sonnet-4-6_high; run=oXZ7ZVZtADrwFPKAe6zTvL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":75.55555555555556,"warning":null},{"id":"epoch-run-P4LXQ9PyDDziEANMB5jVRd","modelId":"claude-opus-4-7","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-4-7_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-7_max; run=P4LXQ9PyDDziEANMB5jVRd; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":31.70731707317073,"warning":null},{"id":"epoch-run-P5L8qbXo42vTvbU3UvTNjk","modelId":"gemma-4-31b-it","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"minimal","sourceLabel":"gemma-4-31b-it_minimal","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemma-4-31b-it_minimal; run=P5L8qbXo42vTvbU3UvTNjk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":75.75757575757575,"warning":null},{"id":"epoch-run-PBRdMZyz4L4CqPpXyLD2Lk","modelId":"qwen-3.8-max","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"qwen3.8-max_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.8-max_xhigh; run=PBRdMZyz4L4CqPpXyLD2Lk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":46.34146341463415,"warning":null},{"id":"epoch-run-PBrQpAgRCXeUSNveEn4aDE","modelId":"gpt-5.6-luna","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"none","sourceLabel":"gpt-5.6-luna_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-luna_none; run=PBrQpAgRCXeUSNveEn4aDE; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20,"warning":null},{"id":"epoch-run-Pc6CyTEYKbcdDUDKzGZnS9","modelId":"gemma-3-1b-it","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"gemma-3-1b-it","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemma-3-1b-it; run=Pc6CyTEYKbcdDUDKzGZnS9; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19.94949494949495,"warning":null},{"id":"epoch-run-PEZpuEuU65bu3ZtvjLytHq","modelId":"claude-opus-4-8","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-opus-4-8_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-8_max; run=PEZpuEuU65bu3ZtvjLytHq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":52,"warning":null},{"id":"epoch-run-PKSMspJiJdNThB6qNvaBSM","modelId":"gpt-5.5","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"gpt-5.5_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.5_xhigh; run=PKSMspJiJdNThB6qNvaBSM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":72.5,"warning":null},{"id":"epoch-run-PN9exWpeJGXJB8tZGX32e4","modelId":"claude-sonnet-5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-sonnet-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-5_max; run=PN9exWpeJGXJB8tZGX32e4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45,"warning":null},{"id":"epoch-run-PppwnX7EjXx2SeTcFN8Ujn","modelId":"gpt-6-astra","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gpt-6-astra_max; run=PppwnX7EjXx2SeTcFN8Ujn; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":75.6,"warning":null},{"id":"epoch-run-PtxHF8BH8WuNdNfDnutGBb","modelId":"qwq-plus","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"qwq-plus","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=qwq-plus; run=PtxHF8BH8WuNdNfDnutGBb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":65.40404040404042,"warning":null},{"id":"epoch-run-Q3WpHdurYNvawj6ycXyj6e","modelId":"claude-opus-4-8","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"claude-opus-4-8_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-opus-4-8_max; run=Q3WpHdurYNvawj6ycXyj6e; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36,"warning":null},{"id":"epoch-run-QDZRoAVaEDtdS6nECEeE35","modelId":"gemini-3.5-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"minimal","sourceLabel":"gemini-3.5-flash_minimal","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.5-flash_minimal; run=QDZRoAVaEDtdS6nECEeE35; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.36363636363636,"warning":null},{"id":"epoch-run-QfS2PrjUFMNnceRZmCmewP","modelId":"gpt-5.5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"none","sourceLabel":"gpt-5.5_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.5_none; run=QfS2PrjUFMNnceRZmCmewP; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":9,"warning":null},{"id":"epoch-run-QHfAMMDjjzQF6xitN2vBbA","modelId":"gpt-5.5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"low","sourceLabel":"gpt-5.5_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.5_low; run=QHfAMMDjjzQF6xitN2vBbA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.000000000000004,"warning":null},{"id":"epoch-run-QJ5rJMVypB4PMxPcBGftc8","modelId":"gpt-5.6-sol","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-sol_max; run=QJ5rJMVypB4PMxPcBGftc8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.92682926829268,"warning":null},{"id":"epoch-run-QuqPmMUTbiEsUvoFYZx5Mk","modelId":"gpt-5.5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"none","sourceLabel":"gpt-5.5_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.5_none; run=QuqPmMUTbiEsUvoFYZx5Mk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":57.77777777777777,"warning":null},{"id":"epoch-run-Qw4tGw38pk4sdsEehhrLSR","modelId":"claude-sonnet-5","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"xhigh","sourceLabel":"claude-sonnet-5_xhigh","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-sonnet-5_xhigh; run=Qw4tGw38pk4sdsEehhrLSR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32.9,"warning":null},{"id":"epoch-run-QWw65xeTn8x42SymMsVREN","modelId":"mistral-medium-2505","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"mistral-medium-2505","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=mistral-medium-2505; run=QWw65xeTn8x42SymMsVREN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32.22222222222222,"warning":null},{"id":"epoch-run-QYZ6wRuT7WMkaYPKc9dRPy","modelId":"gpt-5.6-luna","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"none","sourceLabel":"gpt-5.6-luna_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-luna_none; run=QYZ6wRuT7WMkaYPKc9dRPy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":40,"warning":null},{"id":"epoch-run-R2fMFAtRXah864hMifdQGV","modelId":"glm-5.1","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"unspecified","sourceLabel":"glm-5.1","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=glm-5.1; run=R2fMFAtRXah864hMifdQGV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34,"warning":null},{"id":"epoch-run-R7XgJMi2R9sv7tZdx5SZTv","modelId":"glm-5.1","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"glm-5.1","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=glm-5.1; run=R7XgJMi2R9sv7tZdx5SZTv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20,"warning":null},{"id":"epoch-run-RA5MRbmWJpFPR29BQxfvZt","modelId":"kimi-k2.6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"kimi-k2.6","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=kimi-k2.6; run=RA5MRbmWJpFPR29BQxfvZt; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-RAwuPRwJedy7P2KeQhEmr8","modelId":"phi-4","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"phi-4","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=phi-4; run=RAwuPRwJedy7P2KeQhEmr8; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":56.060606060606055,"warning":null},{"id":"epoch-run-ReDAtJGGfRtrEGGWf4Zbsd","modelId":"claude-fable-5","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-fable-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-fable-5_max; run=ReDAtJGGfRtrEGGWf4Zbsd; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.2,"warning":null},{"id":"epoch-run-RG2biwnmGybSVTVxxMdRWa","modelId":"qwen3.7-max","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.7-max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.7-max; run=RG2biwnmGybSVTVxxMdRWa; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":64.56140350877193,"warning":null},{"id":"epoch-run-RHbNnWqK6x8f6pR2AZ8VAq","modelId":"claude-opus-4-7","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"claude-opus-4-7_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-7_xhigh; run=RHbNnWqK6x8f6pR2AZ8VAq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36,"warning":null},{"id":"epoch-run-RHjHEPHEsWZqFuMjBTeC6r","modelId":"gpt-6-sol","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-6-sol_max; run=RHjHEPHEsWZqFuMjBTeC6r; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":89.82456140350877,"warning":null},{"id":"epoch-run-RLN98bmQSMnc5Zv39BF4M2","modelId":"claude-opus-5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-opus-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-5_max; run=RLN98bmQSMnc5Zv39BF4M2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":43,"warning":null},{"id":"epoch-run-RMcys99fv45vvgTgakXXWn","modelId":"qwen3-14b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3-14b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3-14b_none; run=RMcys99fv45vvgTgakXXWn; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25.833333333333336,"warning":null},{"id":"epoch-run-RMeUVa4tq3fURqCMtQWkfe","modelId":"gpt-5.6-sol","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"low","sourceLabel":"gpt-5.6-sol_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-sol_low; run=RMeUVa4tq3fURqCMtQWkfe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26,"warning":null},{"id":"epoch-run-RoscWx4NUGVqcHdko6QhKd","modelId":"qwen3.6-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.6-flash_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-flash_none; run=RoscWx4NUGVqcHdko6QhKd; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":22.45614035087719,"warning":null},{"id":"epoch-run-RuQX7KA7owzqzf8Le9veeM","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_59K (thinking 59k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-4-5-20250929_59K; run=RuQX7KA7owzqzf8Le9veeM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":10,"warning":null},{"id":"epoch-run-RWjuCuVGvaRd4vDjJJ6aP2","modelId":"qwen3.5-35b-a3b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.5-35b-a3b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-35b-a3b; run=RWjuCuVGvaRd4vDjJJ6aP2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":83.45959595959596,"warning":null},{"id":"epoch-run-S5pm7ctk9zgPfbyHrbD9AV","modelId":"claude-fable-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"claude-fable-5_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-fable-5_max; run=S5pm7ctk9zgPfbyHrbD9AV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":85.85858585858585,"warning":null},{"id":"epoch-run-S5QYXSvQBRSbUbXSnAGbMm","modelId":"llama-3.1-405b-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"Llama-3.1-405B-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=Llama-3.1-405B-Instruct; run=S5QYXSvQBRSbUbXSnAGbMm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":50.91540404040404,"warning":null},{"id":"epoch-run-S7wbot7JR53BAjM8F9gaws","modelId":"qwen3.5-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.5-flash","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.5-flash; run=S7wbot7JR53BAjM8F9gaws; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.32323232323232,"warning":null},{"id":"epoch-run-SAZy8rk2HgyPjA8XZJFUmR","modelId":"qwen3.5-4b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.5-4b_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-4b_none; run=SAZy8rk2HgyPjA8XZJFUmR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":55.833333333333336,"warning":null},{"id":"epoch-run-SB6Qj6rCBDk8jZmFcCTzWc","modelId":"glm-5.2","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"glm-5.2_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-5.2_low; run=SB6Qj6rCBDk8jZmFcCTzWc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.87878787878788,"warning":null},{"id":"epoch-run-SEePRKRcv8oqn5HpD2KFxB","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"max","sourceLabel":"claude-sonnet-4-6_max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-sonnet-4-6_max; run=SEePRKRcv8oqn5HpD2KFxB; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":78.78787878787878,"warning":null},{"id":"epoch-run-SEnvgtunRa5GwAtU2sHa87","modelId":"claude-fable-5","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"xhigh","sourceLabel":"claude-fable-5_xhigh","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=claude-fable-5_xhigh; run=SEnvgtunRa5GwAtU2sHa87; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70.7,"warning":null},{"id":"epoch-run-SFKXCPUu73mwxMoUotiZM2","modelId":"gpt-5.5-pro","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"gpt-5.5-pro_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.5-pro_xhigh; run=SFKXCPUu73mwxMoUotiZM2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":78.04878048780488,"warning":null},{"id":"epoch-run-SH5jr76ni3RnpXVMETMy2H","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-2-model-extracted-exact-match","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_16K (thinking 16k)","configuration":"Epoch AI Inspect; task 1.0.2; scorer model_extracted_exact_match; model=claude-opus-4-5-20251101_16K; run=SH5jr76ni3RnpXVMETMy2H; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":81.66666666666667,"warning":null},{"id":"epoch-run-SLJyD8yew4TxcmprJfVvuE","modelId":"qwen3-4b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3-4b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-4b_none; run=SLJyD8yew4TxcmprJfVvuE; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":43.56060606060606,"warning":null},{"id":"epoch-run-SP9UFPzxVnsbzs2kExP2bh","modelId":"qwen3.6-plus","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"qwen3.6-plus","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=qwen3.6-plus; run=SP9UFPzxVnsbzs2kExP2bh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":14.000000000000002,"warning":null},{"id":"epoch-run-STxFZiSPAPScUuENvXBWLc","modelId":"grok-4.20-0309-reasoning","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"grok-4.20-0309-reasoning","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=grok-4.20-0309-reasoning; run=STxFZiSPAPScUuENvXBWLc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":89.33080808080808,"warning":null},{"id":"epoch-run-SX5FCWMk3HF83gGsP3WFA6","modelId":"grok-4.20-0309-reasoning","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"grok-4.20-0309-reasoning","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.20-0309-reasoning; run=SX5FCWMk3HF83gGsP3WFA6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17.073170731707318,"warning":null},{"id":"epoch-run-SZDgstFQJAGeGQwMXgNx6w","modelId":"qwen3.7-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.7-flash_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.7-flash_none; run=SZDgstFQJAGeGQwMXgNx6w; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":77.77777777777779,"warning":null},{"id":"epoch-run-T2wFfuSkUdyjpyqurxNmNK","modelId":"claude-sonnet-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"xhigh","sourceLabel":"claude-sonnet-5_xhigh","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-sonnet-5_xhigh; run=T2wFfuSkUdyjpyqurxNmNK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.72222222222221,"warning":null},{"id":"epoch-run-T3X5DeC47iqCaoRYXu5k8L","modelId":"llama-4-scout","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"Llama-4-Scout-17B-16E-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=Llama-4-Scout-17B-16E-Instruct; run=T3X5DeC47iqCaoRYXu5k8L; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.777777777777778,"warning":null},{"id":"epoch-run-T4y4iGHMqA8tXCdBKr5GjS","modelId":"claude-opus-4-7","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"claude-opus-4-7_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-opus-4-7_max; run=T4y4iGHMqA8tXCdBKr5GjS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.66666666666667,"warning":null},{"id":"epoch-run-T56N3X4KLA3xz7tf5aCMGb","modelId":"claude-fable-5-1","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-fable-5-1_max; run=T56N3X4KLA3xz7tf5aCMGb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.8048780487805,"warning":null},{"id":"epoch-run-TEVNDC9QXBh7PQvX9kHZip","modelId":"qwen-3.8-max","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"xhigh","sourceLabel":"qwen3.8-max_xhigh","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.8-max_xhigh; run=TEVNDC9QXBh7PQvX9kHZip; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":92.67676767676768,"warning":null},{"id":"epoch-run-TijCafTTJUFrJ6jPGvBbgq","modelId":"gpt-5.5","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"xhigh","sourceLabel":"gpt-5.5_xhigh","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=gpt-5.5_xhigh; run=TijCafTTJUFrJ6jPGvBbgq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":63,"warning":null},{"id":"epoch-run-TjoS7sNuqVTc4LVyGRo4wa","modelId":"qwen3.7-flash","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"qwen3.7-flash","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.7-flash; run=TjoS7sNuqVTc4LVyGRo4wa; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.66666666666667,"warning":null},{"id":"epoch-run-TMM3rPrAHcpBG5Pvw9kr4L","modelId":"glm-5.3-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"glm-5.3-flash_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=glm-5.3-flash_max; run=TMM3rPrAHcpBG5Pvw9kr4L; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":8,"warning":null},{"id":"epoch-run-TpHYtuPG6kYZFvMK7Z6Hqj","modelId":"grok-4.6","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"grok-4.6_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=grok-4.6_high; run=TpHYtuPG6kYZFvMK7Z6Hqj; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":49.3,"warning":null},{"id":"epoch-run-TqFmiGvqozt7TSZYmhgF8m","modelId":"deepseek-v4-flash","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"deepseek-v4-flash-0731_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=deepseek-v4-flash-0731_max; run=TqFmiGvqozt7TSZYmhgF8m; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33.633633633633636,"warning":null},{"id":"epoch-run-Tt8tBXHoCEYqGLePWXkM8u","modelId":"qwen3.7-plus","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.7-plus_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.7-plus_none; run=Tt8tBXHoCEYqGLePWXkM8u; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34.385964912280706,"warning":null},{"id":"epoch-run-TTuCPacMDpNaCSs3Hpbdjp","modelId":"glm-5.2","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"glm-5.2_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.2_max; run=TTuCPacMDpNaCSs3Hpbdjp; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":59.205776173285194,"warning":null},{"id":"epoch-run-TUPC8VdX4jApFa4MGfNP4t","modelId":"claude-opus-4-6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-4-model-extracted-exact-match","level":"thinking","sourceLabel":"claude-opus-4-6_64K (thinking 64k)","configuration":"Epoch AI Inspect; task 1.0.4; scorer model_extracted_exact_match; model=claude-opus-4-6_64K; run=TUPC8VdX4jApFa4MGfNP4t; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.44444444444444,"warning":null},{"id":"epoch-run-TxXS78Wg2pSCDbpmeJuzQn","modelId":"llama-3.3-70b-instruct","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"Llama-3.3-70B-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=Llama-3.3-70B-Instruct; run=TxXS78Wg2pSCDbpmeJuzQn; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":47.44318181818182,"warning":null},{"id":"epoch-run-TY8R8FV7zCwByEipkqyg2P","modelId":"glm-5.3","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"glm-5.3_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=glm-5.3_max; run=TY8R8FV7zCwByEipkqyg2P; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":41,"warning":null},{"id":"epoch-run-U5PsUdX2ak4kJ2hgnQkWnX","modelId":"claude-fable-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"claude-fable-5_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-fable-5_low; run=U5PsUdX2ak4kJ2hgnQkWnX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":78.78787878787878,"warning":null},{"id":"epoch-run-U8H8juE3Ko83jsR6TRntbh","modelId":"nemotron-3-ultra","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"unspecified","sourceLabel":"nemotron-3-ultra","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=nemotron-3-ultra; run=U8H8juE3Ko83jsR6TRntbh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20,"warning":null},{"id":"epoch-run-Ua5RK6JRxdyrHRa3hVFQwq","modelId":"qwen3.6-35b-a3b","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.6-35b-a3b_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.6-35b-a3b_none; run=Ua5RK6JRxdyrHRa3hVFQwq; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":22,"warning":null},{"id":"epoch-run-UnKHyc2idtTmoTUGxGeNJV","modelId":"glm-5.2","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"low","sourceLabel":"glm-5.2_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.2_low; run=UnKHyc2idtTmoTUGxGeNJV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":54.736842105263165,"warning":null},{"id":"epoch-run-UnRUSuZzm2X27TUbpeMPva","modelId":"gpt-5.6-sol","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=gpt-5.6-sol_max; run=UnRUSuZzm2X27TUbpeMPva; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":69.69999999999999,"warning":null},{"id":"epoch-run-URoWjw9Htc6QPwABVcrybw","modelId":"qwen3.7-plus","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.7-plus","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.7-plus; run=URoWjw9Htc6QPwABVcrybw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.87878787878788,"warning":null},{"id":"epoch-run-Usx9KaiXvHZU4qrzgc4sLg","modelId":"gemma-4-26b-a4b-it","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"minimal","sourceLabel":"gemma-4-26b-a4b-it_minimal","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemma-4-26b-a4b-it_minimal; run=Usx9KaiXvHZU4qrzgc4sLg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":73.23232323232324,"warning":null},{"id":"epoch-run-Ut4d4rvkUygdLTRomFb5rh","modelId":"qwen-3.8-max-0902","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"xhigh","sourceLabel":"qwen3.8-max-0902_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=qwen3.8-max-0902_xhigh; run=Ut4d4rvkUygdLTRomFb5rh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":37,"warning":null},{"id":"epoch-run-Utgs6DuaQDDdi98ZWH22qL","modelId":"gemini-3.8-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"high","sourceLabel":"gemini-3.8-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.8-flash_high; run=Utgs6DuaQDDdi98ZWH22qL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":68.42105263157895,"warning":null},{"id":"epoch-run-UTVSgirxQpTgmTHbgE9EUk","modelId":"kimi-k2.7-code","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"kimi-k2.7-code","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=kimi-k2.7-code; run=UTVSgirxQpTgmTHbgE9EUk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12.195121951219512,"warning":null},{"id":"epoch-run-UUNZJtgib9XbCjLEdhtGaT","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"gemini-3-flash-preview_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3-flash-preview_high; run=UUNZJtgib9XbCjLEdhtGaT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":39,"warning":null},{"id":"epoch-run-UUq99Qhq3qizcoJo5vdhuK","modelId":"glm-5.3","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"glm-5.3_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=glm-5.3_max; run=UUq99Qhq3qizcoJo5vdhuK; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33,"warning":null},{"id":"epoch-run-V6aVGtn9gAmRep2oBGsqVN","modelId":"gpt-5.6-sol","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"gpt-5.6-sol_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-sol_low; run=V6aVGtn9gAmRep2oBGsqVN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":31,"warning":null},{"id":"epoch-run-V6YuGpcu8MCbJNXKoJMA6b","modelId":"claude-sonnet-5","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-sonnet-5_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-sonnet-5_max; run=V6YuGpcu8MCbJNXKoJMA6b; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":65.6140350877193,"warning":null},{"id":"epoch-run-V7seJ9yeHu4LYjrkomhEQm","modelId":"claude-opus-4-5-20251101","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"thinking","sourceLabel":"claude-opus-4-5-20251101_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-5-20251101_32K; run=V7seJ9yeHu4LYjrkomhEQm; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":5,"warning":null},{"id":"epoch-run-VajBDnUCyX7uqAfMTbi63j","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"claude-sonnet-4-5-20250929","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-sonnet-4-5-20250929; run=VajBDnUCyX7uqAfMTbi63j; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":6,"warning":null},{"id":"epoch-run-VBpT2WWYKBZtUxG9Z3svGh","modelId":"gpt-5.6-sol","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-sol_max; run=VBpT2WWYKBZtUxG9Z3svGh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":89.12280701754386,"warning":null},{"id":"epoch-run-VcuyQvF8LWuYhQCx2FGDHh","modelId":"gemini-3.1-flash-lite","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"low","sourceLabel":"gemini-3.1-flash-lite_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3.1-flash-lite_low; run=VcuyQvF8LWuYhQCx2FGDHh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":22.45614035087719,"warning":null},{"id":"epoch-run-VFYhE54autgdYcMJNXicwU","modelId":"claude-opus-4-7","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"claude-opus-4-7_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-opus-4-7_max; run=VFYhE54autgdYcMJNXicwU; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":28.000000000000004,"warning":null},{"id":"epoch-run-VGeF8Xac4WcSAGgSXVdEBi","modelId":"gemini-3.1-pro","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"high","sourceLabel":"gemini-3.1-pro-preview_high","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=gemini-3.1-pro-preview_high; run=VGeF8Xac4WcSAGgSXVdEBi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":73.5,"warning":null},{"id":"epoch-run-VhBM9HgJ2RuSVwXriRXMVJ","modelId":"claude-haiku-4-5-20251001","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"thinking","sourceLabel":"claude-haiku-4-5-20251001_32K (thinking 32k)","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-haiku-4-5-20251001_32K; run=VhBM9HgJ2RuSVwXriRXMVJ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":12.6,"warning":null},{"id":"epoch-run-VhQBa9AViRt2UKSNpEAyjN","modelId":"qwen3-4b-instruct-2507","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3-4b-instruct-2507","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-4b-instruct-2507; run=VhQBa9AViRt2UKSNpEAyjN; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45.83333333333333,"warning":null},{"id":"epoch-run-ViC6pxQZUVS3E9M9iEubF7","modelId":"claude-opus-4-8","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"claude-opus-4-8_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-opus-4-8_max; run=ViC6pxQZUVS3E9M9iEubF7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":53,"warning":null},{"id":"epoch-run-VJezHrJ5YPCbpui8pZgcJg","modelId":"claude-opus-4-7","benchmarkId":"epoch-gpqa-diamond-1-0-7-choice","level":"xhigh","sourceLabel":"claude-opus-4-7_xhigh","configuration":"Epoch AI Inspect; task 1.0.7; scorer choice; model=claude-opus-4-7_xhigh; run=VJezHrJ5YPCbpui8pZgcJg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.15151515151516,"warning":null},{"id":"epoch-run-VjfWxdhdNerrh4huSHT8PT","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"unspecified","sourceLabel":"claude-sonnet-4-6","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=claude-sonnet-4-6; run=VjfWxdhdNerrh4huSHT8PT; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":14.000000000000002,"warning":null},{"id":"epoch-run-VonoEhAsGSJ2FHU33qnhui","modelId":"qwen3.6-27b","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"qwen3.6-27b","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-27b; run=VonoEhAsGSJ2FHU33qnhui; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":35.08771929824561,"warning":null},{"id":"epoch-run-VPSX7EnwWuBrCPnpYGMUEM","modelId":"kimi-k3","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"kimi-k3_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=kimi-k3_low; run=VPSX7EnwWuBrCPnpYGMUEM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":18,"warning":null},{"id":"epoch-run-Vw7spGUnJyf3vuzSjTcj2F","modelId":"grok-4.3","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"grok-4.3_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=grok-4.3_high; run=Vw7spGUnJyf3vuzSjTcj2F; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.82575757575758,"warning":null},{"id":"epoch-run-VZAUcNkzeMSMm9paxBhj6B","modelId":"gpt-6-sol","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-6-sol_max; run=VZAUcNkzeMSMm9paxBhj6B; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":100,"warning":null},{"id":"epoch-run-VzeQXKBH9maR2nDcZJ8b7S","modelId":"qwen-3.8-max-0902","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"xhigh","sourceLabel":"qwen3.8-max-0902_xhigh","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.8-max-0902_xhigh; run=VzeQXKBH9maR2nDcZJ8b7S; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":92.2979797979798,"warning":null},{"id":"epoch-run-W6mTanyjvXhasFBzHukv4S","modelId":"qwen3.7-max","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3.7-max","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.7-max; run=W6mTanyjvXhasFBzHukv4S; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":90.9090909090909,"warning":null},{"id":"epoch-run-WcCQQSHTFAhXk6qCWEeSLh","modelId":"gpt-5.6-terra","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"medium","sourceLabel":"gpt-5.6-terra_medium","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gpt-5.6-terra_medium; run=WcCQQSHTFAhXk6qCWEeSLh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":10,"warning":null},{"id":"epoch-run-WeP7Q27N5PzxSgg4ybgijX","modelId":"qwen3.6-27b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3.6-27b_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.6-27b_none; run=WeP7Q27N5PzxSgg4ybgijX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":84.84848484848484,"warning":null},{"id":"epoch-run-WfaJS4L7Xdz5ZBXgJMbRGb","modelId":"claude-opus-5","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"low","sourceLabel":"claude-opus-5_low","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=claude-opus-5_low; run=WfaJS4L7Xdz5ZBXgJMbRGb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.87878787878788,"warning":null},{"id":"epoch-run-WgjFZzojpiv54XqUhvfuUG","modelId":"glm-5.2","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"glm-5.2_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=glm-5.2_max; run=WgjFZzojpiv54XqUhvfuUG; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34.2,"warning":null},{"id":"epoch-run-WjXdTPMnJUAT27azUKo3Ka","modelId":"qwen3.6-27b","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.6-27b_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-27b_none; run=WjXdTPMnJUAT27azUKo3Ka; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":34.03508771929825,"warning":null},{"id":"epoch-run-WNE7cveVN8rsi7DmsFsFVy","modelId":"gemma-2-9b-it","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"gemma-2-9b-it","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=gemma-2-9b-it; run=WNE7cveVN8rsi7DmsFsFVy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":0.5555555555555556,"warning":null},{"id":"epoch-run-Ws3g7xtto2fZ7AhuXJfecf","modelId":"minimax-m3","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"non-reasoning","sourceLabel":"MiniMax-M3_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=MiniMax-M3_none; run=Ws3g7xtto2fZ7AhuXJfecf; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.000000000000001,"warning":null},{"id":"epoch-run-WueZbBM3NwMUZPveuagrZo","modelId":"gemini-3.1-pro","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"medium","sourceLabel":"gemini-3.1-pro-preview_medium","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3.1-pro-preview_medium; run=WueZbBM3NwMUZPveuagrZo; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32,"warning":null},{"id":"epoch-run-WvV6QREDDXeDEGdrKztDsy","modelId":"gpt-5.6-sol","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"none","sourceLabel":"gpt-5.6-sol_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gpt-5.6-sol_none; run=WvV6QREDDXeDEGdrKztDsy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":7.000000000000001,"warning":null},{"id":"epoch-run-WYs83xB6NwrkMNyQJtuMcQ","modelId":"deepseek-v4-flash","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"deepseek-v4-flash-0731_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=deepseek-v4-flash-0731_max; run=WYs83xB6NwrkMNyQJtuMcQ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":24.390243902439025,"warning":null},{"id":"epoch-run-WYTWr3Xk4qxymhMsm7vRQ2","modelId":"nemotron-3-ultra","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"nemotron-3-ultra","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=nemotron-3-ultra; run=WYTWr3Xk4qxymhMsm7vRQ2; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.66666666666667,"warning":null},{"id":"epoch-run-X29V6KG3aVe5Nv5r3tsLHX","modelId":"claude-fable-5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"claude-fable-5_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-fable-5_max; run=X29V6KG3aVe5Nv5r3tsLHX; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":52,"warning":null},{"id":"epoch-run-X2QNPHFXuWosA6AiuvVtDv","modelId":"kimi-k3","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"kimi-k3_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=kimi-k3_high; run=X2QNPHFXuWosA6AiuvVtDv; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":37,"warning":null},{"id":"epoch-run-X2tQXGELXtEbc2VX6gARd7","modelId":"llama-3.1-70b-instruct","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"Llama-3.1-70B-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=Llama-3.1-70B-Instruct; run=X2tQXGELXtEbc2VX6gARd7; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":3.6111111111111107,"warning":null},{"id":"epoch-run-X3HfncZtnLKUPoUuRc9YsR","modelId":"qwen3-30b-a3b-instruct-2507","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3-30b-a3b-instruct-2507","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-30b-a3b-instruct-2507; run=X3HfncZtnLKUPoUuRc9YsR; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":55.61868686868687,"warning":null},{"id":"epoch-run-X3RxRkt2pBUXYJjAQHsUo6","modelId":"kimi-k2.7-code","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"kimi-k2.7-code","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=kimi-k2.7-code; run=X3RxRkt2pBUXYJjAQHsUo6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":54.03508771929825,"warning":null},{"id":"epoch-run-X5FL4gkJQYjzJ7axz23rfd","modelId":"qwen3.5-122b-a10b","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.5-122b-a10b_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.5-122b-a10b_none; run=X5FL4gkJQYjzJ7axz23rfd; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17,"warning":null},{"id":"epoch-run-XaGuWDxPtqVvyxqdMncKFj","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_59K (thinking 59k)","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=claude-sonnet-4-5-20250929_59K; run=XaGuWDxPtqVvyxqdMncKFj; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.32323232323232,"warning":null},{"id":"epoch-run-XbtWdvRz65dbPPnHcSBvmF","modelId":"qwen3.5-plus","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"non-reasoning","sourceLabel":"qwen3.5-plus_none","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=qwen3.5-plus_none; run=XbtWdvRz65dbPPnHcSBvmF; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17,"warning":null},{"id":"epoch-run-XCSWtbDquA7FUiWfd7kh9q","modelId":"open-mistral-nemo-2407","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"open-mistral-nemo-2407","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=open-mistral-nemo-2407; run=XCSWtbDquA7FUiWfd7kh9q; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":29.892676767676768,"warning":null},{"id":"epoch-run-XEaDaqM3oaPcrtQ6XH5fGS","modelId":"qwen3.5-plus","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"unspecified","sourceLabel":"qwen3.5-plus","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=qwen3.5-plus; run=XEaDaqM3oaPcrtQ6XH5fGS; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":8,"warning":null},{"id":"epoch-run-XfjRkSyP4ub4gDsWcgS7KY","modelId":"qwen3-30b-a3b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"unspecified","sourceLabel":"Qwen3-30B-A3B","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=Qwen3-30B-A3B; run=XfjRkSyP4ub4gDsWcgS7KY; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":62.77777777777778,"warning":null},{"id":"epoch-run-XGK9DjJHHVWr2LyT298ooL","modelId":"qwen3-14b","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"qwen3-14b","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3-14b; run=XGK9DjJHHVWr2LyT298ooL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":63.76262626262626,"warning":null},{"id":"epoch-run-Xi45sLZGyn5sutf7nKRHTk","modelId":"glm-5.2","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"non-reasoning","sourceLabel":"glm-5.2_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=glm-5.2_none; run=Xi45sLZGyn5sutf7nKRHTk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":19,"warning":null},{"id":"epoch-run-XjFqou6MyT2XLpkUsCrgxM","modelId":"glm-5.2","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"glm-5.2_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=glm-5.2_max; run=XjFqou6MyT2XLpkUsCrgxM; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.38888888888889,"warning":null},{"id":"epoch-run-XneEig6pACkAWeekvgCwNk","modelId":"kimi-k2.7-code","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"unspecified","sourceLabel":"kimi-k2.7-code","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=kimi-k2.7-code; run=XneEig6pACkAWeekvgCwNk; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":87.87878787878788,"warning":null},{"id":"epoch-run-XnQVH5qSfxVjpzoPwUEmnh","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"claude-sonnet-4-6_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-sonnet-4-6_max; run=XnQVH5qSfxVjpzoPwUEmnh; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.11111111111111,"warning":null},{"id":"epoch-run-XQEGUQjpXupjdyKoGuSmum","modelId":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"thinking","sourceLabel":"claude-sonnet-4-5-20250929_59K (thinking 59k)","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-sonnet-4-5-20250929_59K; run=XQEGUQjpXupjdyKoGuSmum; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":30.7,"warning":null},{"id":"epoch-run-XRutLYUVUePe69APTSTw8S","modelId":"claude-sonnet-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"max","sourceLabel":"claude-sonnet-5_max","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-sonnet-5_max; run=XRutLYUVUePe69APTSTw8S; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80,"warning":null},{"id":"epoch-run-XTW92PMsypptnTGBFuLCxe","modelId":"claude-opus-5","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"low","sourceLabel":"claude-opus-5_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-5_low; run=XTW92PMsypptnTGBFuLCxe; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":27,"warning":null},{"id":"epoch-run-Xu34dKTAdpBWL6advf6DMg","modelId":"gemini-3.7-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"gemini-3.7-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.7-flash_high; run=Xu34dKTAdpBWL6advf6DMg; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66,"warning":null},{"id":"epoch-run-XUDdWKUZBhdWPPQBRgx8Pw","modelId":"gemini-3.5-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"gemini-3.5-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.5-flash_high; run=XUDdWKUZBhdWPPQBRgx8Pw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":45,"warning":null},{"id":"epoch-run-XWpHpVPCaVjcwPpdpXuuGH","modelId":"qwen3-235b-a22b-thinking-2507","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-3-model-extracted-exact-match","level":"unspecified","sourceLabel":"Qwen3-235B-A22B-Thinking-2507","configuration":"Epoch AI Inspect; task 1.0.3; scorer model_extracted_exact_match; model=Qwen3-235B-A22B-Thinking-2507; run=XWpHpVPCaVjcwPpdpXuuGH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":86.66666666666667,"warning":null},{"id":"epoch-run-XXPqVNR7WCFFA2Rnzjcbk6","modelId":"claude-opus-4-7","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"max","sourceLabel":"claude-opus-4-7_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=claude-opus-4-7_max; run=XXPqVNR7WCFFA2Rnzjcbk6; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":41,"warning":null},{"id":"epoch-run-XyVL3VffJ6UboY97vxpohL","modelId":"gemini-3.7-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"high","sourceLabel":"gemini-3.7-flash_high","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=gemini-3.7-flash_high; run=XyVL3VffJ6UboY97vxpohL; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":94.82323232323232,"warning":null},{"id":"epoch-run-Y39ZRp553ULpUDBV2XrBbc","modelId":"grok-4.3","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"high","sourceLabel":"grok-4.3_high","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=grok-4.3_high; run=Y39ZRp553ULpUDBV2XrBbc; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33.2,"warning":null},{"id":"epoch-run-YBMTDnW5LmCN6tBp2QYHzi","modelId":"qwen3.5-9b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"qwen3.5-9B_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=qwen3.5-9B_none; run=YBMTDnW5LmCN6tBp2QYHzi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":61.66666666666667,"warning":null},{"id":"epoch-run-YBqiCXPjMUSAgAjDUoz5sy","modelId":"gemini-3.6-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"high","sourceLabel":"gemini-3.6-flash_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gemini-3.6-flash_high; run=YBqiCXPjMUSAgAjDUoz5sy; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":30,"warning":null},{"id":"epoch-run-YC2u9wykbYQg4gdkf76j26","modelId":"glm-5.3-flash","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"glm-5.3-flash_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=glm-5.3-flash_max; run=YC2u9wykbYQg4gdkf76j26; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":55.78947368421052,"warning":null},{"id":"epoch-run-YdcyYXLzqfPUSSgGv4MTED","modelId":"claude-sonnet-5","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"max","sourceLabel":"claude-sonnet-5_max","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=claude-sonnet-5_max; run=YdcyYXLzqfPUSSgGv4MTED; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":35,"warning":null},{"id":"epoch-run-YDQJAMgMjLaikPkeVDbAjr","modelId":"glm-5.2","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"glm-5.2_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=glm-5.2_none; run=YDQJAMgMjLaikPkeVDbAjr; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":71.21212121212122,"warning":null},{"id":"epoch-run-YdyvoMQ4mA5V8Ymg6Tksxu","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"low","sourceLabel":"claude-sonnet-4-6_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=claude-sonnet-4-6_low; run=YdyvoMQ4mA5V8Ymg6Tksxu; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":16,"warning":null},{"id":"epoch-run-Yhds8GYkCXWgRQhuLiURua","modelId":"qwen3.7-flash","benchmarkId":"epoch-gpqa-diamond-1-0-11-choice","level":"non-reasoning","sourceLabel":"qwen3.7-flash_none","configuration":"Epoch AI Inspect; task 1.0.11; scorer choice; model=qwen3.7-flash_none; run=Yhds8GYkCXWgRQhuLiURua; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":80.8080808080808,"warning":null},{"id":"epoch-run-YLAk5J3Wd8EruVXee8bAQb","modelId":"gpt-5.6-luna","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-5.6-luna_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-luna_max; run=YLAk5J3Wd8EruVXee8bAQb; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":82.10526315789474,"warning":null},{"id":"epoch-run-YoAJyC3j8dZ6ZpDZ5rSDLw","modelId":"grok-4.6","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"grok-4.6_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=grok-4.6_high; run=YoAJyC3j8dZ6ZpDZ5rSDLw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":33,"warning":null},{"id":"epoch-run-YRvn36dmMyDbS5e4XkF4Xi","modelId":"qwen3.6-35b-a3b","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.6-35b-a3b_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-35b-a3b_none; run=YRvn36dmMyDbS5e4XkF4Xi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":20.350877192982455,"warning":null},{"id":"epoch-run-YwHKNFQjbat97wVLqachmi","modelId":"gpt-5.6-luna","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"none","sourceLabel":"gpt-5.6-luna_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-luna_none; run=YwHKNFQjbat97wVLqachmi; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":39.64912280701755,"warning":null},{"id":"epoch-run-Z4xfwTke5pR6RrY952bttA","modelId":"qwen3.6-plus","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"non-reasoning","sourceLabel":"qwen3.6-plus_none","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=qwen3.6-plus_none; run=Z4xfwTke5pR6RrY952bttA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32.280701754385966,"warning":null},{"id":"epoch-run-Z5FAHm8aqmk8NUUm7tojng","modelId":"claude-opus-5","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"claude-opus-5_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=claude-opus-5_low; run=Z5FAHm8aqmk8NUUm7tojng; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":93.33333333333333,"warning":null},{"id":"epoch-run-Z6S3nMC2tnkpKUixCPLiDj","modelId":"gpt-5.6-terra","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gpt-5.6-terra_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-terra_low; run=Z6S3nMC2tnkpKUixCPLiDj; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":88.88888888888889,"warning":null},{"id":"epoch-run-ZavSMMq9jk5s9WSbARZr6X","modelId":"gemini-3.7-flash","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-agent-submitted-next-move-scorer","level":"high","sourceLabel":"gemini-3.7-flash_high","configuration":"Epoch AI Inspect; task 1.0.4; scorer agent_submitted_next_move_scorer; model=gemini-3.7-flash_high; run=ZavSMMq9jk5s9WSbARZr6X; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":37,"warning":null},{"id":"epoch-run-Zb2AKwa942D7dBTTcebAba","modelId":"grok-4.6","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"xhigh","sourceLabel":"grok-4.6_xhigh","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=grok-4.6_xhigh; run=Zb2AKwa942D7dBTTcebAba; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":65.96491228070175,"warning":null},{"id":"epoch-run-ZczEhu69FQ9Cnns8NT6ew4","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"unspecified","sourceLabel":"gemini-3-flash-preview","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gemini-3-flash-preview; run=ZczEhu69FQ9Cnns8NT6ew4; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":17.073170731707318,"warning":null},{"id":"epoch-run-ZDeGG5QYkdgVqzGPmmnbfx","modelId":"gemini-3.6-flash","benchmarkId":"epoch-chess-puzzles-2-0-0-exact","level":"high","sourceLabel":"gemini-3.6-flash_high","configuration":"Epoch AI Inspect; task 2.0.0; scorer exact; model=gemini-3.6-flash_high; run=ZDeGG5QYkdgVqzGPmmnbfx; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":40,"warning":null},{"id":"epoch-run-Zftz2yk4X5NF28JQzjN3hZ","modelId":"claude-opus-5","benchmarkId":"epoch-simpleqa-verified-1-0-0-simpleqa-scorer","level":"max","sourceLabel":"claude-opus-5_max","configuration":"Epoch AI Inspect; task 1.0.0; scorer simpleqa_scorer; model=claude-opus-5_max; run=Zftz2yk4X5NF28JQzjN3hZ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":59.9,"warning":null},{"id":"epoch-run-ZgU8BagFmaj7BE94y3JDZA","modelId":"gpt-5.6-luna","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"low","sourceLabel":"gpt-5.6-luna_low","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=gpt-5.6-luna_low; run=ZgU8BagFmaj7BE94y3JDZA; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":66.66666666666666,"warning":null},{"id":"epoch-run-ZjsXQpZ7SsejXDo52aetZZ","modelId":"llama-3.3-70b-instruct","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-0-0-model-graded","level":"unspecified","sourceLabel":"Llama-3.3-70B-Instruct","configuration":"Epoch AI Inspect; task 1.0.0; scorer model_graded; model=Llama-3.3-70B-Instruct; run=ZjsXQpZ7SsejXDo52aetZZ; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":5.138888888888888,"warning":null},{"id":"epoch-run-ZkNQ7GER7Mc9FSPmpUzgKH","modelId":"claude-opus-4-7","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"max","sourceLabel":"claude-opus-4-7_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=claude-opus-4-7_max; run=ZkNQ7GER7Mc9FSPmpUzgKH; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70.17543859649122,"warning":null},{"id":"epoch-run-ZQMcTQjHcDSevCN2dYYHKY","modelId":"gemini-3-flash-preview","benchmarkId":"epoch-mystery-game-puzzles-1-0-4-exact-match-next-move-scorer","level":"low","sourceLabel":"gemini-3-flash-preview_low","configuration":"Epoch AI Inspect; task 1.0.4; scorer exact_match_next_move_scorer; model=gemini-3-flash-preview_low; run=ZQMcTQjHcDSevCN2dYYHKY; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":26,"warning":null},{"id":"epoch-run-ZryXhNUpPedPvGhbqdNBEw","modelId":"gpt-5.6-luna","benchmarkId":"epoch-frontiermath-tiers-1-3-v2-2-0-0-verification-code","level":"low","sourceLabel":"gpt-5.6-luna_low","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-luna_low; run=ZryXhNUpPedPvGhbqdNBEw; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":41.40350877192983,"warning":null},{"id":"epoch-run-Zs5fCvNzaz8LhqgwDSAztU","modelId":"claude-sonnet-4-6","benchmarkId":"epoch-simpleqa-verified-1-2-0-simpleqa-scorer","level":"max","sourceLabel":"claude-sonnet-4-6_max","configuration":"Epoch AI Inspect; task 1.2.0; scorer simpleqa_scorer; model=claude-sonnet-4-6_max; run=Zs5fCvNzaz8LhqgwDSAztU; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":32.800000000000004,"warning":null},{"id":"epoch-run-ZUJxsJDpHnLXxNiDcvj22W","modelId":"gemma-2-27b-it","benchmarkId":"epoch-gpqa-diamond-1-0-0-choice","level":"unspecified","sourceLabel":"gemma-2-27b-it","configuration":"Epoch AI Inspect; task 1.0.0; scorer choice; model=gemma-2-27b-it; run=ZUJxsJDpHnLXxNiDcvj22W; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":36.48989898989899,"warning":null},{"id":"epoch-run-ZUygpkXqg3NjchXXfedAWV","modelId":"gpt-5.6-terra","benchmarkId":"epoch-frontiermath-tier-4-v2-2-0-0-verification-code","level":"max","sourceLabel":"gpt-5.6-terra_max","configuration":"Epoch AI Inspect; task 2.0.0; scorer verification_code; model=gpt-5.6-terra_max; run=ZUygpkXqg3NjchXXfedAWV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":70.73170731707317,"warning":null},{"id":"epoch-run-ZzbWv2QGVfg4o2DTX45nqV","modelId":"qwen3-30b-a3b","benchmarkId":"epoch-otis-mock-aime-2024-2025-1-1-6-model-extracted-exact-match","level":"non-reasoning","sourceLabel":"Qwen3-30B-A3B_none","configuration":"Epoch AI Inspect; task 1.1.6; scorer model_extracted_exact_match; model=Qwen3-30B-A3B_none; run=ZzbWv2QGVfg4o2DTX45nqV; mean accuracy; https://epoch.ai/data/benchmarks.csv","score":25.555555555555554,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-4-argon","modelId":"gemini-4-argon","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"{\"checkpoint\":\"gemini-4-argon\",\"evaluatorModel\":\"google/gemini-4-argon\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":65.401,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":61.435,"warning":null},{"id":"vals-finance-agent-v2:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-finance-agent-v2","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":60.599,"warning":null},{"id":"vals-finance-agent-v2:meta/muse-wonder-full","modelId":"muse-wonder-full","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Wonder Full (max)","configuration":"{\"checkpoint\":\"straw-110_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder-full\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":59.061,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":59.042,"warning":null},{"id":"vals-finance-agent-v2:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-finance-agent-v2","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":58.901,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":58.877,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":58.633,"warning":"One run-task used an Opus 4.8 refusal fallback; mixed checkpoint result excluded from ranking"},{"id":"vals-finance-agent-v2:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":58.103,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":57.861,"warning":null},{"id":"vals-finance-agent-v2:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Fireworks AI\",\"harness\":null}","score":57.85,"warning":null},{"id":"vals-finance-agent-v2:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-finance-agent-v2","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":57.207,"warning":null},{"id":"vals-finance-agent-v2:meta/muse-wonder","modelId":"muse-wonder","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Wonder (max)","configuration":"{\"checkpoint\":\"restful-rf120_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":56.637,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":56.314,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":56.296,"warning":null},{"id":"vals-finance-agent-v2:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Fireworks AI\",\"harness\":null}","score":55.84,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":55.044,"warning":null},{"id":"vals-finance-agent-v2:mistralai/mistral-large-4","modelId":"mistral-large-4","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Mistral Large 4 (high)","configuration":"{\"checkpoint\":\"mistral-large-4\",\"evaluatorModel\":\"mistralai/mistral-large-4\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":54.678,"warning":"Vals reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"vals-finance-agent-v2:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":54.436,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":53.918,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":53.909,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":53.756,"warning":null},{"id":"vals-finance-agent-v2:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":53.682,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":53.54,"warning":null},{"id":"vals-finance-agent-v2:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":53.114,"warning":null},{"id":"vals-finance-agent-v2:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-finance-agent-v2","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":52.251,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-6.1-sol","modelId":"gpt-6.1-sol","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6.1-sol\",\"evaluatorModel\":\"openai/gpt-6.1-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":52.033,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-finance-agent-v2","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":51.76,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Claude Opus 4.7 (high)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"high\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":51.509,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":51.035,"warning":null},{"id":"vals-finance-agent-v2:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":50.393,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-6-luna\",\"evaluatorModel\":\"openai/gpt-6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":49.873,"warning":null},{"id":"vals-finance-agent-v2:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"GLM 5.2 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":49.699,"warning":null},{"id":"vals-finance-agent-v2:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":49.517,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"vals-finance-agent-v2","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6-sol\",\"evaluatorModel\":\"openai/gpt-6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":49.05,"warning":null},{"id":"vals-finance-agent-v2:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":48.349,"warning":null},{"id":"vals-finance-agent-v2:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":48.269,"warning":null},{"id":"vals-finance-agent-v2:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":47.776,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":47.442,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-finance-agent-v2","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":45.36,"warning":null},{"id":"vals-finance-agent-v2:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":44.899,"warning":null},{"id":"vals-finance-agent-v2:zai/glm-5.1","modelId":"glm-5.1","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"GLM 5.1 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.1\",\"evaluatorModel\":\"zai/glm-5.1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":44.792,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":42.982,"warning":null},{"id":"vals-finance-agent-v2:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":42.551,"warning":null},{"id":"vals-finance-agent-v2:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":41.502,"warning":null},{"id":"vals-finance-agent-v2:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":40.846,"warning":null},{"id":"vals-finance-agent-v2:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":38.22,"warning":null},{"id":"vals-finance-agent-v2:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":38.217,"warning":null},{"id":"vals-finance-agent-v2:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":37.728,"warning":null},{"id":"vals-finance-agent-v2:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":36.728,"warning":null},{"id":"vals-finance-agent-v2:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-finance-agent-v2","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":32.103,"warning":null},{"id":"vals-finance-agent-v2:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":31.01,"warning":null},{"id":"vals-finance-agent-v2:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":28.492,"warning":null},{"id":"vals-finance-agent-v2:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-finance-agent-v2","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":196608,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":27.887,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":69.825,"warning":null},{"id":"vals-code-migration-v1:google/gemini-4-argon","modelId":"gemini-4-argon","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"{\"checkpoint\":\"gemini-4-argon\",\"evaluatorModel\":\"google/gemini-4-argon\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":68.173,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":67.74,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-6.1-sol","modelId":"gpt-6.1-sol","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6.1-sol\",\"evaluatorModel\":\"openai/gpt-6.1-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":65.122,"warning":null},{"id":"vals-code-migration-v1:meta/muse-wonder","modelId":"muse-wonder","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Wonder (max)","configuration":"{\"checkpoint\":\"restful-rf120_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":62.552,"warning":null},{"id":"vals-code-migration-v1:meta/muse-wonder-full","modelId":"muse-wonder-full","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Wonder Full (max)","configuration":"{\"checkpoint\":\"straw-110_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder-full\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":61.033,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":57.473,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6-sol\",\"evaluatorModel\":\"openai/gpt-6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":57.195,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":55.065,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":54.607,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":52.916,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":47.804,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":47.25,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-code-migration-v1","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":45.161,"warning":null},{"id":"vals-code-migration-v1:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-code-migration-v1","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":44.816,"warning":null},{"id":"vals-code-migration-v1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":44.572,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":44.55,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":44.392,"warning":null},{"id":"vals-code-migration-v1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":44.222,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":43.878,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-6-luna\",\"evaluatorModel\":\"openai/gpt-6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":42.554,"warning":null},{"id":"vals-code-migration-v1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":41.54,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":39.892,"warning":null},{"id":"vals-code-migration-v1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":38.631,"warning":null},{"id":"vals-code-migration-v1:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GLM 5.2 (max)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":37.87,"warning":null},{"id":"vals-code-migration-v1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":36.596,"warning":null},{"id":"vals-code-migration-v1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":36.546,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","benchmarkId":"vals-code-migration-v1","level":"xhigh","sourceLabel":"GPT 5.4 (xhigh) (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-2026-03-05\",\"evaluatorModel\":\"openai/gpt-5.4-2026-03-05\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":34.984,"warning":null},{"id":"vals-code-migration-v1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":34.8,"warning":null},{"id":"vals-code-migration-v1:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-code-migration-v1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":31.115,"warning":null},{"id":"vals-code-migration-v1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":30.928,"warning":null},{"id":"vals-code-migration-v1:mistralai/mistral-large-4","modelId":"mistral-large-4","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Mistral Large 4 (high)","configuration":"{\"checkpoint\":\"mistral-large-4\",\"evaluatorModel\":\"mistralai/mistral-large-4\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":30.561,"warning":"Vals reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"vals-code-migration-v1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-code-migration-v1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":29.954,"warning":null},{"id":"vals-code-migration-v1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":27.768,"warning":null},{"id":"vals-code-migration-v1:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-code-migration-v1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":27.58,"warning":null},{"id":"vals-code-migration-v1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":26.745,"warning":null},{"id":"vals-code-migration-v1:zai/glm-5.1","modelId":"glm-5.1","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"GLM 5.1 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.1\",\"evaluatorModel\":\"zai/glm-5.1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":25.768,"warning":null},{"id":"vals-code-migration-v1:kimi/kimi-k2.7-code","modelId":"kimi-k2.7-code","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"Kimi K2.7 Code (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.7-code\",\"evaluatorModel\":\"kimi/kimi-k2.7-code\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":32768,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":25.392,"warning":null},{"id":"vals-code-migration-v1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":21.557,"warning":null},{"id":"vals-code-migration-v1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":20.515,"warning":null},{"id":"vals-code-migration-v1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":19.925,"warning":null},{"id":"vals-code-migration-v1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":17.313,"warning":null},{"id":"vals-code-migration-v1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-code-migration-v1","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":16.099,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":14.467,"warning":null},{"id":"vals-code-migration-v1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":14.228,"warning":null},{"id":"vals-code-migration-v1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":13.069,"warning":null},{"id":"vals-code-migration-v1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-code-migration-v1","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":12.936,"warning":null},{"id":"vals-code-migration-v1:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":12.856,"warning":null},{"id":"vals-code-migration-v1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":11.102,"warning":null},{"id":"vals-code-migration-v1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":8.685,"warning":null},{"id":"vals-code-migration-v1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":7.551,"warning":null},{"id":"vals-code-migration-v1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":6.795,"warning":null},{"id":"vals-code-migration-v1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":6.391,"warning":null},{"id":"vals-code-migration-v1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":6.148,"warning":null},{"id":"vals-code-migration-v1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-code-migration-v1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":5.127,"warning":null},{"id":"vals-code-migration-v1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-code-migration-v1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":0.314,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":55.288,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":55.288,"warning":null},{"id":"vals-legal-research-v1:google/gemini-4-argon","modelId":"gemini-4-argon","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"{\"checkpoint\":\"gemini-4-argon\",\"evaluatorModel\":\"google/gemini-4-argon\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":54.808,"warning":null},{"id":"vals-legal-research-v1:meta/muse-wonder-full","modelId":"muse-wonder-full","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Wonder Full (max)","configuration":"{\"checkpoint\":\"straw-110_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder-full\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":53.365,"warning":null},{"id":"vals-legal-research-v1:meta/muse-wonder","modelId":"muse-wonder","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Wonder (max)","configuration":"{\"checkpoint\":\"restful-rf120_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":50.481,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":49.519,"warning":null},{"id":"vals-legal-research-v1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":49.038,"warning":null},{"id":"vals-legal-research-v1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":48.077,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":48.077,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":48.077,"warning":null},{"id":"vals-legal-research-v1:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-legal-research-v1","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":47.115,"warning":null},{"id":"vals-legal-research-v1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":46.154,"warning":null},{"id":"vals-legal-research-v1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":45.192,"warning":null},{"id":"vals-legal-research-v1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-legal-research-v1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":43.75,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":43.75,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":41.827,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":41.346,"warning":null},{"id":"vals-legal-research-v1:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-legal-research-v1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":40.865,"warning":null},{"id":"vals-legal-research-v1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":40.865,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-legal-research-v1","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":40.385,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":39.423,"warning":null},{"id":"vals-legal-research-v1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":38.942,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":38.462,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":38.462,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-6.1-sol","modelId":"gpt-6.1-sol","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6.1-sol\",\"evaluatorModel\":\"openai/gpt-6.1-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":38.462,"warning":null},{"id":"vals-legal-research-v1:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-legal-research-v1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":37.981,"warning":null},{"id":"vals-legal-research-v1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":37.981,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":36.538,"warning":null},{"id":"vals-legal-research-v1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":34.615,"warning":null},{"id":"vals-legal-research-v1:mistralai/mistral-large-4","modelId":"mistral-large-4","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Mistral Large 4 (high)","configuration":"{\"checkpoint\":\"mistral-large-4\",\"evaluatorModel\":\"mistralai/mistral-large-4\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":31.731,"warning":"Vals reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"vals-legal-research-v1:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GLM 5.2 (max)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":31.25,"warning":null},{"id":"vals-legal-research-v1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":30.769,"warning":null},{"id":"vals-legal-research-v1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":30.288,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-6-luna\",\"evaluatorModel\":\"openai/gpt-6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":30.288,"warning":null},{"id":"vals-legal-research-v1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":29.808,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"vals-legal-research-v1","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6-sol\",\"evaluatorModel\":\"openai/gpt-6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":28.846,"warning":null},{"id":"vals-legal-research-v1:zai/glm-5.1","modelId":"glm-5.1","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"GLM 5.1 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.1\",\"evaluatorModel\":\"zai/glm-5.1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":27.885,"warning":null},{"id":"vals-legal-research-v1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":25.481,"warning":null},{"id":"vals-legal-research-v1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":25,"warning":null},{"id":"vals-legal-research-v1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":20.673,"warning":null},{"id":"vals-legal-research-v1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":18.269,"warning":null},{"id":"vals-legal-research-v1:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":16.346,"warning":null},{"id":"vals-legal-research-v1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":15.865,"warning":null},{"id":"vals-legal-research-v1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":15.865,"warning":null},{"id":"vals-legal-research-v1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":15.385,"warning":null},{"id":"vals-legal-research-v1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":14.904,"warning":null},{"id":"vals-legal-research-v1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":13.942,"warning":null},{"id":"vals-legal-research-v1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":13.942,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-legal-research-v1","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":12.5,"warning":null},{"id":"vals-legal-research-v1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":10.577,"warning":null},{"id":"vals-legal-research-v1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":10.577,"warning":null},{"id":"vals-legal-research-v1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-legal-research-v1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":9.135,"warning":null},{"id":"vals-legal-research-v1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":9.135,"warning":null},{"id":"vals-legal-research-v1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-legal-research-v1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":6.25,"warning":null},{"id":"vals-emb-v1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":76.67,"warning":null},{"id":"vals-emb-v1:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":75.713,"warning":null},{"id":"vals-emb-v1:google/gemini-4-argon","modelId":"gemini-4-argon","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"{\"checkpoint\":\"gemini-4-argon\",\"evaluatorModel\":\"google/gemini-4-argon\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":75.244,"warning":null},{"id":"vals-emb-v1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":73.666,"warning":null},{"id":"vals-emb-v1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":73.556,"warning":null},{"id":"vals-emb-v1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":72.339,"warning":null},{"id":"vals-emb-v1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":72.196,"warning":null},{"id":"vals-emb-v1:meta/muse-wonder-full","modelId":"muse-wonder-full","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Wonder Full (max)","configuration":"{\"checkpoint\":\"straw-110_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder-full\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":71.917,"warning":null},{"id":"vals-emb-v1:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":71.7,"warning":null},{"id":"vals-emb-v1:openai/gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6-sol\",\"evaluatorModel\":\"openai/gpt-6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":71.53,"warning":null},{"id":"vals-emb-v1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":71.326,"warning":null},{"id":"vals-emb-v1:openai/gpt-6.1-sol","modelId":"gpt-6.1-sol","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6.1-sol\",\"evaluatorModel\":\"openai/gpt-6.1-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":70.773,"warning":null},{"id":"vals-emb-v1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":69.373,"warning":null},{"id":"vals-emb-v1:openai/gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-6-luna\",\"evaluatorModel\":\"openai/gpt-6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":68.515,"warning":null},{"id":"vals-emb-v1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":67.116,"warning":null},{"id":"vals-emb-v1:meta/muse-wonder","modelId":"muse-wonder","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Wonder (max)","configuration":"{\"checkpoint\":\"restful-rf120_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":67.084,"warning":null},{"id":"vals-emb-v1:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-emb-v1","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":66.991,"warning":null},{"id":"vals-emb-v1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":66.677,"warning":null},{"id":"vals-emb-v1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":66.316,"warning":null},{"id":"vals-emb-v1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":66.205,"warning":null},{"id":"vals-emb-v1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":65.411,"warning":null},{"id":"vals-emb-v1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-emb-v1","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":64.543,"warning":null},{"id":"vals-emb-v1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":63.728,"warning":null},{"id":"vals-emb-v1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":63.548,"warning":null},{"id":"vals-emb-v1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":62.727,"warning":null},{"id":"vals-emb-v1:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-emb-v1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":62.712,"warning":null},{"id":"vals-emb-v1:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"GLM 5.2 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":61.532,"warning":null},{"id":"vals-emb-v1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":60.154,"warning":null},{"id":"vals-emb-v1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":57.855,"warning":null},{"id":"vals-emb-v1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":57.041,"warning":null},{"id":"vals-emb-v1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-emb-v1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":56.978,"warning":null},{"id":"vals-emb-v1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":56.977,"warning":null},{"id":"vals-emb-v1:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-emb-v1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":56.361,"warning":null},{"id":"vals-emb-v1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":56.344,"warning":null},{"id":"vals-emb-v1:mistralai/mistral-large-4","modelId":"mistral-large-4","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Mistral Large 4 (high)","configuration":"{\"checkpoint\":\"mistral-large-4\",\"evaluatorModel\":\"mistralai/mistral-large-4\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":56.213,"warning":"Vals reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"vals-emb-v1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":55.932,"warning":null},{"id":"vals-emb-v1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":55.228,"warning":null},{"id":"vals-emb-v1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":55.092,"warning":null},{"id":"vals-emb-v1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":52.911,"warning":null},{"id":"vals-emb-v1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-emb-v1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":52.803,"warning":null},{"id":"vals-emb-v1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":52.616,"warning":null},{"id":"vals-emb-v1:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":49.345,"warning":null},{"id":"vals-emb-v1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":47.759,"warning":null},{"id":"vals-emb-v1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-emb-v1","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":45.426,"warning":null},{"id":"vals-emb-v1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":44.751,"warning":null},{"id":"vals-emb-v1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":43.231,"warning":null},{"id":"vals-emb-v1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":32.867,"warning":null},{"id":"vals-emb-v1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":28.451,"warning":null},{"id":"vals-emb-v1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":26.017,"warning":null},{"id":"vals-emb-v1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":22.664,"warning":null},{"id":"vals-emb-v1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":18.394,"warning":null},{"id":"vals-emb-v1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-emb-v1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":11.739,"warning":null},{"id":"vals-emb-v1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-emb-v1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":11.185,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":92.375,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":91.591,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":91.502,"warning":null},{"id":"vals-mmlu-pro-v1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65535,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":90.987,"warning":null},{"id":"vals-mmlu-pro-v1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":90.221,"warning":null},{"id":"vals-mmlu-pro-v1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":90.121,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":89.871,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":89.585,"warning":null},{"id":"vals-mmlu-pro-v1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":89.515,"warning":null},{"id":"vals-mmlu-pro-v1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":89.4,"warning":null},{"id":"vals-mmlu-pro-v1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":89.311,"warning":null},{"id":"vals-mmlu-pro-v1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":89.276,"warning":null},{"id":"vals-mmlu-pro-v1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":89.216,"warning":null},{"id":"vals-mmlu-pro-v1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":89.1,"warning":null},{"id":"vals-mmlu-pro-v1:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-mmlu-pro-v1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":88.726,"warning":null},{"id":"vals-mmlu-pro-v1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":88.592,"warning":null},{"id":"vals-mmlu-pro-v1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-mmlu-pro-v1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":88.275,"warning":null},{"id":"vals-mmlu-pro-v1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-mmlu-pro-v1","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":88.144,"warning":null},{"id":"vals-mmlu-pro-v1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":87.973,"warning":null},{"id":"vals-mmlu-pro-v1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":87.668,"warning":null},{"id":"vals-mmlu-pro-v1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":87.572,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":87.545,"warning":null},{"id":"vals-mmlu-pro-v1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","benchmarkId":"vals-mmlu-pro-v1","level":"xhigh","sourceLabel":"GPT 5.4 (xhigh) (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-2026-03-05\",\"evaluatorModel\":\"openai/gpt-5.4-2026-03-05\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":87.482,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":87.341,"warning":null},{"id":"vals-mmlu-pro-v1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":86.972,"warning":null},{"id":"vals-mmlu-pro-v1:zai/glm-5.1","modelId":"glm-5.1","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"GLM 5.1 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.1\",\"evaluatorModel\":\"zai/glm-5.1\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":86.9,"warning":null},{"id":"vals-mmlu-pro-v1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":86.77,"warning":null},{"id":"vals-mmlu-pro-v1:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"GLM 5.2 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":86.714,"warning":null},{"id":"vals-mmlu-pro-v1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-mmlu-pro-v1","level":"xhigh","sourceLabel":"GPT-5.6 Terra (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":86.659,"warning":null},{"id":"vals-mmlu-pro-v1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":86.254,"warning":null},{"id":"vals-mmlu-pro-v1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":86.206,"warning":null},{"id":"vals-mmlu-pro-v1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":86.059,"warning":null},{"id":"vals-mmlu-pro-v1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-mmlu-pro-v1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":86.036,"warning":null},{"id":"vals-mmlu-pro-v1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":85.838,"warning":null},{"id":"vals-mmlu-pro-v1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":85.836,"warning":null},{"id":"vals-mmlu-pro-v1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":84.595,"warning":null},{"id":"vals-mmlu-pro-v1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-mmlu-pro-v1","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":84.554,"warning":null},{"id":"vals-mmlu-pro-v1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":84.22,"warning":null},{"id":"vals-mmlu-pro-v1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":82.931,"warning":null},{"id":"vals-mmlu-pro-v1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":80.427,"warning":null},{"id":"vals-mmlu-pro-v1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-mmlu-pro-v1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":78.715,"warning":null},{"id":"vals-mmlu-pro-v1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":77.172,"warning":null},{"id":"vals-mmlu-pro-v1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-mmlu-pro-v1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-09-01\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":75.335,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":87.266,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":85.768,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Claude Fable 5.1 (high)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"high\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":85.019,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Claude Opus 5 (high)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"high\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":84.644,"warning":"Source discloses an Opus 4.8 refusal fallback; mixed checkpoint result excluded from ranking"},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6-sol\",\"evaluatorModel\":\"openai/gpt-6-sol\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":83.146,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":81.273,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Kimi K3 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":80.899,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":80.524,"warning":"Source discloses an Opus 4.8 refusal fallback; mixed checkpoint result excluded from ranking"},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":79.026,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":78.277,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":77.528,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":77.528,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"GPT 5.5 (high)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":76.404,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":74.532,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":74.157,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":73.783,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":73.408,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-6-luna\",\"evaluatorModel\":\"openai/gpt-6-luna\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":73.034,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":72.285,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":71.91,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":71.536,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":70.787,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":69.663,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":69.288,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Claude Opus 4.7 (high)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"high\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":68.539,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":67.79,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GLM 5.2 (max)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":1,\"max_output_tokens\":48000,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":67.79,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"DeepSeek V4 Flash 0731 (unspecified)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":false,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":67.041,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:kimi/kimi-k2.7-code","modelId":"kimi-k2.7-code","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Kimi K2.7 Code (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.7-code\",\"evaluatorModel\":\"kimi/kimi-k2.7-code\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":32768,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":67.041,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Fireworks AI\",\"harness\":null}","score":62.921,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":61.049,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":60.674,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":false,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":57.303,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":57.303,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:zai/glm-5.1","modelId":"glm-5.1","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"GLM 5.1 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.1\",\"evaluatorModel\":\"zai/glm-5.1\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":56.929,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"GPT 5.4 Mini (high)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":54.682,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":54.682,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":53.933,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":53.558,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":53.558,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":53.184,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":52.809,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":50.187,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":196608,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":48.689,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":44.195,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":43.82,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":41.948,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":41.573,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":38.951,"warning":null},{"id":"vals-terminal-bench-2-1-v2.1:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-terminal-bench-2-1-v2.1","level":"high","sourceLabel":"Claude Sonnet 5.5 (high)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-09-28\",\"reasoning_effort\":null,\"compute_effort\":\"high\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":83.146,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":92.392,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-4-argon","modelId":"gemini-4-argon","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"{\"checkpoint\":\"gemini-4-argon\",\"evaluatorModel\":\"google/gemini-4-argon\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":91.906,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":90.352,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":90.263,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":89.594,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-6.1-sol","modelId":"gpt-6.1-sol","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6.1-sol\",\"evaluatorModel\":\"openai/gpt-6.1-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":88.931,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Claude Opus 5 (unspecified)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":88.404,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6-sol\",\"evaluatorModel\":\"openai/gpt-6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":87.824,"warning":null},{"id":"vals-vibe-code-bench-v1.1:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-vibe-code-bench-v1.1","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":\"OpenHands\"}","score":86.175,"warning":null},{"id":"vals-vibe-code-bench-v1.1:meta/muse-wonder-full","modelId":"muse-wonder-full","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Wonder Full (max)","configuration":"{\"checkpoint\":\"straw-110_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder-full\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":\"OpenHands\"}","score":85.996,"warning":null},{"id":"vals-vibe-code-bench-v1.1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":\"OpenHands\"}","score":84.968,"warning":null},{"id":"vals-vibe-code-bench-v1.1:meta/muse-wonder","modelId":"muse-wonder","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Wonder (max)","configuration":"{\"checkpoint\":\"restful-rf120_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":\"OpenHands\"}","score":83.089,"warning":null},{"id":"vals-vibe-code-bench-v1.1:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-vibe-code-bench-v1.1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":\"OpenHands\"}","score":82.859,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":82.725,"warning":null},{"id":"vals-vibe-code-bench-v1.1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":\"OpenHands\"}","score":82.297,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-6-luna\",\"evaluatorModel\":\"openai/gpt-6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":81.649,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":81.325,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":80.495,"warning":null},{"id":"vals-vibe-code-bench-v1.1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-vibe-code-bench-v1.1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":\"OpenHands\"}","score":79.1,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":78.651,"warning":null},{"id":"vals-vibe-code-bench-v1.1:mistralai/mistral-large-4","modelId":"mistral-large-4","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Mistral Large 4 (high)","configuration":"{\"checkpoint\":\"mistral-large-4\",\"evaluatorModel\":\"mistralai/mistral-large-4\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":\"OpenHands\"}","score":78.403,"warning":"Vals reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"vals-vibe-code-bench-v1.1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":\"OpenHands\"}","score":78.125,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":77.055,"warning":null},{"id":"vals-vibe-code-bench-v1.1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":\"OpenHands\"}","score":76.239,"warning":null},{"id":"vals-vibe-code-bench-v1.1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":\"OpenHands\"}","score":74.736,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":74.594,"warning":null},{"id":"vals-vibe-code-bench-v1.1:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-vibe-code-bench-v1.1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":\"OpenHands\"}","score":72.158,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":71.003,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":70.395,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":69.847,"warning":null},{"id":"vals-vibe-code-bench-v1.1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":\"OpenHands\"}","score":69.002,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","benchmarkId":"vals-vibe-code-bench-v1.1","level":"xhigh","sourceLabel":"GPT 5.4 (xhigh) (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-2026-03-05\",\"evaluatorModel\":\"openai/gpt-5.4-2026-03-05\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":67.421,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":64.005,"warning":null},{"id":"vals-vibe-code-bench-v1.1:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GLM 5.2 (max)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":\"OpenHands\"}","score":63.961,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":51.476,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":48.683,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-vibe-code-bench-v1.1","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":47.969,"warning":null},{"id":"vals-vibe-code-bench-v1.1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":\"OpenHands\"}","score":47.671,"warning":null},{"id":"vals-vibe-code-bench-v1.1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":\"OpenHands\"}","score":47.566,"warning":null},{"id":"vals-vibe-code-bench-v1.1:kimi/kimi-k2.7-code","modelId":"kimi-k2.7-code","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Kimi K2.7 Code (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.7-code\",\"evaluatorModel\":\"kimi/kimi-k2.7-code\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":32768,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":\"OpenHands\"}","score":47.211,"warning":null},{"id":"vals-vibe-code-bench-v1.1:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":\"OpenHands\"}","score":46.391,"warning":null},{"id":"vals-vibe-code-bench-v1.1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":\"OpenHands\"}","score":42.174,"warning":null},{"id":"vals-vibe-code-bench-v1.1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":\"OpenHands\"}","score":37.891,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":37.161,"warning":null},{"id":"vals-vibe-code-bench-v1.1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":\"OpenHands\"}","score":34.113,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":32.034,"warning":null},{"id":"vals-vibe-code-bench-v1.1:zai/glm-5.1","modelId":"glm-5.1","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"GLM 5.1 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.1\",\"evaluatorModel\":\"zai/glm-5.1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":\"OpenHands\"}","score":31.456,"warning":null},{"id":"vals-vibe-code-bench-v1.1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-vibe-code-bench-v1.1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":\"OpenHands\"}","score":30.759,"warning":null},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":\"OpenHands\"}","score":26.097,"warning":null},{"id":"vals-vibe-code-bench-v1.1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":\"OpenHands\"}","score":25.565,"warning":null},{"id":"vals-vibe-code-bench-v1.1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":\"OpenHands\"}","score":20.204,"warning":null},{"id":"vals-vibe-code-bench-v1.1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":\"OpenHands\"}","score":19.403,"warning":null},{"id":"vals-vibe-code-bench-v1.1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":196608,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":\"OpenHands\"}","score":11.926,"warning":null},{"id":"vals-vibe-code-bench-v1.1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":\"OpenHands\"}","score":11.393,"warning":null},{"id":"vals-vibe-code-bench-v1.1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-vibe-code-bench-v1.1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":\"OpenHands\"}","score":4.063,"warning":null},{"id":"vals-vibe-code-bench-v1.1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-vibe-code-bench-v1.1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":\"OpenHands\"}","score":2.888,"warning":null},{"id":"vals-legal-agent-benchmark-v1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-legal-agent-benchmark-v1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":25.417,"warning":null},{"id":"vals-legal-agent-benchmark-v1:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-legal-agent-benchmark-v1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":22.917,"warning":null},{"id":"vals-legal-agent-benchmark-v1:meta/muse-wonder-full","modelId":"muse-wonder-full","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Wonder Full (max)","configuration":"{\"checkpoint\":\"straw-110_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder-full\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":21.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:meta/muse_spark_1_1","modelId":"muse-spark-1.1","benchmarkId":"vals-legal-agent-benchmark-v1","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"{\"checkpoint\":\"muse_spark_1_1\",\"evaluatorModel\":\"meta/muse_spark_1_1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":20,"warning":null},{"id":"vals-legal-agent-benchmark-v1:google/gemini-4-argon","modelId":"gemini-4-argon","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"{\"checkpoint\":\"gemini-4-argon\",\"evaluatorModel\":\"google/gemini-4-argon\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":19.583,"warning":null},{"id":"vals-legal-agent-benchmark-v1:meta/muse-wonder","modelId":"muse-wonder","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Wonder (max)","configuration":"{\"checkpoint\":\"restful-rf120_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":17.083,"warning":null},{"id":"vals-legal-agent-benchmark-v1:mistralai/mistral-large-4","modelId":"mistral-large-4","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Mistral Large 4 (high)","configuration":"{\"checkpoint\":\"mistral-large-4\",\"evaluatorModel\":\"mistralai/mistral-large-4\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":15.833,"warning":"Vals reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"vals-legal-agent-benchmark-v1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":15.833,"warning":null},{"id":"vals-legal-agent-benchmark-v1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":12.917,"warning":null},{"id":"vals-legal-agent-benchmark-v1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":12.917,"warning":null},{"id":"vals-legal-agent-benchmark-v1:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-legal-agent-benchmark-v1","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":12.5,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":11.25,"warning":"Source discloses an Opus 4.8 refusal fallback; mixed checkpoint result excluded from ranking"},{"id":"vals-legal-agent-benchmark-v1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":10,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":9.583,"warning":null},{"id":"vals-legal-agent-benchmark-v1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":8.75,"warning":null},{"id":"vals-legal-agent-benchmark-v1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":8.333,"warning":null},{"id":"vals-legal-agent-benchmark-v1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":8.333,"warning":null},{"id":"vals-legal-agent-benchmark-v1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":7.5,"warning":null},{"id":"vals-legal-agent-benchmark-v1:zai/glm-5.2","modelId":"glm-5.2","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"GLM 5.2 (max)","configuration":"{\"checkpoint\":\"glm-5.2\",\"evaluatorModel\":\"zai/glm-5.2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":7.083,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":6.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":6.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":6.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":6.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":5.417,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":5,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":5,"warning":null},{"id":"vals-legal-agent-benchmark-v1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":4.167,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":3.75,"warning":null},{"id":"vals-legal-agent-benchmark-v1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":3.333,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":2.917,"warning":null},{"id":"vals-legal-agent-benchmark-v1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":2.5,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":2.5,"warning":null},{"id":"vals-legal-agent-benchmark-v1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":2.083,"warning":null},{"id":"vals-legal-agent-benchmark-v1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":1.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":1.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":1.667,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":1.25,"warning":null},{"id":"vals-legal-agent-benchmark-v1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":1.25,"warning":null},{"id":"vals-legal-agent-benchmark-v1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":0.833,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-legal-agent-benchmark-v1","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":0.833,"warning":null},{"id":"vals-legal-agent-benchmark-v1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":0.417,"warning":null},{"id":"vals-legal-agent-benchmark-v1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":0.417,"warning":null},{"id":"vals-legal-agent-benchmark-v1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-legal-agent-benchmark-v1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:zai/glm-5.1","modelId":"glm-5.1","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"GLM 5.1 (unspecified)","configuration":"{\"checkpoint\":\"glm-5.1\",\"evaluatorModel\":\"zai/glm-5.1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-legal-agent-benchmark-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-legal-agent-benchmark-v1","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":0,"warning":null},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","benchmarkId":"vals-legal-agent-benchmark-v1","level":"xhigh","sourceLabel":"GPT 5.4 (xhigh) (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-2026-03-05\",\"evaluatorModel\":\"openai/gpt-5.4-2026-03-05\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":0,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Fable 5.1 (max)","configuration":"{\"checkpoint\":\"claude-fable-5-1\",\"evaluatorModel\":\"anthropic/claude-fable-5-1\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":77.642,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-4-argon","modelId":"gemini-4-argon","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 4 Argon (high)","configuration":"{\"checkpoint\":\"gemini-4-argon\",\"evaluatorModel\":\"google/gemini-4-argon\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":76.231,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-opus-5","modelId":"claude-opus-5","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Opus 5 (max)","configuration":"{\"checkpoint\":\"claude-opus-5\",\"evaluatorModel\":\"anthropic/claude-opus-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":75.06,"warning":null},{"id":"vals-tax-agent-bench-v1:meta/muse-wonder-full","modelId":"muse-wonder-full","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Wonder Full (max)","configuration":"{\"checkpoint\":\"straw-110_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder-full\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":75.001,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-sonnet-5-5","modelId":"claude-sonnet-5-5","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Sonnet 5.5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":73.395,"warning":null},{"id":"vals-tax-agent-bench-v1:zai/glm-5.3","modelId":"glm-5.3","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GLM 5.3 (max)","configuration":"{\"checkpoint\":\"glm-5.3\",\"evaluatorModel\":\"zai/glm-5.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":73.089,"warning":null},{"id":"vals-tax-agent-bench-v1:meta/muse_spark_1_3","modelId":"muse-spark-1.3","benchmarkId":"vals-tax-agent-bench-v1","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.3\",\"evaluatorModel\":\"meta/muse_spark_1_3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":71.934,"warning":null},{"id":"vals-tax-agent-bench-v1:meta/muse-wonder","modelId":"muse-wonder","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Wonder (max)","configuration":"{\"checkpoint\":\"restful-rf120_evals_ext\",\"evaluatorModel\":\"meta/muse-wonder\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":71.161,"warning":null},{"id":"vals-tax-agent-bench-v1:grok/grok-4.6","modelId":"grok-4.6","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Grok 4.6 (high)","configuration":"{\"checkpoint\":\"grok-4.6\",\"evaluatorModel\":\"grok/grok-4.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":70.788,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-fable-5","modelId":"claude-fable-5","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Fable 5 (max)","configuration":"{\"checkpoint\":\"claude-fable-5\",\"evaluatorModel\":\"anthropic/claude-fable-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":69.822,"warning":null},{"id":"vals-tax-agent-bench-v1:kimi/kimi-k3","modelId":"kimi-k3","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"{\"checkpoint\":\"kimi-k3\",\"evaluatorModel\":\"kimi/kimi-k3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":262144,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":68.672,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-5.6-sol","modelId":"gpt-5.6-sol","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GPT-5.6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-5.6-sol\",\"evaluatorModel\":\"openai/gpt-5.6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":67.955,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-3.8-flash","modelId":"gemini-3.8-flash","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 3.8 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.8-flash\",\"evaluatorModel\":\"google/gemini-3.8-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":66.771,"warning":null},{"id":"vals-tax-agent-bench-v1:grok/grok-4.7","modelId":"grok-4.7","benchmarkId":"vals-tax-agent-bench-v1","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh)","configuration":"{\"checkpoint\":\"grok-4.7\",\"evaluatorModel\":\"grok/grok-4.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":65.596,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-5.6-terra","modelId":"gpt-5.6-terra","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GPT-5.6 Terra (max)","configuration":"{\"checkpoint\":\"gpt-5.6-terra\",\"evaluatorModel\":\"openai/gpt-5.6-terra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":65.199,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-opus-4-8","modelId":"claude-opus-4-8","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-8\",\"evaluatorModel\":\"anthropic/claude-opus-4-8\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":64.906,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-sonnet-4-6","modelId":"claude-sonnet-4-6","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Sonnet 4.6 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-4-6\",\"evaluatorModel\":\"anthropic/claude-sonnet-4-6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":true,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":64.854,"warning":null},{"id":"vals-tax-agent-bench-v1:grok/grok-4.5","modelId":"grok-4.5","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Grok 4.5 (high)","configuration":"{\"checkpoint\":\"grok-4.5\",\"evaluatorModel\":\"grok/grok-4.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":64.323,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-opus-4-7","modelId":"claude-opus-4-7","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Opus 4.7 (max)","configuration":"{\"checkpoint\":\"claude-opus-4-7\",\"evaluatorModel\":\"anthropic/claude-opus-4-7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":63.864,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GPT-6 Astra (max)","configuration":"{\"checkpoint\":\"gpt-6-astra\",\"evaluatorModel\":\"openai/gpt-6-astra\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":63.335,"warning":null},{"id":"vals-tax-agent-bench-v1:mistralai/mistral-large-4","modelId":"mistral-large-4","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Mistral Large 4 (high)","configuration":"{\"checkpoint\":\"mistral-large-4\",\"evaluatorModel\":\"mistralai/mistral-large-4\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":256000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":63.294,"warning":"Vals reports High, but the official Mistral Large 4 model card does not establish a named High API effort setting. This is an evaluator configuration label."},{"id":"vals-tax-agent-bench-v1:deepseek/deepseek-v4-flash-0731","modelId":"deepseek-v4-flash","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"DeepSeek V4 Flash 0731 (high)","configuration":"{\"checkpoint\":\"deepseek-v4-flash\",\"evaluatorModel\":\"deepseek/deepseek-v4-flash-0731\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":62.563,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-6.1-sol","modelId":"gpt-6.1-sol","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GPT-6.1 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6.1-sol\",\"evaluatorModel\":\"openai/gpt-6.1-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":62.31,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-sonnet-5","modelId":"claude-sonnet-5","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"{\"checkpoint\":\"claude-sonnet-5\",\"evaluatorModel\":\"anthropic/claude-sonnet-5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":\"max\",\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":62.273,"warning":null},{"id":"vals-tax-agent-bench-v1:zai/glm-5.3-flash","modelId":"glm-5.3-flash","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"{\"checkpoint\":\"glm-5.3-flash\",\"evaluatorModel\":\"zai/glm-5.3-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Zhipu AI\",\"harness\":null}","score":61.956,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-5.6-luna","modelId":"gpt-5.6-luna","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GPT-5.6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-5.6-luna\",\"evaluatorModel\":\"openai/gpt-5.6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":60.815,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-5.5","modelId":"gpt-5.5","benchmarkId":"vals-tax-agent-bench-v1","level":"xhigh","sourceLabel":"GPT 5.5 (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.5\",\"evaluatorModel\":\"openai/gpt-5.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":60.456,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GPT-6 Luna (max)","configuration":"{\"checkpoint\":\"gpt-6-luna\",\"evaluatorModel\":\"openai/gpt-6-luna\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":58.864,"warning":null},{"id":"vals-tax-agent-bench-v1:deepseek/deepseek-v4-pro-0813","modelId":"deepseek-v4-pro","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"{\"checkpoint\":\"deepseek-v4-pro\",\"evaluatorModel\":\"deepseek/deepseek-v4-pro-0813\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":384000,\"verbosity\":null,\"provider\":\"DeepSeek\",\"harness\":null}","score":58.663,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-3.7-flash","modelId":"gemini-3.7-flash","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 3.7 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.7-flash\",\"evaluatorModel\":\"google/gemini-3.7-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":57.665,"warning":null},{"id":"vals-tax-agent-bench-v1:meta/muse_spark_1_2","modelId":"muse-spark-1.2","benchmarkId":"vals-tax-agent-bench-v1","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"{\"checkpoint\":\"muse-spark-1.2\",\"evaluatorModel\":\"meta/muse_spark_1_2\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":131072,\"verbosity\":null,\"provider\":\"Meta\",\"harness\":null}","score":56.86,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-3.5-flash","modelId":"gemini-3.5-flash","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash\",\"evaluatorModel\":\"google/gemini-3.5-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":53.526,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"vals-tax-agent-bench-v1","level":"max","sourceLabel":"GPT-6 Sol (max)","configuration":"{\"checkpoint\":\"gpt-6-sol\",\"evaluatorModel\":\"openai/gpt-6-sol\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"max\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":53.045,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-3.6-flash","modelId":"gemini-3.6-flash","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 3.6 Flash (high)","configuration":"{\"checkpoint\":\"gemini-3.6-flash\",\"evaluatorModel\":\"google/gemini-3.6-flash\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":49.895,"warning":null},{"id":"vals-tax-agent-bench-v1:minimax/MiniMax-M3","modelId":"minimax-m3","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"MiniMax-M3 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M3\",\"evaluatorModel\":\"minimax/MiniMax-M3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":512000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":49.691,"warning":null},{"id":"vals-tax-agent-bench-v1:alibaba/qwen3.7-max","modelId":"qwen3.7-max","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Max (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-max\",\"evaluatorModel\":\"alibaba/qwen3.7-max\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":47.407,"warning":null},{"id":"vals-tax-agent-bench-v1:kimi/kimi-k2.6","modelId":"kimi-k2.6","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"Kimi K2.6 (unspecified)","configuration":"{\"checkpoint\":\"kimi-k2.6\",\"evaluatorModel\":\"kimi/kimi-k2.6\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Moonshot AI\",\"harness\":null}","score":45.135,"warning":null},{"id":"vals-tax-agent-bench-v1:alibaba/qwen3.7-plus","modelId":"qwen3.7-plus","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"Qwen 3.7 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.7-plus\",\"evaluatorModel\":\"alibaba/qwen3.7-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":38.713,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-3.1-pro-preview","modelId":"gemini-3.1-pro","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 3.1 Pro Preview (02/26) (high)","configuration":"{\"checkpoint\":\"gemini-3.1-pro-preview\",\"evaluatorModel\":\"google/gemini-3.1-pro-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":37.842,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-3-flash-preview","modelId":"gemini-3-flash-preview","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 3 Flash (12/25) (high)","configuration":"{\"checkpoint\":\"gemini-3-flash-preview\",\"evaluatorModel\":\"google/gemini-3-flash-preview\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":37.557,"warning":null},{"id":"vals-tax-agent-bench-v1:google/gemini-3.5-flash-lite","modelId":"gemini-3.5-flash-lite","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Gemini 3.5 Flash Lite (high)","configuration":"{\"checkpoint\":\"gemini-3.5-flash-lite\",\"evaluatorModel\":\"google/gemini-3.5-flash-lite\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Google\",\"harness\":null}","score":37.451,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-5.4-mini-2026-03-17","modelId":"gpt-5.4-mini","benchmarkId":"vals-tax-agent-bench-v1","level":"xhigh","sourceLabel":"GPT 5.4 Mini (xhigh)","configuration":"{\"checkpoint\":\"gpt-5.4-mini-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-mini-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"xhigh\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":36.586,"warning":null},{"id":"vals-tax-agent-bench-v1:xiaomi/mimo-v2.5-pro","modelId":"mimo-v2.5-pro","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"MiMo V2.5 Pro (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5-pro\",\"evaluatorModel\":\"xiaomi/mimo-v2.5-pro\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":34.787,"warning":null},{"id":"vals-tax-agent-bench-v1:grok/grok-4.3","modelId":"grok-4.3","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Grok 4.3 (high)","configuration":"{\"checkpoint\":\"grok-4.3\",\"evaluatorModel\":\"grok/grok-4.3\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":null,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":34.416,"warning":null},{"id":"vals-tax-agent-bench-v1:alibaba/qwen3.6-plus","modelId":"qwen3.6-plus","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"Qwen 3.6 Plus (unspecified)","configuration":"{\"checkpoint\":\"qwen3.6-plus\",\"evaluatorModel\":\"alibaba/qwen3.6-plus\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":null,\"max_output_tokens\":65536,\"verbosity\":null,\"provider\":\"Alibaba\",\"harness\":null}","score":32.844,"warning":null},{"id":"vals-tax-agent-bench-v1:grok/grok-4.20-0309-reasoning","modelId":"grok-4.20-0309-reasoning","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"Grok 4.20 (Reasoning) (unspecified)","configuration":"{\"checkpoint\":\"grok-4.20-0309-reasoning\",\"evaluatorModel\":\"grok/grok-4.20-0309-reasoning\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":2000000,\"verbosity\":null,\"provider\":\"SpaceXAI\",\"harness\":null}","score":31.241,"warning":null},{"id":"vals-tax-agent-bench-v1:xiaomi/mimo-v2.5","modelId":"mimo-v2.5","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"MiMo V2.5 (unspecified)","configuration":"{\"checkpoint\":\"mimo-v2.5\",\"evaluatorModel\":\"xiaomi/mimo-v2.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"Xiaomi\",\"harness\":null}","score":29.163,"warning":null},{"id":"vals-tax-agent-bench-v1:anthropic/claude-haiku-4-5-20251001-thinking","modelId":"claude-haiku-4-5-20251001","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"Claude Haiku 4.5 (Thinking) (unspecified)","configuration":"{\"checkpoint\":\"claude-haiku-4-5-20251001\",\"evaluatorModel\":\"anthropic/claude-haiku-4-5-20251001-thinking\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":null,\"max_output_tokens\":64000,\"verbosity\":null,\"provider\":\"Anthropic\",\"harness\":null}","score":28.314,"warning":null},{"id":"vals-tax-agent-bench-v1:mistralai/mistral-medium-3.5","modelId":"mistral-medium-3.5","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"Mistral Medium 3.5 (high)","configuration":"{\"checkpoint\":\"mistral-medium-3.5\",\"evaluatorModel\":\"mistralai/mistral-medium-3.5\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":0.7,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"Mistral AI\",\"harness\":null}","score":27.291,"warning":null},{"id":"vals-tax-agent-bench-v1:openai/gpt-5.4-nano-2026-03-17","modelId":"gpt-5.4-nano","benchmarkId":"vals-tax-agent-bench-v1","level":"high","sourceLabel":"GPT 5.4 Nano (high)","configuration":"{\"checkpoint\":\"gpt-5.4-nano-2026-03-17\",\"evaluatorModel\":\"openai/gpt-5.4-nano-2026-03-17\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":\"high\",\"compute_effort\":null,\"reasoning\":null,\"temperature\":null,\"top_p\":null,\"max_output_tokens\":128000,\"verbosity\":null,\"provider\":\"OpenAI\",\"harness\":null}","score":26.578,"warning":null},{"id":"vals-tax-agent-bench-v1:minimax/MiniMax-M2.7","modelId":"minimax-m2.7","benchmarkId":"vals-tax-agent-bench-v1","level":"unspecified","sourceLabel":"MiniMax-M2.7 (unspecified)","configuration":"{\"checkpoint\":\"MiniMax-M2.7\",\"evaluatorModel\":\"minimax/MiniMax-M2.7\",\"sourceUpdatedOn\":\"2026-10-16\",\"reasoning_effort\":null,\"compute_effort\":null,\"reasoning\":null,\"temperature\":1,\"top_p\":0.95,\"max_output_tokens\":80000,\"verbosity\":null,\"provider\":\"MiniMax\",\"harness\":null}","score":25.382,"warning":null},{"id":"36d6df353b16e177cd44e3ec","modelId":"gpt-6-astra","benchmarkId":"real-swe-gpt-6-astra-codex-cli","level":"high","sourceLabel":"GPT-6 Astra · Codex CLI","configuration":"GPT-6 Astra; native harness: Codex CLI; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":46.25,"warning":null},{"id":"d886f6a08b775c8731f741ca","modelId":"claude-fable-5-1","benchmarkId":"real-swe-fable-5-1-claude-code","level":"high","sourceLabel":"Fable 5.1 · Claude Code","configuration":"Fable 5.1; native harness: Claude Code; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":45,"warning":null},{"id":"3d5f434c3cbe85ca3b4ccb8d","modelId":"gemini-3.8-flash","benchmarkId":"real-swe-gemini-3-8-flash-gemini-cli","level":"high","sourceLabel":"Gemini 3.8 Flash · Gemini CLI","configuration":"Gemini 3.8 Flash; native harness: Gemini CLI; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":38.75,"warning":null},{"id":"a98dea2e9467129aeca8c8e1","modelId":"glm-5.3","benchmarkId":"real-swe-glm-5-3-claude-code","level":"high","sourceLabel":"GLM 5.3 · Claude Code","configuration":"GLM 5.3; native harness: Claude Code; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":37.5,"warning":null},{"id":"f07a75cfe72239d502ceb6ac","modelId":"muse-spark-1.3","benchmarkId":"real-swe-muse-spark-1-3-muse-code","level":"high","sourceLabel":"Muse Spark 1.3 · Muse Code","configuration":"Muse Spark 1.3; native harness: Muse Code; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":36.25,"warning":null},{"id":"bf75f254142209a0515e1cc0","modelId":"grok-4.6","benchmarkId":"real-swe-grok-4-6-grok-build","level":"high","sourceLabel":"Grok 4.6 · Grok Build","configuration":"Grok 4.6; native harness: Grok Build; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":32.5,"warning":null},{"id":"3db0287be8f5aec2207ea3d8","modelId":"gpt-5.6-sol","benchmarkId":"real-swe-gpt-5-6-sol-codex-cli","level":"high","sourceLabel":"GPT-5.6 Sol · Codex CLI","configuration":"GPT-5.6 Sol; native harness: Codex CLI; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":26.25,"warning":null},{"id":"1e52efc02b7e8795cc3d9569","modelId":"kimi-k3","benchmarkId":"real-swe-kimi-k3-kimi-code","level":"high","sourceLabel":"Kimi K3 · Kimi Code","configuration":"Kimi K3; native harness: Kimi Code; high reasoning; pass@1 averaged over 8 runs; 10 private tasks; board label September 2026. This score ranks the model with that harness.","score":20,"warning":null},{"id":"aa-cyber-index-claude-fable-5-1","modelId":"claude-fable-5-1","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"Claude Fable 5.1 (Max, Default Fallback)","configuration":"Artificial Analysis Stirrup; v1.0; Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback); source label Claude Fable 5.1 (Max, Default Fallback); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.508333333333333; refusal 0.15966386554621848; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.23461876950156; refusal 0; CyberGym-E2E-AA pass@1 0.007633587786259542; refusal 0.9923664122137404; https://artificialanalysis.ai/models/claude-fable-5-1; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":25.0195230207051,"warning":null},{"id":"aa-cyber-index-claude-opus-5-5","modelId":"claude-opus-5-5","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"Claude Opus 5.5 (Max, Default Fallback)","configuration":"Artificial Analysis Stirrup; v1.0; Claude Opus 5.5 (Adaptive Reasoning, Max Effort, Default Fallback); source label Claude Opus 5.5 (Max, Default Fallback); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.583333333333333; refusal 0.09166666666666666; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.26894153661864; refusal 0; CyberGym-E2E-AA pass@1 0.007633587786259542; refusal 0.9847328244274809; https://artificialanalysis.ai/models/claude-opus-5-5; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":28.6636152579411,"warning":null},{"id":"aa-cyber-index-deepseek-v4-1-flash","modelId":"deepseek-v4.1-flash","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"DeepSeek V4.1 Flash (Max)","configuration":"Artificial Analysis Stirrup; v1.0; DeepSeek V4.1 Flash (Max); source label DeepSeek V4.1 Flash (Max); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.683333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.31199859972578; refusal 0; CyberGym-E2E-AA pass@1 0.22900763358778625; refusal 0; https://artificialanalysis.ai/models/deepseek-v4-1-flash; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":40.81131888823,"warning":null},{"id":"aa-cyber-index-gemini-3-8-flash","modelId":"gemini-3.8-flash","benchmarkId":"aa-cyber-index","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Artificial Analysis Stirrup; v1.0; Gemini 3.8 Flash (high); source label Gemini 3.8 Flash (High); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.525; refusal 0.09166666666666666; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.145379023883697; refusal 0; CyberGym-E2E-AA pass@1 0.08396946564885496; refusal 0.8625954198473282; https://artificialanalysis.ai/models/gemini-3-8-flash; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":25.1449496510851,"warning":null},{"id":"aa-cyber-index-glm-5-3","modelId":"glm-5.3","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"GLM-5.3 (Max)","configuration":"Artificial Analysis Stirrup; v1.0; GLM-5.3 (max); source label GLM-5.3 (Max); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.558333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.242861348248454; refusal 0; CyberGym-E2E-AA pass@1 0.2900763358778626; refusal 0; https://artificialanalysis.ai/models/glm-5-3; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":36.3757005819883,"warning":null},{"id":"aa-cyber-index-glm-5-3-flash","modelId":"glm-5.3-flash","benchmarkId":"aa-cyber-index","level":"unspecified","sourceLabel":"GLM-5.3-Flash","configuration":"Artificial Analysis Stirrup; v1.0; GLM-5.3-Flash; source label GLM-5.3-Flash; mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.558333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.198915233995969; refusal 0; CyberGym-E2E-AA pass@1 0.7404580152671756; refusal 0; https://artificialanalysis.ai/models/glm-5-3-flash; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":49.9235527532159,"warning":null},{"id":"aa-cyber-index-gpt-6-astra","modelId":"gpt-6-astra","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"GPT-6 Astra (Max)","configuration":"Artificial Analysis Stirrup; v1.0; GPT-6 Astra (max); source label GPT-6 Astra (Max); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.633333333333333; refusal 0.13333333333333333; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.369261477045908; refusal 0; CyberGym-E2E-AA pass@1 0; refusal 1; https://artificialanalysis.ai/models/gpt-6-astra; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":33.4198270126414,"warning":null},{"id":"aa-cyber-index-gpt-6-luna","modelId":"gpt-6-luna","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"GPT-6 Luna (Max)","configuration":"Artificial Analysis Stirrup; v1.0; GPT-6 Luna (max); source label GPT-6 Luna (Max); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.566666666666667; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.235268161578884; refusal 0; CyberGym-E2E-AA pass@1 0.7786259541984732; refusal 0; https://artificialanalysis.ai/models/gpt-6-luna; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":52.6853594148008,"warning":null},{"id":"aa-cyber-index-gpt-6-sol","modelId":"gpt-6-sol","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"GPT-6 Sol (Max)","configuration":"Artificial Analysis Stirrup; v1.0; GPT-6 Sol (max); source label GPT-6 Sol (Max); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.641666666666667; refusal 0.09166666666666666; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.463158172139356; refusal 0; CyberGym-E2E-AA pass@1 0; refusal 1; https://artificialanalysis.ai/models/gpt-6-sol; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":36.8274946268674,"warning":null},{"id":"aa-cyber-index-grok-4-7","modelId":"grok-4.7","benchmarkId":"aa-cyber-index","level":"xhigh","sourceLabel":"Grok 4.7 (Xhigh)","configuration":"Artificial Analysis Stirrup; v1.0; Grok 4.7 (xhigh); source label Grok 4.7 (Xhigh); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.683333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.269178891978752; refusal 0; CyberGym-E2E-AA pass@1 0.7404580152671756; refusal 0; https://artificialanalysis.ai/models/grok-4-7; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":56.432341352642,"warning":null},{"id":"aa-cyber-index-kimi-k3","modelId":"kimi-k3","benchmarkId":"aa-cyber-index","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Artificial Analysis Stirrup; v1.0; Kimi K3 (max); source label Kimi K3 (Max); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.533333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.124543341082697; refusal 0; CyberGym-E2E-AA pass@1 0.5801526717557252; refusal 0; https://artificialanalysis.ai/models/kimi-k3; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":41.2676448723918,"warning":null},{"id":"aa-cyber-index-mimo-v2-6-pro","modelId":"mimo-v2.6-pro","benchmarkId":"aa-cyber-index","level":"unspecified","sourceLabel":"MiMo-V2.6-Pro","configuration":"Artificial Analysis Stirrup; v1.0; MiMo-V2.6-Pro; source label MiMo-V2.6-Pro; mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.633333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.26237518589335; refusal 0; CyberGym-E2E-AA pass@1 0.7862595419847328; refusal 0; https://artificialanalysis.ai/models/mimo-v2-6-pro; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":56.0656020403805,"warning":null},{"id":"aa-cyber-index-minimax-m3","modelId":"minimax-m3","benchmarkId":"aa-cyber-index","level":"unspecified","sourceLabel":"MiniMax-M3","configuration":"Artificial Analysis Stirrup; v1.0; MiniMax-M3; source label MiniMax-M3; mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.416666666666667; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.0106990014265335; refusal 0; CyberGym-E2E-AA pass@1 0.24427480916030533; refusal 0; https://artificialanalysis.ai/models/minimax-m3; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":22.3880159084502,"warning":null},{"id":"aa-cyber-index-mistral-large-4","modelId":"mistral-large-4","benchmarkId":"aa-cyber-index","level":"unspecified","sourceLabel":"Mistral Large 4 Preview","configuration":"Artificial Analysis Stirrup; v1.0; Mistral Large 4 Preview; source label Mistral Large 4 Preview; mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.508333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.160214245442586; refusal 0; CyberGym-E2E-AA pass@1 0.816793893129771; refusal 0; https://artificialanalysis.ai/models/mistral-large-4; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":49.5113823968563,"warning":null},{"id":"aa-cyber-index-muse-spark-1-3-xhigh","modelId":"muse-spark-1.3","benchmarkId":"aa-cyber-index","level":"xhigh","sourceLabel":"Muse Spark 1.3 (Xhigh)","configuration":"Artificial Analysis Stirrup; v1.0; Muse Spark 1.3 (xhigh); source label Muse Spark 1.3 (Xhigh); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.608333333333333; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.125391849529781; refusal 0; CyberGym-E2E-AA pass@1 0.5954198473282443; refusal 0; https://artificialanalysis.ai/models/muse-spark-1-3-xhigh; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":44.3048343397119,"warning":null},{"id":"aa-cyber-index-nvidia-nemotron-3-ultra-550b-a55b","modelId":"nemotron-3-ultra","benchmarkId":"aa-cyber-index","level":"reasoning","sourceLabel":"Nemotron 3 Ultra 550B A55B (Reasoning)","configuration":"Artificial Analysis Stirrup; v1.0; Nemotron 3 Ultra 550B A55B (Reasoning); source label Nemotron 3 Ultra 550B A55B (Reasoning); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.275; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.010482180293501; refusal 0; CyberGym-E2E-AA pass@1 0.10687022900763359; refusal 0; https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":13.0784136433712,"warning":null},{"id":"aa-cyber-index-qwen3-8-27b","modelId":"qwen3.8-27b","benchmarkId":"aa-cyber-index","level":"xhigh","sourceLabel":"Qwen3.8 27B (Xhigh)","configuration":"Artificial Analysis Stirrup; v1.0; Qwen3.8 27B (xhigh); source label Qwen3.8 27B (Xhigh); mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.266666666666667; refusal 0.6134453781512605; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.115062761506276; refusal 0.09090909090909091; CyberGym-E2E-AA pass@1 0; refusal 1; https://artificialanalysis.ai/models/qwen3-8-27b; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":12.7243142724314,"warning":null},{"id":"aa-cyber-index-step-5","modelId":"step-5-preview","benchmarkId":"aa-cyber-index","level":"unspecified","sourceLabel":"Step 5 Preview","configuration":"Artificial Analysis Stirrup; v1.0; Step 5 Preview; source label Step 5 Preview; mitigation standard; public model; source field cyberIndex; index 0–100; CWE-Bench-AA pass@1 0.525; refusal 0; DeepsecBench-AA median F2, 3 repeats, GPT-5.6 Sol High judge 0.158112846668792; refusal 0; CyberGym-E2E-AA pass@1 0.7480916030534351; refusal 0; https://artificialanalysis.ai/models/step-5; page sha256 344ca66f5aa3f3abd1901b12bd29db73950d90e553fcb3ae900dcf590d7a6db2; encrypted payload sha256 b5e6ae259181eaed471e288c4bbdbd1088438e0a81caf19ec0fa60f7092d48f7; Display only. This composite measures defensive source-code tasks. Successful capability scores depend on safety refusals, and refusal rates are reported separately. It does not enter weighted Overall.","score":47.7068149907409,"warning":null},{"id":"cb2e3515b80b12af3303866a","modelId":"claude-fable-5-1","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"Claude Fable 5.1","configuration":"Claude Fable 5.1; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Official board truncates the FasterGCG run at 11.9 hours after a safety-filter cutoff. The published AUARC includes that cutoff. Display only. Excluded from weighted Overall.","score":0.602,"warning":null},{"id":"ea1ff84c5a751f4d8b772b62","modelId":"claude-opus-5","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"Claude Opus 5","configuration":"Claude Opus 5; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Display only. Excluded from weighted Overall.","score":0.579,"warning":null},{"id":"1b012811a9b967f7ffd89f2c","modelId":"gpt-5.6-sol","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"GPT-5.6 Sol","configuration":"GPT-5.6 Sol; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Display only. Excluded from weighted Overall.","score":0.513,"warning":null},{"id":"7ecb39c599db2cdf2d237740","modelId":"kimi-k3","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"Kimi K3","configuration":"Kimi K3; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Display only. Excluded from weighted Overall.","score":0.435,"warning":null},{"id":"668de7b6d6dc4a5d877776b7","modelId":"qwen-3.8-max","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Display only. Excluded from weighted Overall.","score":0.423,"warning":null},{"id":"50c9d59622809091b00ba9d6","modelId":"gemini-3.8-flash","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"Gemini 3.8 Flash","configuration":"Gemini 3.8 Flash; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Display only. Excluded from weighted Overall.","score":0.484,"warning":null},{"id":"5be8f052b58e62435776e382","modelId":"muse-spark-1.3","benchmarkId":"autoresearchexam","level":"xhigh","sourceLabel":"Muse Spark 1.3","configuration":"Muse Spark 1.3; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Source names xhigh. Max was not available for these rollouts. Display only. Excluded from weighted Overall.","score":0.379,"warning":null},{"id":"d383f75d80bc4b00f394602f","modelId":"grok-4.6","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"Grok 4.6","configuration":"Grok 4.6; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Display only. Excluded from weighted Overall.","score":0.493,"warning":null},{"id":"a1a12c7c5eca0c65ca4cc66b","modelId":"gpt-6-astra","benchmarkId":"autoresearchexam","level":"unspecified","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; Terminus 2; Harbor; 24-hour window; hidden-test AUARC; 29 tasks; board snapshot 2026-09-09. Highest effort available at rollout; the board does not name the knob. Display only. Excluded from weighted Overall.","score":0.6,"warning":null},{"id":"d069057c75d5f66dec371d2d","modelId":"gpt-6-astra","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"GPT-6 Astra (max effort) - mean of 3","configuration":"GPT-6 Astra; Codex CLI; effort max (first_party); source ceiling. Mean of 3 runs; fixed 45 of 105 planted bugs. partial 1; claimed_only 0; extras 55; evaluation 2026-09-14. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: Every figure on this row is a MEAN of three runs, carried to one decimal - 45.0 of 105. The three runs scored 48, 40 and 47: a range of 8, which is the widest spread measured on any row above 30 here and is wider than the gap between this [source caveat truncated].","score":42.86,"warning":null},{"id":"e95242c2a0ee491639498eaa","modelId":"gpt-5.6-sol","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"GPT-5.6 Sol (max effort) - mean of 2","configuration":"GPT-5.6 Sol; Codex CLI; effort max (first_party); source ceiling. Mean of 2 runs; fixed 43.5 of 105 planted bugs. partial 2; claimed_only 0.5; extras 48.5; evaluation 2026-09-16. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: TWO RUNS IS TWO POINTS. The range between them is 7 of 105, which is wider than the gap between several adjacent rows on this board, so treat 43.5 as an estimate with a known spread rather than a settled number - and note the repo figures [source caveat truncated].","score":41.43,"warning":null},{"id":"79867e42f55ea4a657c228d6","modelId":"gpt-6-astra","benchmarkId":"bughunt-bench","level":"xhigh","sourceLabel":"GPT-6 Astra (xhigh effort)","configuration":"GPT-6 Astra; Codex CLI; effort xhigh (first_party). Single run; fixed 43 of 105 planted bugs. partial 1; claimed_only 1; extras 53; evaluation 2026-09-04. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored.","score":40.95,"warning":null},{"id":"e3421f54ae0ea255729f0eff","modelId":"claude-opus-5-5","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"Opus 5.5 (max effort)","configuration":"Opus 5.5; Claude Code; effort max (first_party); source ceiling. Single run; fixed 43 of 105 planted bugs. partial 3; claimed_only 3; extras 9; evaluation 2026-09-22. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: ONE RUN, not a mean - and this board has already measured a 13-point spread between two runs of one identical max-effort configuration, so treat 43 as a single draw, not a settled level. Runs a 1M context window with zero compactions; [source caveat truncated].","score":40.95,"warning":null},{"id":"69b8475f557646e661eaa00d","modelId":"claude-fable-5-1","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"Fable 5.1 (max effort)","configuration":"Fable 5.1; Claude Code; effort max (first_party); source ceiling. Single run; fixed 43 of 105 planted bugs. partial 2; claimed_only 4; extras 11; evaluation 2026-09-01. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored.","score":40.95,"warning":null},{"id":"1ca7abeef4fe3150af253415","modelId":"gpt-5.6-sol","benchmarkId":"bughunt-bench","level":"high","sourceLabel":"GPT-5.6 Sol (re-run Jul 31)","configuration":"GPT-5.6 Sol; Codex CLI; effort high (first_party). Single run; fixed 34 of 105 planted bugs. partial 4; claimed_only 0; extras 28; evaluation 2026-07-31. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored.","score":32.38,"warning":null},{"id":"4b93ad5f00ede5ee6f7eab30","modelId":"claude-fable-5-1","benchmarkId":"bughunt-bench","level":"high","sourceLabel":"Fable 5.1","configuration":"Fable 5.1; Claude Code; effort high (first_party). Single run; fixed 33 of 105 planted bugs. partial 2; claimed_only 0; extras 7; evaluation 2026-09-01. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored.","score":31.43,"warning":null},{"id":"4ec2e25ea3dec7936c3149e9","modelId":"muse-spark-1.3","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"Muse Spark 1.3 (max effort, Muse Code, Meta API) - mean of 5","configuration":"Muse Spark 1.3; Muse Code / Meta API; effort max (verified_ceiling); source ceiling. Mean of 5 runs; fixed 32.2 of 105 planted bugs. partial 1.6; claimed_only 1.2; extras 24; evaluation 2026-09-17. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: FIVE RUNS, RANGE 6. The spread is the finding, not the mean: at n=3 this arm looked tight (range 2), and two more runs tripled the error bar. Treat 32.2 as an estimate with a wide known spread. TWO OF THE FIVE RUNS ARE SPLIT-VERSION - the [source caveat truncated].","score":30.67,"warning":null},{"id":"b1bb14b9843b3977bed0580f","modelId":"gpt-5.6-luna","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"GPT-5.6 Luna (max effort) - mean of 3","configuration":"GPT-5.6 Luna; Codex CLI; effort max (first_party); source ceiling. Mean of 3 runs; fixed 31.3 of 105 planted bugs. partial 3.3; claimed_only 0; extras 54.7; evaluation 2026-09-22. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: Three runs of an identical configuration spanning 30 to 33 of 105 - read the mean with that width in mind, not as a point. Cost is a TOKEN ESTIMATE at list rates, not a bill, and is comparable only to other list-priced rows. Runs shared [source caveat truncated].","score":29.81,"warning":null},{"id":"aefc1e5e687f5fccd2b5b878","modelId":"grok-4.7","benchmarkId":"bughunt-bench","level":"xhigh","sourceLabel":"Grok 4.7 (xhigh) seq - mean of 4","configuration":"Grok 4.7; Grok Build CLI (ACP); effort xhigh (verified); source ceiling. Mean of 4 runs; fixed 28.8 of 105 planted bugs. partial 1.3; claimed_only 0.2; extras 31.8; evaluation 2026-09-21. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: A mean of FOUR runs against a 4.6 row that is a mean of three - the samples are not the same size. The four span 25 to 31 of 105 against 4.6's 27 to 30, so the tie still rests on the wider distribution. Cost uses a per-session [source caveat truncated].","score":27.43,"warning":null},{"id":"cfc2b44701cc8afdf833989a","modelId":"grok-4.6","benchmarkId":"bughunt-bench","level":"xhigh","sourceLabel":"Grok 4.6 (xhigh) seq - mean of 3","configuration":"Grok 4.6; Grok Build CLI (ACP); effort xhigh (verified); source ceiling. Mean of 3 runs; fixed 28.7 of 105 planted bugs. partial 2.7; claimed_only 0.7; extras 18; evaluation 2026-09-14. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: A mean of three runs, not a single measurement; the three span 27 to 30. This row SUPERSEDES the single-run sequential row (27/105) published 2026-08-28, which was the LOWEST of the three. Every figure on this row is a MEAN of the three [source caveat truncated].","score":27.33,"warning":null},{"id":"b222fcaa4ad0fedda81b2c5b","modelId":"claude-opus-5","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"Opus 5 (max effort)","configuration":"Opus 5; Claude Code; effort max (first_party); source ceiling. Single run; fixed 27 of 105 planted bugs. partial 2; claimed_only 5; extras 2; evaluation 2026-08-01. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored.","score":25.71,"warning":null},{"id":"8c81bb270658bc572ac23413","modelId":"qwen-3.8-max","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"Qwen3.8-Max (max effort, Alibaba API) - mean of 3","configuration":"Qwen3.8-Max; Claude Code / Alibaba API; effort max (first_party). Mean of 3 runs; fixed 25.7 of 105 planted bugs. partial 1.4; claimed_only 2; extras 7; evaluation 2026-09-15. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: Superseded row: the single 28/105 run published 2026-09-11 stays on the board, marked superseded, not deleted - it is the best of the three, and the mean is the number the board leads with per the standing rule that a rung measured three [source caveat truncated].","score":24.48,"warning":null},{"id":"d718dd7c1ee1a7bd0a102862","modelId":"kimi-k3","benchmarkId":"bughunt-bench","level":"default","sourceLabel":"Kimi K3","configuration":"Kimi K3; Claude Code / OpenRouter; effort default (inert_default); source ceiling. Single run; fixed 21 of 105 planted bugs. partial 0; claimed_only 6; extras 6; evaluation 2026-07-26. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored.","score":20,"warning":null},{"id":"78a9c78f4cb3495ab080ba7c","modelId":"claude-opus-5","benchmarkId":"bughunt-bench","level":"high","sourceLabel":"Opus 5","configuration":"Opus 5; Claude Code; effort high (first_party). Single run; fixed 21 of 105 planted bugs. partial 0; claimed_only 3; extras 6; evaluation 2026-07-26. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored.","score":20,"warning":null},{"id":"98c6c82eda4d51ad59d6d651","modelId":"muse-spark-1.3","benchmarkId":"bughunt-bench","level":"xhigh","sourceLabel":"Muse Spark 1.3 (xhigh effort, Muse Code, Meta API) - mean of 3","configuration":"Muse Spark 1.3; Muse Code / Meta API; effort xhigh (verified). Mean of 3 runs; fixed 20.3 of 105 planted bugs. partial 0.7; claimed_only 0; extras 13.3; evaluation 2026-09-14. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: A mean of three runs, not a single measurement; the three span 16 to 25. This row SUPERSEDES the single-run xhigh row (20/105) published 2026-09-13, which landed within a point of the mean - the supersession is about method, not a [source caveat truncated].","score":19.33,"warning":null},{"id":"7f7d2d73ed6a8716bfedc645","modelId":"glm-5.3","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"GLM-5.3 (max effort, Z.ai API)","configuration":"GLM-5.3; Claude Code / Z.ai API; effort max (first_party); source ceiling. Single run; fixed 19 of 105 planted bugs. partial 0; claimed_only 0; extras 2; evaluation 2026-09-10. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: One run on each side of the comparison. Two single runs landing on the same total is not a demonstration that the paths are equivalent - the board has watched one model at one setting score 16, 13 and 17 - and the matching totals here are [source caveat truncated].","score":18.1,"warning":null},{"id":"119b4de295f86b5a42fe9bf2","modelId":"gemini-3.8-flash","benchmarkId":"bughunt-bench","level":"high","sourceLabel":"Gemini 3.8 Flash - mean of 3","configuration":"Gemini 3.8 Flash; Antigravity CLI; effort high (verified_ceiling); source ceiling. Mean of 3 runs; fixed 18 of 105 planted bugs. partial 1.3; claimed_only 0.3; extras 4; evaluation 2026-09-15. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: A mean of three runs, not a single measurement; the three span 15 to 20. This row SUPERSEDES the single-run row (20/105) published 2026-09-02, which was the HIGHEST of the three; the supersession is about method, not a correction. Every [source caveat truncated].","score":17.14,"warning":null},{"id":"c454aaeeca53fc8beee03ad8","modelId":"glm-5.3-flash","benchmarkId":"bughunt-bench","level":"max","sourceLabel":"GLM-5.3 Flash (max effort, Z.ai API) - mean of 3","configuration":"GLM-5.3 Flash; Claude Code / Z.ai API; effort max (verified_ceiling); source ceiling. Mean of 3 runs; fixed 17.7 of 105 planted bugs. partial 1; claimed_only 1.3; extras 3.3; evaluation 2026-09-15. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: A mean of three runs, not a single measurement; the three span 16 to 19. This row SUPERSEDES the single-run max row (19/105) published 2026-09-12, which was the HIGHEST of the three; the supersession is about method, not a correction. [source caveat truncated].","score":16.86,"warning":null},{"id":"4c3732f9bf8749e94adb0d68","modelId":"qwen3.8-27b","benchmarkId":"bughunt-bench","level":"xhigh","sourceLabel":"Qwen3.8-27B (xhigh effort, Alibaba API) - mean of 3","configuration":"Qwen3.8-27B; Claude Code / Alibaba API; effort xhigh (first_party). Mean of 3 runs; fixed 15 of 105 planted bugs. partial 0.7; claimed_only 4.3; extras 2.3; evaluation 2026-09-15. Score is (fixed / 105) * 100. Extras, partials and claimed-only are not scored. Source caveat: Every figure on this row is a MEAN of three runs, carried to one decimal - 15.0 of 105, range 4 (13 to 17). This row SUPERSEDES the single-run row (15/105) published 2026-09-12, which landed exactly on this mean by coincidence of one draw, [source caveat truncated].","score":14.29,"warning":null},{"id":"8013f488929a27c953d0969a","modelId":"claude-opus-5-5","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Claude Opus 5.5","configuration":"Claude Opus 5.5; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":24.7,"warning":null},{"id":"e58469b1ddb1dd8c54509141","modelId":"gpt-6-astra","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"GPT 6 Astra","configuration":"GPT 6 Astra; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":11.6,"warning":null},{"id":"3a062742525c19ec510ae326","modelId":"claude-fable-5-1","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Claude Fable 5.1","configuration":"Claude Fable 5.1; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":9.6,"warning":null},{"id":"b51f4f94cce16b585d3c6596","modelId":"gemini-4-argon","benchmarkId":"dayjob-healthcare","level":"high","sourceLabel":"Gemini 4 Argon","configuration":"Gemini 4 Argon; effort high. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":9.6,"warning":null},{"id":"df9df7eaca39913a5b4ca84c","modelId":"claude-opus-5","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Claude Opus 5","configuration":"Claude Opus 5; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":8.4,"warning":null},{"id":"2c3b1aa370c48a623350ef1a","modelId":"grok-4.7","benchmarkId":"dayjob-healthcare","level":"xhigh","sourceLabel":"Grok 4.7","configuration":"Grok 4.7; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":8.4,"warning":null},{"id":"8778b1b5186749b45aa39c62","modelId":"muse-spark-1.3","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Muse Spark 1.3","configuration":"Muse Spark 1.3; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":7.8,"warning":null},{"id":"10715c1dd8c16e77f6575e4f","modelId":"claude-fable-5","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Claude Fable 5","configuration":"Claude Fable 5; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":5.6,"warning":null},{"id":"b5b40be041ab8147dd1281c5","modelId":"gpt-6-sol","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"GPT 6 Sol","configuration":"GPT 6 Sol; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":3.6,"warning":null},{"id":"23a57717b22b7f3394b62313","modelId":"grok-4.6","benchmarkId":"dayjob-healthcare","level":"xhigh","sourceLabel":"Grok 4.6","configuration":"Grok 4.6; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2.8,"warning":null},{"id":"c7f06f5e3c3bedd1a2ef4c69","modelId":"muse-spark-1.2","benchmarkId":"dayjob-healthcare","level":"xhigh","sourceLabel":"Muse Spark 1.2","configuration":"Muse Spark 1.2; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2,"warning":null},{"id":"ed78cae115b8991a86c3e524","modelId":"qwen-3.8-max","benchmarkId":"dayjob-healthcare","level":"xhigh","sourceLabel":"Qwen 3.8 Max","configuration":"Qwen 3.8 Max; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2,"warning":null},{"id":"c697fccffb7cd3fbe0c3f595","modelId":"gpt-5.6-sol","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"GPT 5.6 Sol","configuration":"GPT 5.6 Sol; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":1.6,"warning":null},{"id":"23f334ec20ec3b7754b09fc2","modelId":"gpt-5.6-terra","benchmarkId":"dayjob-healthcare","level":"xhigh","sourceLabel":"GPT 5.6 Terra","configuration":"GPT 5.6 Terra; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":1.6,"warning":null},{"id":"e770c30699b657336304ff1d","modelId":"kimi-k3","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Kimi K3","configuration":"Kimi K3; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":1.2,"warning":null},{"id":"e81945732df3cbf866f89ad9","modelId":"gemini-3.8-flash","benchmarkId":"dayjob-healthcare","level":"high","sourceLabel":"Gemini 3.8 Flash","configuration":"Gemini 3.8 Flash; effort high. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0.8,"warning":null},{"id":"b904deb8ba6d1f0fbb5abf52","modelId":"hy4-preview","benchmarkId":"dayjob-healthcare","level":"high","sourceLabel":"Hy4 Preview","configuration":"Hy4 Preview; effort high. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0.8,"warning":null},{"id":"0dcf485cda88a796ea217e9b","modelId":"glm-5.3","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"GLM 5.3","configuration":"GLM 5.3; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0.4,"warning":null},{"id":"30a1b9fb05051a44e7b29e97","modelId":"glm-5.3-flash","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"GLM 5.3 Flash","configuration":"GLM 5.3 Flash; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0.4,"warning":null},{"id":"cc05393a564054b0ba49074f","modelId":"gpt-5.6-luna","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"GPT 5.6 Luna","configuration":"GPT 5.6 Luna; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0.4,"warning":null},{"id":"db7d58ab530db9a06359328c","modelId":"hy3","benchmarkId":"dayjob-healthcare","level":"high","sourceLabel":"Hy3","configuration":"Hy3; effort high. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0.4,"warning":null},{"id":"571df1e8692ba33af9730e5e","modelId":"claude-sonnet-5","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Claude Sonnet 5","configuration":"Claude Sonnet 5; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"21a80c1cc36483d8cbbb51c3","modelId":"gemini-3.7-flash","benchmarkId":"dayjob-healthcare","level":"high","sourceLabel":"Gemini 3.7 Flash","configuration":"Gemini 3.7 Flash; effort high. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"f8657c77aa8f68ed4c4caf98","modelId":"gemini-3.1-pro","benchmarkId":"dayjob-healthcare","level":"high","sourceLabel":"Gemini 3.1 Pro","configuration":"Gemini 3.1 Pro; effort high. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"42f706e6aed03b32501d07b6","modelId":"kimi-k2.7-code","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"cc67f59bc6a200eabb3e48dd","modelId":"muse-glimmer-30b","benchmarkId":"dayjob-healthcare","level":"xhigh","sourceLabel":"Muse Glimmer 30B","configuration":"Muse Glimmer 30B; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"c79b4e01f0296958f4e65108","modelId":"nemotron-3-ultra","benchmarkId":"dayjob-healthcare","level":"unspecified","sourceLabel":"Nemotron 3 Ultra","configuration":"Nemotron 3 Ultra; effort unspecified. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"4d180b6e6141701640fba711","modelId":"mistral-large-3","benchmarkId":"dayjob-healthcare","level":"unspecified","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; effort unspecified. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"9a074a7174126349ffd8c560","modelId":"gpt-6-luna","benchmarkId":"dayjob-healthcare","level":"max","sourceLabel":"GPT 6 Luna","configuration":"GPT 6 Luna; effort max. Strict pass@1 percent, tau = 1, five attempts, 50 DAYJOB: Healthcare tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"f51c91469a44f016098a8082","modelId":"claude-opus-5-5","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Claude Opus 5.5","configuration":"Claude Opus 5.5; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":23.9,"warning":null},{"id":"ac82b48e4e980b527e69036f","modelId":"gpt-6-astra","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"GPT 6 Astra","configuration":"GPT 6 Astra; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":21.5,"warning":null},{"id":"55267a53a60ce4470d84800f","modelId":"gemini-4-argon","benchmarkId":"dayjob-finance","level":"high","sourceLabel":"Gemini 4 Argon","configuration":"Gemini 4 Argon; effort high. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":20.3,"warning":null},{"id":"6f15b12f701a09c117318c08","modelId":"claude-fable-5-1","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Claude Fable 5.1","configuration":"Claude Fable 5.1; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":19.8,"warning":null},{"id":"3ac96c565d7d95c5322117f2","modelId":"muse-spark-1.3","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Muse Spark 1.3","configuration":"Muse Spark 1.3; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":14.8,"warning":null},{"id":"2f1659e33b93ef2e6c349481","modelId":"grok-4.7","benchmarkId":"dayjob-finance","level":"xhigh","sourceLabel":"Grok 4.7","configuration":"Grok 4.7; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":14.5,"warning":null},{"id":"ad304890a5734df3813236b3","modelId":"grok-4.6","benchmarkId":"dayjob-finance","level":"xhigh","sourceLabel":"Grok 4.6","configuration":"Grok 4.6; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":12.8,"warning":null},{"id":"7a1c2608ea8895d1a4d7b471","modelId":"claude-opus-5","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Claude Opus 5","configuration":"Claude Opus 5; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":11.3,"warning":null},{"id":"a01c1873f75e2a70abdf61e2","modelId":"gpt-6-sol","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"GPT 6 Sol","configuration":"GPT 6 Sol; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":9.3,"warning":null},{"id":"a6a3b9b6d8edb7dd7f9ddbc7","modelId":"claude-fable-5","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Claude Fable 5","configuration":"Claude Fable 5; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":9,"warning":null},{"id":"319cd5622cf0976b6dbcc4b7","modelId":"hy4-preview","benchmarkId":"dayjob-finance","level":"high","sourceLabel":"Hy4 Preview","configuration":"Hy4 Preview; effort high. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":7.3,"warning":null},{"id":"36778994cf17b13267691d55","modelId":"glm-5.3","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"GLM 5.3","configuration":"GLM 5.3; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":6,"warning":null},{"id":"025e1dec2153bb55b486a00e","modelId":"gpt-5.6-sol","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"GPT 5.6 Sol","configuration":"GPT 5.6 Sol; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":5.5,"warning":null},{"id":"e8b9d521c56996897b3ce54a","modelId":"kimi-k3","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Kimi K3","configuration":"Kimi K3; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":3.8,"warning":null},{"id":"2a81aed5bec1a34551da1b5f","modelId":"qwen-3.8-max","benchmarkId":"dayjob-finance","level":"xhigh","sourceLabel":"Qwen 3.8 Max","configuration":"Qwen 3.8 Max; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":3,"warning":null},{"id":"af9cc3681ea9343205ff64e9","modelId":"gemini-3.7-flash","benchmarkId":"dayjob-finance","level":"high","sourceLabel":"Gemini 3.7 Flash","configuration":"Gemini 3.7 Flash; effort high. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2.8,"warning":null},{"id":"c93e77782f6d7094c410a0d5","modelId":"gemini-3.8-flash","benchmarkId":"dayjob-finance","level":"high","sourceLabel":"Gemini 3.8 Flash","configuration":"Gemini 3.8 Flash; effort high. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2.8,"warning":null},{"id":"c147538da8246c3f2fdf28b3","modelId":"glm-5.3-flash","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"GLM 5.3 Flash","configuration":"GLM 5.3 Flash; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2.3,"warning":null},{"id":"73aabc3298ba34589e56d013","modelId":"muse-spark-1.2","benchmarkId":"dayjob-finance","level":"xhigh","sourceLabel":"Muse Spark 1.2","configuration":"Muse Spark 1.2; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2.3,"warning":null},{"id":"c4873c69d18cf220189ce836","modelId":"claude-sonnet-5","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Claude Sonnet 5","configuration":"Claude Sonnet 5; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2.3,"warning":null},{"id":"a638443a7f29edaab94f9f5c","modelId":"gpt-6-luna","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"GPT 6 Luna","configuration":"GPT 6 Luna; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":2.3,"warning":null},{"id":"d694f8a18fee8ea1437b2810","modelId":"gpt-5.6-luna","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"GPT 5.6 Luna","configuration":"GPT 5.6 Luna; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":1.8,"warning":null},{"id":"50a7dd1087d296dce7264938","modelId":"gpt-5.6-terra","benchmarkId":"dayjob-finance","level":"xhigh","sourceLabel":"GPT 5.6 Terra","configuration":"GPT 5.6 Terra; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":1.8,"warning":null},{"id":"a9b2e868353b5c8513d0ee62","modelId":"hy3","benchmarkId":"dayjob-finance","level":"high","sourceLabel":"Hy3","configuration":"Hy3; effort high. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0.3,"warning":null},{"id":"3cfecdce0dc5c8bb730f6237","modelId":"gemini-3.1-pro","benchmarkId":"dayjob-finance","level":"high","sourceLabel":"Gemini 3.1 Pro","configuration":"Gemini 3.1 Pro; effort high. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"ef0ac274d47dee038a98960f","modelId":"kimi-k2.7-code","benchmarkId":"dayjob-finance","level":"max","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; effort max. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"044f47e30edf2bd7ef2f7e11","modelId":"mistral-large-3","benchmarkId":"dayjob-finance","level":"unspecified","sourceLabel":"Mistral Large 3","configuration":"Mistral Large 3; effort unspecified. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"1508d434a180819441288e92","modelId":"muse-glimmer-30b","benchmarkId":"dayjob-finance","level":"xhigh","sourceLabel":"Muse Glimmer 30B","configuration":"Muse Glimmer 30B; effort xhigh. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"383e806eadeaa5b8ecdf2a7b","modelId":"nemotron-3-ultra","benchmarkId":"dayjob-finance","level":"unspecified","sourceLabel":"Nemotron 3 Ultra","configuration":"Nemotron 3 Ultra; effort unspecified. Strict pass@1 percent, tau = 1, five attempts, 80 DAYJOB: Finance tasks. An attempt passes only when every binary rubric criterion is met. OpenHands SDK 1.43.1 in Harbor 0.22.0. Judge: Claude Opus 4.8.","score":0,"warning":null},{"id":"4ded310383d6b785107f1ade","modelId":"claude-opus-5-5","benchmarkId":"sudo-l7","level":"max","sourceLabel":"Claude Opus 5.5","configuration":"Claude Opus 5.5; effort max; agent Claude Code; 60 tasks.","score":41.7,"warning":null},{"id":"fbc0f0f865af728782bed96d","modelId":"claude-sonnet-5-5","benchmarkId":"sudo-l7","level":"max","sourceLabel":"Claude Sonnet 5.5","configuration":"Claude Sonnet 5.5; effort max; agent Claude Code; 60 tasks.","score":41.7,"warning":null},{"id":"00219f3f31423184f4551a76","modelId":"gpt-6-astra","benchmarkId":"sudo-l7","level":"max","sourceLabel":"GPT 6 Astra","configuration":"GPT 6 Astra; effort max; agent Codex; 60 tasks.","score":33.9,"warning":null},{"id":"dc0a5cb0ca2545c23c7b5471","modelId":"claude-fable-5-1","benchmarkId":"sudo-l7","level":"max","sourceLabel":"Claude Fable 5.1","configuration":"Claude Fable 5.1; effort max; agent Claude Code; 60 tasks.","score":31.1,"warning":null},{"id":"d787b9d0f4ebffdda7ed4fc6","modelId":"gpt-6.1-sol","benchmarkId":"sudo-l7","level":"max","sourceLabel":"GPT 6.1 Sol","configuration":"GPT 6.1 Sol; effort max; agent Codex; 60 tasks.","score":28.9,"warning":null},{"id":"40b376f16dbe985410e92967","modelId":"qwen-3.8-max","benchmarkId":"sudo-l7","level":"xhigh","sourceLabel":"Qwen 3.8 Max","configuration":"Qwen 3.8 Max; effort xhigh; agent Qwen Code; 60 tasks.","score":20.6,"warning":null},{"id":"3e501f4d0b20047e61d77985","modelId":"gemini-4-argon","benchmarkId":"sudo-l7","level":"high","sourceLabel":"Gemini 4 Argon","configuration":"Gemini 4 Argon; effort high; agent Antigravity; 60 tasks.","score":18.9,"warning":null},{"id":"e1e12d053abcdc3d71fc670d","modelId":"grok-4.7","benchmarkId":"sudo-l7","level":"xhigh","sourceLabel":"Grok 4.7","configuration":"Grok 4.7; effort xhigh; agent Grok Build; 60 tasks.","score":15.6,"warning":null},{"id":"56b7367f12ba25eff56d498e","modelId":"muse-wonder","benchmarkId":"sudo-l7","level":"max","sourceLabel":"Muse Wonder","configuration":"Muse Wonder; effort max; agent Muse Code; 60 tasks.","score":15.6,"warning":null},{"id":"26e073377f2d99d8a5641e53","modelId":"glm-5.3","benchmarkId":"sudo-l7","level":"max","sourceLabel":"GLM 5.3","configuration":"GLM 5.3; effort max; agent OpenCode; 60 tasks.","score":14.4,"warning":null},{"id":"8d3e2266956b474121a94b0b","modelId":"muse-spark-1.3","benchmarkId":"sudo-l7","level":"max","sourceLabel":"Muse Spark 1.3","configuration":"Muse Spark 1.3; effort max; agent Muse Code; 60 tasks.","score":13.9,"warning":null},{"id":"82e3d518d84c3fa1ed18811f","modelId":"kimi-k3","benchmarkId":"sudo-l7","level":"max","sourceLabel":"Kimi K3","configuration":"Kimi K3; effort max; agent OpenCode; 60 tasks.","score":13.3,"warning":null},{"id":"65a34d1fa2ecb6776ad932e8","modelId":"hy4-preview","benchmarkId":"sudo-l7","level":"high","sourceLabel":"Hy4 Preview","configuration":"Hy4 Preview; effort high; agent OpenCode; 60 tasks.","score":12.8,"warning":null},{"id":"9de87fb166bcde7c385fe449","modelId":"gemini-3.8-flash","benchmarkId":"sudo-l7","level":"high","sourceLabel":"Gemini 3.8 Flash","configuration":"Gemini 3.8 Flash; effort high; agent Antigravity; 60 tasks.","score":11.7,"warning":null},{"id":"3d2825650bc7c062cb4fcaa5","modelId":"mistral-large-4","benchmarkId":"sudo-l7","level":"high","sourceLabel":"Mistral Large 4","configuration":"Mistral Large 4; effort high; agent OpenCode; 60 tasks.","score":11.7,"warning":null},{"id":"93e8c9c93a2b86f54cc4ddfe","modelId":"glm-5.3-flash","benchmarkId":"toolathlon-verified","level":"max","sourceLabel":"GLM 5.3 Flash (max)","configuration":"GLM 5.3 Flash (max); Open-Weights; Default agent. Pass@1 78.4 ± 1.9; 108 tasks; evaluation 2026-08-30. Pass@3 not published; Pass^3 not published; turns not published; tool calls not published. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":78.4,"warning":null},{"id":"ed142fb4c94a739efadbde83","modelId":"kimi-k3","benchmarkId":"toolathlon-verified","level":"max","sourceLabel":"Kimi K3 (max)","configuration":"Kimi K3 (max); Open-Weights; Default agent. Pass@1 76.5 ± 1.9; 108 tasks; evaluation 2026-07-16. Pass@3 83.3; Pass^3 68.5; turns 22.8; tool calls 39.1. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":76.5,"warning":null},{"id":"307df7fc7a3d2d3ff13aace2","modelId":"claude-opus-4-8","benchmarkId":"toolathlon-verified","level":"max","sourceLabel":"Claude Opus 4.8 (max)","configuration":"Claude Opus 4.8 (max); Proprietary; Default agent. Pass@1 76.2 ± 3.4; 108 tasks; evaluation 2026-06-30. Pass@3 84.3; Pass^3 66.7; turns 19.9; tool calls 36.3. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":76.2,"warning":null},{"id":"b05cfbac1378b456dd269c8e","modelId":"muse-spark-1.2","benchmarkId":"toolathlon-verified","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"Muse Spark 1.2 (xhigh); Proprietary; Default agent. Pass@1 75.9 ± 1.3; 108 tasks; evaluation 2026-08-05. Pass@3 87.0; Pass^3 63.0; turns 44.2; tool calls 48.6. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":75.9,"warning":null},{"id":"4387ffd488958a7423b7bfdc","modelId":"muse-spark-1.1","benchmarkId":"toolathlon-verified","level":"xhigh","sourceLabel":"Muse Spark 1.1 (xhigh)","configuration":"Muse Spark 1.1 (xhigh); Proprietary; Default agent. Pass@1 75.6 ± 0.4; 108 tasks; evaluation 2026-07-09. Pass@3 86.1; Pass^3 63.9; turns 28.9; tool calls 42.7. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":75.6,"warning":null},{"id":"a121b744f81681eeb0a5bdcd","modelId":"deepseek-v4-pro","benchmarkId":"toolathlon-verified","level":"max","sourceLabel":"DeepSeek V4 Pro 0813 (max)","configuration":"DeepSeek V4 Pro 0813 (max); Open-Weights; Default agent. Pass@1 74.4 ± 1.2; 108 tasks; evaluation 2026-08-13. Pass@3 not published; Pass^3 not published; turns not published; tool calls not published. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":74.4,"warning":null},{"id":"0dccc75a81b29bb448bbb01f","modelId":"gpt-5.5","benchmarkId":"toolathlon-verified","level":"xhigh","sourceLabel":"GPT-5.5 (xhigh)","configuration":"GPT-5.5 (xhigh); Proprietary; Default agent. Pass@1 73.5 ± 1.2; 108 tasks; evaluation 2026-06-30. Pass@3 82.4; Pass^3 62.0; turns 24.4; tool calls 37.8. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":73.5,"warning":null},{"id":"93d29ea51280a4052f57e0b0","modelId":"claude-sonnet-5","benchmarkId":"toolathlon-verified","level":"max","sourceLabel":"Claude Sonnet 5 (max)","configuration":"Claude Sonnet 5 (max); Proprietary; Default agent. Pass@1 71.6 ± 1.2; 108 tasks; evaluation 2026-07-01. Pass@3 83.3; Pass^3 53.7; turns 27.1; tool calls 51.5. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":71.6,"warning":null},{"id":"f0e924dd1959da90f31e2e52","modelId":"deepseek-v4-flash","benchmarkId":"toolathlon-verified","level":"max","sourceLabel":"DeepSeek V4 Flash 0731 (max)","configuration":"DeepSeek V4 Flash 0731 (max); Open-Weights; Default agent. Pass@1 70.7 ± 0.9; 108 tasks; evaluation 2026-07-31. Pass@3 81.5; Pass^3 58.3; turns 26.3; tool calls 49.0. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":70.7,"warning":null},{"id":"988bef362bd1da11693ac376","modelId":"gemini-3.5-flash","benchmarkId":"toolathlon-verified","level":"high","sourceLabel":"Gemini 3.5 Flash (high)","configuration":"Gemini 3.5 Flash (high); Proprietary; Default agent. Pass@1 67.3 ± 1.2; 108 tasks; evaluation 2026-06-30. Pass@3 79.6; Pass^3 53.7; turns 45.3; tool calls 45.6. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":67.3,"warning":null},{"id":"f05f6616344b6f716ee1b57b","modelId":"gemini-3.1-pro","benchmarkId":"toolathlon-verified","level":"high","sourceLabel":"Gemini 3.1 Pro (high)","configuration":"Gemini 3.1 Pro (high); Proprietary; Default agent. Pass@1 61.1 ± 1.3; 108 tasks; evaluation 2026-07-01. Pass@3 72.2; Pass^3 48.1; turns 26.2; tool calls 32.2. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":61.1,"warning":null},{"id":"4c831ab57740da7b10d0478f","modelId":"glm-5.2","benchmarkId":"toolathlon-verified","level":"max","sourceLabel":"GLM 5.2 (max)","configuration":"GLM 5.2 (max); Open-Weights; Default agent. Pass@1 59.9 ± 1.9; 108 tasks; evaluation 2026-06-30. Pass@3 78.7; Pass^3 41.7; turns 23.3; tool calls 41.9. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":59.9,"warning":null},{"id":"5b23fe5f72e981524de657d2","modelId":"kimi-k2.6","benchmarkId":"toolathlon-verified","level":"unspecified","sourceLabel":"Kimi K2.6","configuration":"Kimi K2.6; Open-Weights; Default agent. Pass@1 58.0 ± 4.9; 108 tasks; evaluation 2026-07-15. Pass@3 72.2; Pass^3 41.7; turns 30.8; tool calls 42.3. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":58,"warning":null},{"id":"f18d0bf828c5790b39fd9ee1","modelId":"kimi-k2.7-code","benchmarkId":"toolathlon-verified","level":"unspecified","sourceLabel":"Kimi K2.7 Code","configuration":"Kimi K2.7 Code; Open-Weights; Default agent. Pass@1 58.0 ± 4.3; 108 tasks; evaluation 2026-06-30. Pass@3 74.1; Pass^3 41.7; turns 23.1; tool calls 37.6. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":58,"warning":null},{"id":"ce84a5798f0f7841fc7a0de5","modelId":"gemini-3.5-flash-lite","benchmarkId":"toolathlon-verified","level":"high","sourceLabel":"Gemini 3.5 Flash-Lite (high)","configuration":"Gemini 3.5 Flash-Lite (high); Proprietary; Default agent. Pass@1 57.1 ± 3.6; 108 tasks; evaluation 2026-07-30. Pass@3 71.3; Pass^3 41.7; turns 37.9; tool calls 38.3. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":57.1,"warning":null},{"id":"f6479b251dbe0fc5d9a2f509","modelId":"hy3","benchmarkId":"toolathlon-verified","level":"high","sourceLabel":"Hy3 (high)","configuration":"Hy3 (high); Open-Weights; Default agent. Pass@1 56.8 ± 3.7; 108 tasks; evaluation 2026-07-30. Pass@3 72.2; Pass^3 40.7; turns 20.5; tool calls 37.9. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":56.8,"warning":null},{"id":"b7828d377cc351ebffaf8ee7","modelId":"mimo-v2.5","benchmarkId":"toolathlon-verified","level":"unspecified","sourceLabel":"MiMo V2.5","configuration":"MiMo V2.5; Open-Weights; Default agent. Pass@1 49.1 ± 3.9; 108 tasks; evaluation 2026-07-30. Pass@3 65.7; Pass^3 30.6; turns 23.2; tool calls 38.5. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":49.1,"warning":null},{"id":"85d0aff28b7c742f5fea4f6d","modelId":"minimax-m2.7","benchmarkId":"toolathlon-verified","level":"unspecified","sourceLabel":"MiniMax M2.7","configuration":"MiniMax M2.7; Open-Weights; Default agent. Pass@1 47.5 ± 3.8; 108 tasks; evaluation 2026-07-30. Pass@3 62.0; Pass^3 32.4; turns 24.6; tool calls 43.2. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":47.5,"warning":null},{"id":"8c4f541f61193e7f74975f3a","modelId":"qwen3.5-397b-a17b","benchmarkId":"toolathlon-verified","level":"unspecified","sourceLabel":"Qwen3.5 397B-A17B","configuration":"Qwen3.5 397B-A17B; Open-Weights; Default agent. Pass@1 40.7 ± 2.0; 108 tasks; evaluation 2026-07-30. Pass@3 51.9; Pass^3 27.8; turns 19.5; tool calls 35.7. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":40.7,"warning":null},{"id":"a01a99f4b87edab5843043fe","modelId":"nemotron-3-ultra","benchmarkId":"toolathlon-verified","level":"unspecified","sourceLabel":"Nemotron 3 Ultra","configuration":"Nemotron 3 Ultra; Open-Weights; Default agent. Pass@1 34.3 ± 1.5; 108 tasks; evaluation 2026-07-15. Pass@3 50.0; Pass^3 19.4; turns 28.6; tool calls 31.9. Headline score is the printed Pass@1 percent. The ± is not the score. The June 30 baseline defines ± as an across-run standard deviation; later rows print ± without a restated trial count, so this row does not store n_runs.","score":34.3,"warning":null},{"id":"928a6287955f80f9c247d66c","modelId":"claude-opus-5-5","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Claude Opus 5.5","configuration":"Claude Opus 5.5; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 67.16. Retail 80.97 (98); Auto insurance 68.40 (100); Travel 54.28 (104); Neobank 71.25 (104); Consulting 61.58 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":67.16,"warning":null},{"id":"9071c8ae8d43dc082d693afa","modelId":"claude-opus-5","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Claude Opus 5","configuration":"Claude Opus 5; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 66.50. Retail 80.71 (98); Auto insurance 65.80 (100); Travel 49.95 (104); Neobank 70.62 (104); Consulting 66.19 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":66.5,"warning":null},{"id":"76bf67b013e8cce12b46f750","modelId":"gpt-5.4","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"GPT-5.4","configuration":"GPT-5.4; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 65.36. Retail 76.33 (98); Auto insurance 62.65 (100); Travel 68.12 (104); Neobank 65.34 (104); Consulting 54.60 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":65.36,"warning":null},{"id":"37ab3536a9071a1c6c69e56f","modelId":"gpt-5.6-sol","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"GPT-5.6 Sol","configuration":"GPT-5.6 Sol; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 61.91. Retail 67.65 (98); Auto insurance 65.30 (100); Travel 60.34 (104); Neobank 59.09 (104); Consulting 57.52 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":61.91,"warning":null},{"id":"a5fda3d68db6727d126c2bc3","modelId":"claude-sonnet-4-6","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Claude Sonnet 4.6","configuration":"Claude Sonnet 4.6; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 59.19. Retail 72.35 (98); Auto insurance 54.40 (100); Travel 58.94 (104); Neobank 56.39 (104); Consulting 54.31 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":59.19,"warning":null},{"id":"6443ce88b910f5f8cec310e4","modelId":"gpt-6-astra","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 58.31. Retail 71.73 (98); Auto insurance 46.55 (100); Travel 55.87 (104); Neobank 60.87 (104); Consulting 56.83 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":58.31,"warning":null},{"id":"721f8b012c30e72f2728dec0","modelId":"gpt-5.2","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"GPT-5.2","configuration":"GPT-5.2; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 46.28. Retail 70.20 (98); Auto insurance 22.40 (100); Travel 53.70 (104); Neobank 51.15 (104); Consulting 34.06 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":46.28,"warning":null},{"id":"12d2562c02df44096210cbea","modelId":"claude-opus-4-6","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Claude Opus 4.6","configuration":"Claude Opus 4.6; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 32.09. Retail 68.62 (98); Auto insurance 8.30 (100); Travel 21.11 (104); Neobank 35.67 (104); Consulting 27.82 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":32.09,"warning":null},{"id":"b53d062aa5e966280e6fda54","modelId":"o3-pro","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"o3-pro","configuration":"o3-pro; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 19.31. Retail 37.70 (98); Auto insurance 2.95 (100); Travel 17.31 (104); Neobank 24.28 (104); Consulting 14.60 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. The August 2026 paper excluded 636 o3-pro system/harness-error trials and published 20.60 overall. This October row is the blog figure and does not restate that exclusion. Display only. Excluded from weighted Overall.","score":19.31,"warning":null},{"id":"fdc2300ebb4c72d77640314f","modelId":"grok-4.3","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Grok-4.3","configuration":"Grok-4.3; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 14.38. Retail 43.93 (98); Auto insurance 2.60 (100); Travel 15.14 (104); Neobank 1.78 (104); Consulting 9.55 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":14.38,"warning":null},{"id":"5b2a06cf1828e45d7c05a243","modelId":"kimi-k3","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Kimi-K3","configuration":"Kimi-K3; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 57.37. Retail 82.24 (98); Auto insurance 50.80 (100); Travel 61.83 (104); Neobank 41.35 (104); Consulting 51.63 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":57.37,"warning":null},{"id":"9256ed2307760a832eed1407","modelId":"qwen3.8-27b","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Qwen3.8-27B","configuration":"Qwen3.8-27B; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 51.70. Retail 64.03 (98); Auto insurance 47.85 (100); Travel 53.41 (104); Neobank 47.88 (104); Consulting 45.69 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":51.7,"warning":null},{"id":"83f5603530c661eb65ae1ec1","modelId":"kimi-k2.6","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Kimi-K2.6","configuration":"Kimi-K2.6; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 37.66. Retail 53.72 (98); Auto insurance 24.50 (100); Travel 39.52 (104); Neobank 33.65 (104); Consulting 37.33 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":37.66,"warning":null},{"id":"33acac6fb60fe4ed590d73f2","modelId":"glm-5.1","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"GLM-5.1","configuration":"GLM-5.1; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 33.19. Retail 58.67 (98); Auto insurance 25.70 (100); Travel 35.43 (104); Neobank 13.27 (104); Consulting 34.06 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":33.19,"warning":null},{"id":"1e899bc63c10d6af15cb3f90","modelId":"qwen3.6-27b","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Qwen3.6-27B","configuration":"Qwen3.6-27B; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 32.94. Retail 43.11 (98); Auto insurance 29.00 (100); Travel 46.39 (104); Neobank 27.84 (104); Consulting 18.37 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":32.94,"warning":null},{"id":"8a9d50b279ed4dfb4489c2aa","modelId":"qwen3.5-9b","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Qwen3.5-9B","configuration":"Qwen3.5-9B; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 5.65. Retail 19.90 (98); Auto insurance 0.70 (100); Travel 4.71 (104); Neobank 1.15 (104); Consulting 2.33 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":5.65,"warning":null},{"id":"c22ff89deca0e83a54901d48","modelId":"mistral-large-3","benchmarkId":"thinkingbox-bench","level":"unspecified","sourceLabel":"Mistral-Large-3","configuration":"Mistral-Large-3; ThinkingBox sandbox; shared prompts, tools, simulated user, and executable terminal-state checks; pass@1; 20 trials on each of 507 tasks; published task-weighted overall 4.66. Retail 11.28 (98); Auto insurance 1.30 (100); Travel 8.99 (104); Neobank 1.15 (104); Consulting 0.74 (101). Blog post 2026-10-03. The post does not name a reasoning-effort dial. Display only. Excluded from weighted Overall.","score":4.66,"warning":null},{"id":"frontiermath-erdos-bloom68-fable-vn2l2xoe9902k9rt","modelId":"claude-fable-5","benchmarkId":"frontiermath-erdos","level":"max","sourceLabel":"claude-fable-5_max","configuration":"Epoch AI Inspect deepagent; one Lean attempt; $300 and 72h; Comparator proof check; task version unpublished; scorer proof_scorer; model=claude-fable-5_max; run=bloom68-fable-vn2l2xoe9902k9rt; started 2026-08-28; displayed proof_scorer 0; mean_score 0; resolved 0 of 68; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":0,"warning":null},{"id":"frontiermath-erdos-bloom68-fable51-8t9up8nw9358oyrl","modelId":"claude-fable-5-1","benchmarkId":"frontiermath-erdos","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI Inspect deepagent; one Lean attempt; $300 and 72h; Comparator proof check; task version unpublished; scorer proof_scorer; model=claude-fable-5-1_max; run=bloom68-fable51-8t9up8nw9358oyrl; started 2026-09-01; displayed proof_scorer 0; mean_score 0; resolved 0 of 68; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":0,"warning":null},{"id":"frontiermath-erdos-bloom68-gpt55-4nh07xafb8lf7jqp","modelId":"gpt-5.5","benchmarkId":"frontiermath-erdos","level":"xhigh","sourceLabel":"gpt-5.5_xhigh","configuration":"Epoch AI Inspect deepagent; one Lean attempt; $300 and 72h; Comparator proof check; task version unpublished; scorer proof_scorer; model=gpt-5.5_xhigh; run=bloom68-gpt55-4nh07xafb8lf7jqp; started 2026-08-28; displayed proof_scorer 0; mean_score 0; resolved 0 of 68; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":0,"warning":null},{"id":"frontiermath-erdos-bloom68-sol-jx4yy8n4naukwwge","modelId":"gpt-5.6-sol","benchmarkId":"frontiermath-erdos","level":"max","sourceLabel":"gpt-5.6-sol_max","configuration":"Epoch AI Inspect deepagent; one Lean attempt; $300 and 72h; Comparator proof check; task version unpublished; scorer proof_scorer; model=gpt-5.6-sol_max; run=bloom68-sol-jx4yy8n4naukwwge; started 2026-08-28; displayed proof_scorer 0; mean_score 0; resolved 0 of 68; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":0,"warning":null},{"id":"frontiermath-erdos-bloom68-ultima-e1dhs43ykrmakree","modelId":"gpt-6-astra","benchmarkId":"frontiermath-erdos","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI Inspect deepagent; one Lean attempt; $300 and 72h; Comparator proof check; task version unpublished; scorer proof_scorer; model=gpt-6-astra_max; run=bloom68-ultima-e1dhs43ykrmakree; started 2026-08-28; displayed proof_scorer 0.03; mean_score 0.029411764705882353; resolved 2 of 68; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":3,"warning":null},{"id":"spireclimb-steambench-sts1-a20h-3a714bccadb811b3","modelId":"claude-fable-5-1","benchmarkId":"spireclimb","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all; setting Slay the Spire 1, Ascension 20 Heart; harness=spireclimb-sts1-a20h; task version 0.3.3_24156a4; scorer win_rate/all; model=claude-fable-5-1_max; run=steambench-sts1-a20h-3a714bccadb811b3; started 2026-09-25; displayed win_rate/all 0; mean_score 0; wins 0 of 60; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":0,"warning":null},{"id":"spireclimb-steambench-sts1-a20h-c8709a73c9fd54e3","modelId":"gpt-6-astra","benchmarkId":"spireclimb","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all; setting Slay the Spire 1, Ascension 20 Heart; harness=spireclimb-sts1-a20h; task version 0.3.3_24156a4; scorer win_rate/all; model=gpt-6-astra_max; run=steambench-sts1-a20h-c8709a73c9fd54e3; started 2026-09-23; displayed win_rate/all 0.17; mean_score 0.16666666666666666; wins 10 of 60; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":17,"warning":null},{"id":"spireclimb-steambench-sts1-a20h-ee8cb2987a2c6c37","modelId":"gpt-6-sol","benchmarkId":"spireclimb","level":"max","sourceLabel":"gpt-6-sol_max","configuration":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all; setting Slay the Spire 1, Ascension 20 Heart; harness=spireclimb-sts1-a20h; task version 0.3.3_24156a4; scorer win_rate/all; model=gpt-6-sol_max; run=steambench-sts1-a20h-ee8cb2987a2c6c37; started 2026-09-24; displayed win_rate/all 0.02; mean_score 0.016666666666666666; wins 1 of 60; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":2,"warning":null},{"id":"spireclimb-steambench-sts1-a20h-fc581bcaa7f17d12","modelId":"gpt-6-luna","benchmarkId":"spireclimb","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all; setting Slay the Spire 1, Ascension 20 Heart; harness=spireclimb-sts1-a20h; task version 0.3.3_24156a4; scorer win_rate/all; model=gpt-6-luna_max; run=steambench-sts1-a20h-fc581bcaa7f17d12; started 2026-09-28; displayed win_rate/all 0; mean_score 0; wins 0 of 60; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":0,"warning":null},{"id":"spireclimb-steambench-sts2-a10-0e148062f8e47c4a","modelId":"gpt-6-luna","benchmarkId":"spireclimb","level":"max","sourceLabel":"gpt-6-luna_max","configuration":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all; setting Slay the Spire 2, Ascension 10; harness=spireclimb-sts2-a10; task version 0.3.3_e52e280; scorer win_rate/all; model=gpt-6-luna_max; run=steambench-sts2-a10-0e148062f8e47c4a; started 2026-09-28; displayed win_rate/all 0; mean_score 0; wins 0 of 75; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":0,"warning":null},{"id":"spireclimb-steambench-sts2-a10-413d1cc930f6d4c6","modelId":"claude-fable-5-1","benchmarkId":"spireclimb","level":"max","sourceLabel":"claude-fable-5-1_max","configuration":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all; setting Slay the Spire 2, Ascension 10; harness=spireclimb-sts2-a10; task version 0.3.3_e52e280; scorer win_rate/all; model=claude-fable-5-1_max; run=steambench-sts2-a10-413d1cc930f6d4c6; started 2026-09-25; displayed win_rate/all 0.04; mean_score 0.04; wins 3 of 75; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":4,"warning":null},{"id":"spireclimb-steambench-sts2-a10-cdf1c9fa85cb6300","modelId":"gpt-6-astra","benchmarkId":"spireclimb","level":"max","sourceLabel":"gpt-6-astra_max","configuration":"Epoch AI computer-use harness; Slay the Spire 2 Ascension 10, 75 runs; Slay the Spire 1 Ascension 20 Heart, 60 runs; scorer win_rate/all; setting Slay the Spire 2, Ascension 10; harness=spireclimb-sts2-a10; task version 0.3.3_e52e280; scorer win_rate/all; model=gpt-6-astra_max; run=steambench-sts2-a10-cdf1c9fa85cb6300; started 2026-09-22; displayed win_rate/all 0.35; mean_score 0.3466666666666667; wins 26 of 75; https://epoch.ai/data/benchmarks.csv; Display only. Excluded from weighted Overall.","score":35,"warning":null},{"id":"9c995be124820e515fae18f2","modelId":"gpt-5.6-sol","benchmarkId":"swe-sweep","level":"xhigh","sourceLabel":"Sol 5.6 (xhigh)","configuration":"Sol 5.6 (xhigh); OpenAI; mini-SWE-agent; effort xhigh. 191 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 4.7%. USD per repository 72.30. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":4.7,"warning":null},{"id":"76d20e5abf2ab8e7bd3c9193","modelId":"claude-opus-5-5","benchmarkId":"swe-sweep","level":"xhigh","sourceLabel":"Opus 5.5 (xhigh)","configuration":"Opus 5.5 (xhigh); Anthropic; mini-SWE-agent; effort xhigh. 147 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 3.6%. USD per repository 21.98. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":3.61,"warning":null},{"id":"349dde94caf88259e8ae0882","modelId":"gpt-5.6-luna","benchmarkId":"swe-sweep","level":"xhigh","sourceLabel":"Luna 5.6 (xhigh)","configuration":"Luna 5.6 (xhigh); OpenAI; mini-SWE-agent; effort xhigh. 103 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 2.5%. USD per repository 2.24. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":2.53,"warning":null},{"id":"a5ecfe2a2bd08a24e15e424e","modelId":"gpt-5.6-terra","benchmarkId":"swe-sweep","level":"xhigh","sourceLabel":"Terra 5.6 (xhigh)","configuration":"Terra 5.6 (xhigh); OpenAI; mini-SWE-agent; effort xhigh. 63 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 1.5%. USD per repository 3.57. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":1.55,"warning":null},{"id":"692fcd7f94d1bcdbe3cf4c17","modelId":"gpt-5.6-luna","benchmarkId":"swe-sweep","level":"high","sourceLabel":"Luna 5.6 (high)","configuration":"Luna 5.6 (high); OpenAI; mini-SWE-agent; effort high. 57 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 1.4%. USD per repository 0.28. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":1.4,"warning":null},{"id":"2fe37b6bde125afef22914e1","modelId":"claude-opus-5","benchmarkId":"swe-sweep","level":"xhigh","sourceLabel":"Opus 5 (xhigh)","configuration":"Opus 5 (xhigh); Anthropic; mini-SWE-agent; effort xhigh. 52 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 1.3%. USD per repository 53.63. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":1.28,"warning":null},{"id":"480bc916d7108ac0ee2030d6","modelId":"kimi-k3","benchmarkId":"swe-sweep","level":"unspecified","sourceLabel":"Kimi K3","configuration":"Kimi K3; Moonshot AI; mini-SWE-agent; The board does not name a reasoning knob for this row. 23 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 0.6%. USD per repository 24.51. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":0.57,"warning":null},{"id":"c8c9452d1622e46697875765","modelId":"gpt-5.6-luna","benchmarkId":"swe-sweep","level":"unspecified","sourceLabel":"Luna 5.6","configuration":"Luna 5.6; OpenAI; mini-SWE-agent; The board does not name a reasoning knob for this row. 21 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 0.5%. USD per repository 0.04. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":0.52,"warning":null},{"id":"5dc7185c20fd87aef971c46f","modelId":"gpt-5.4-mini","benchmarkId":"swe-sweep","level":"high","sourceLabel":"GPT-5.4 Mini (high)","configuration":"GPT-5.4 Mini (high); OpenAI; mini-SWE-agent; effort high. 20 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 0.5%. USD per repository 1.22. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":0.49,"warning":null},{"id":"291eeb8fecc4dc27ef24cc33","modelId":"gpt-5.4-mini","benchmarkId":"swe-sweep","level":"unspecified","sourceLabel":"GPT-5.4 Mini","configuration":"GPT-5.4 Mini; OpenAI; mini-SWE-agent; The board does not name a reasoning knob for this row. 8 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 0.2%. USD per repository 0.05. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":0.2,"warning":null},{"id":"bb238d1c6ede7b766c6c5b7c","modelId":"gemini-3.5-flash-lite","benchmarkId":"swe-sweep","level":"unspecified","sourceLabel":"Gemini 3.5 Flash Lite","configuration":"Gemini 3.5 Flash Lite; Google; mini-SWE-agent; The board does not name a reasoning knob for this row. 5 of 4068 bugs across 100 repositories. Micro-average of bugs resolved. The public table prints 0.1%. USD per repository 0.06. Display only. 12 integer board rows. Unreviewed rows stay unmapped. Excluded from weighted Overall.","score":0.12,"warning":null},{"id":"8f602902f4f283bbb57f68a6","modelId":"claude-opus-5-5","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Claude Code · claude-opus-5-5 · Max","configuration":"Claude Code · claude-opus-5-5 · Max. Full/Overall pass rate 38.2%. 58 passes, 150 runs, 150 tasks scored, 152 split tasks. Partial-credit score 63.2%. Estimated cost $1,340. Runtime 182h 45m. Pass rate is the published passRate field, not passes divided by split tasks.","score":38.2,"warning":null},{"id":"872386cc451a062a3659974a","modelId":"gpt-6-astra","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Codex · gpt-6-astra · Max","configuration":"Codex · gpt-6-astra · Max. Full/Overall pass rate 34.2%. 52 passes, 151 runs, 151 tasks scored, 152 split tasks. Partial-credit score 59.3%. Estimated cost $1,100. Runtime 146h 16m. Pass rate is the published passRate field, not passes divided by split tasks.","score":34.2,"warning":null},{"id":"bc30496970a9d59834c5cbea","modelId":"gpt-6-sol","benchmarkId":"agents-last-exam","level":"xhigh","sourceLabel":"Codex · gpt-6-sol · XHigh","configuration":"Codex · gpt-6-sol · XHigh. Full/Overall pass rate 32.2%. 49 passes, 151 runs, 151 tasks scored, 152 split tasks. Partial-credit score 56.5%. Estimated cost $318. Runtime 123h 53m. Pass rate is the published passRate field, not passes divided by split tasks.","score":32.2,"warning":null},{"id":"94f176bea93c5ff3cf9fd62d","modelId":"claude-opus-5","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"Claude Code · claude-opus-5 · High","configuration":"Claude Code · claude-opus-5 · High. Full/Overall pass rate 32.2%. 49 passes, 150 runs, 150 tasks scored, 152 split tasks. Partial-credit score 55.9%. Estimated cost $1,108. Runtime 133h 30m. Pass rate is the published passRate field, not passes divided by split tasks.","score":32.2,"warning":null},{"id":"676de6ee897817d1d1995fa9","modelId":"muse-spark-1.3","benchmarkId":"agents-last-exam","level":"xhigh","sourceLabel":"Codex · muse-spark-1-3 · XHigh","configuration":"Codex · muse-spark-1-3 · XHigh. Full/Overall pass rate 32.2%. 49 passes, 148 runs, 148 tasks scored, 152 split tasks. Partial-credit score 55.8%. Estimated cost $428. Runtime 80h 37m. Pass rate is the published passRate field, not passes divided by split tasks.","score":32.2,"warning":null},{"id":"7182b41ada71f4ea331578ae","modelId":"gpt-5.6-sol","benchmarkId":"agents-last-exam","level":"xhigh","sourceLabel":"Codex · GPT-5.6-Sol · XHigh","configuration":"Codex · GPT-5.6-Sol · XHigh. Full/Overall pass rate 30.6%. 93 passes, 301 runs, 152 tasks scored, 152 split tasks. Partial-credit score 53.6%. Estimated cost $772. Runtime 94h 39m. Pass rate is the published passRate field, not passes divided by split tasks.","score":30.6,"warning":null},{"id":"3843a03d8c0c83f7c0f2fd5c","modelId":"gpt-5.6-luna","benchmarkId":"agents-last-exam","level":"xhigh","sourceLabel":"Codex · GPT-5.6-Luna · XHigh","configuration":"Codex · GPT-5.6-Luna · XHigh. Full/Overall pass rate 30.3%. 46 passes, 152 runs, 152 tasks scored, 152 split tasks. Partial-credit score 49.4%. Estimated cost $235. Runtime 66h 7m. Pass rate is the published passRate field, not passes divided by split tasks.","score":30.3,"warning":null},{"id":"d147d87d8825b885876e54e8","modelId":"kimi-k3","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Kimi Code · kimi-k3 · Max","configuration":"Kimi Code · kimi-k3 · Max. Full/Overall pass rate 28.3%. 43 passes, 149 runs, 149 tasks scored, 152 split tasks. Partial-credit score 51.6%. Estimated cost $606. Runtime 186h 5m. Pass rate is the published passRate field, not passes divided by split tasks.","score":28.3,"warning":null},{"id":"da1b7a5f88650cd956a5d434","modelId":"gpt-5.6-terra","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Codex · GPT-5.6-Terra · Max","configuration":"Codex · GPT-5.6-Terra · Max. Full/Overall pass rate 28.0%. 82 passes, 297 runs, 152 tasks scored, 152 split tasks. Partial-credit score 50.7%. Estimated cost $545. Runtime 118h 39m. Pass rate is the published passRate field, not passes divided by split tasks.","score":28,"warning":null},{"id":"e8cb99d37aa15cde0f079ce3","modelId":"qwen-3.8-max","benchmarkId":"agents-last-exam","level":"xhigh","sourceLabel":"Claude Code · qwen-3-8-max · XHigh","configuration":"Claude Code · qwen-3-8-max · XHigh. Full/Overall pass rate 27.0%. 41 passes, 148 runs, 148 tasks scored, 152 split tasks. Partial-credit score 52.5%. Estimated cost $486. Runtime 202h 41m. Pass rate is the published passRate field, not passes divided by split tasks.","score":27,"warning":null},{"id":"40e5b714c33333961838af83","modelId":"grok-4.5","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"Grok Build · grok-4-5 · High","configuration":"Grok Build · grok-4-5 · High. Full/Overall pass rate 27.0%. 41 passes, 151 runs, 151 tasks scored, 152 split tasks. Partial-credit score 51.2%. Estimated cost $337. Runtime 89h 8m. Pass rate is the published passRate field, not passes divided by split tasks.","score":27,"warning":null},{"id":"eb302e48cfc1131b785921eb","modelId":"kimi-k3","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Claude Code · kimi-k3 · Max","configuration":"Claude Code · kimi-k3 · Max. Full/Overall pass rate 27.0%. 41 passes, 149 runs, 149 tasks scored, 152 split tasks. Partial-credit score 50.7%. Estimated cost $464. Runtime 215h 32m. Pass rate is the published passRate field, not passes divided by split tasks.","score":27,"warning":null},{"id":"f9b3e2c57ef651d836e9b514","modelId":"claude-opus-4-8","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Claude Code · claude-opus-4-8 · Max","configuration":"Claude Code · claude-opus-4-8 · Max. Full/Overall pass rate 27.0%. 41 passes, 150 runs, 150 tasks scored, 152 split tasks. Partial-credit score 45.1%. Estimated cost $3,985. Runtime 179h 18m. Pass rate is the published passRate field, not passes divided by split tasks.","score":27,"warning":null},{"id":"99aaf2eec3bdecbcdd6ac1ae","modelId":"gpt-5.5","benchmarkId":"agents-last-exam","level":"xhigh","sourceLabel":"Codex · gpt-5-5 · XHigh","configuration":"Codex · gpt-5-5 · XHigh. Full/Overall pass rate 26.6%. 80 passes, 299 runs, 152 tasks scored, 152 split tasks. Partial-credit score 47.9%. Estimated cost $602. Runtime 97h 8m. Pass rate is the published passRate field, not passes divided by split tasks.","score":26.6,"warning":null},{"id":"4791f23ba358c833ebb34cf0","modelId":"claude-fable-5","benchmarkId":"agents-last-exam","level":"xhigh","sourceLabel":"Claude Code · anthropic-claude-fable-5 · XHigh","configuration":"Claude Code · anthropic-claude-fable-5 · XHigh. Full/Overall pass rate 25.7%. 39 passes, 150 runs, 150 tasks scored, 152 split tasks. Partial-credit score 48.7%. Estimated cost $4,340. Runtime 70h 0m. Pass rate is the published passRate field, not passes divided by split tasks. 59 of 152 benchmark tasks (38.8%) triggered Fable 5's safety fallback to Opus 4.8, rounded to 40%.","score":25.7,"warning":null},{"id":"29ec421af5ec87c489812279","modelId":"gpt-6-luna","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Codex · gpt-6-luna · Max","configuration":"Codex · gpt-6-luna · Max. Full/Overall pass rate 25.0%. 38 passes, 150 runs, 150 tasks scored, 152 split tasks. Partial-credit score 50.9%. Estimated cost $21. Runtime 131h 50m. Pass rate is the published passRate field, not passes divided by split tasks.","score":25,"warning":null},{"id":"6ff9b78c83312245d7e2e18a","modelId":"gpt-5.5","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"ALE-Claw · gpt-5-5 · High","configuration":"ALE-Claw · gpt-5-5 · High. Full/Overall pass rate 23.0%. 35 passes, 145 runs, 145 tasks scored, 152 split tasks. Partial-credit score 45.8%. Estimated cost $310. Runtime 35h 34m. Pass rate is the published passRate field, not passes divided by split tasks.","score":23,"warning":null},{"id":"8b848b06bec07b325290f58b","modelId":"gpt-5.5","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · gpt-5-5 · High","configuration":"OpenClaw · gpt-5-5 · High. Full/Overall pass rate 21.7%. 33 passes, 136 runs, 136 tasks scored, 152 split tasks. Partial-credit score 42.1%. Estimated cost $447. Runtime 82h 38m. Pass rate is the published passRate field, not passes divided by split tasks.","score":21.7,"warning":null},{"id":"b401da83d711f0c01e26e3c2","modelId":"gpt-5.5","benchmarkId":"agents-last-exam","level":"medium","sourceLabel":"Cursor CLI · gpt-5-5 · Medium","configuration":"Cursor CLI · gpt-5-5 · Medium. Full/Overall pass rate 21.4%. 33 passes, 142 runs, 139 tasks scored, 152 split tasks. Partial-credit score 40.7%. Estimated cost $174. Runtime 58h 47m. Pass rate is the published passRate field, not passes divided by split tasks.","score":21.4,"warning":null},{"id":"1c189d546f25ecf35d1cef0f","modelId":"claude-opus-4-7","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"Cursor CLI · claude-opus-4-7 · High","configuration":"Cursor CLI · claude-opus-4-7 · High. Full/Overall pass rate 21.1%. 32 passes, 137 runs, 137 tasks scored, 152 split tasks. Partial-credit score 42.5%. Estimated cost $1,899. Runtime 63h 49m. Pass rate is the published passRate field, not passes divided by split tasks.","score":21.1,"warning":null},{"id":"85767195c6630f480c84d715","modelId":"gpt-5.4","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · gpt-5-4 · High","configuration":"OpenClaw · gpt-5-4 · High. Full/Overall pass rate 20.5%. 40 passes, 179 runs, 142 tasks scored, 152 split tasks. Partial-credit score 37.3%. Estimated cost $274. Runtime 127h 46m. Pass rate is the published passRate field, not passes divided by split tasks.","score":20.5,"warning":null},{"id":"6db38ff867f86b76eb7ce119","modelId":"glm-5.2","benchmarkId":"agents-last-exam","level":"max","sourceLabel":"Claude Code · glm-5-2 · Max","configuration":"Claude Code · glm-5-2 · Max. Full/Overall pass rate 20.4%. 31 passes, 146 runs, 146 tasks scored, 152 split tasks. Partial-credit score 41.1%. Estimated cost $1,086. Runtime 107h 42m. Pass rate is the published passRate field, not passes divided by split tasks.","score":20.4,"warning":null},{"id":"99b9bc14d37a54c4472bf028","modelId":"gpt-5.5","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"Droid · gpt-5-5 · High","configuration":"Droid · gpt-5-5 · High. Full/Overall pass rate 19.7%. 30 passes, 139 runs, 137 tasks scored, 152 split tasks. Partial-credit score 39.2%. Estimated cost $242. Runtime 63h 49m. Pass rate is the published passRate field, not passes divided by split tasks.","score":19.7,"warning":null},{"id":"6ca7f9471855c52d650102c4","modelId":"claude-opus-4-7","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"ALE-Claw · claude-opus-4-7 · High","configuration":"ALE-Claw · claude-opus-4-7 · High. Full/Overall pass rate 18.4%. 28 passes, 143 runs, 143 tasks scored, 152 split tasks. Partial-credit score 40.5%. Estimated cost $1,132. Runtime 77h 39m. Pass rate is the published passRate field, not passes divided by split tasks.","score":18.4,"warning":null},{"id":"4db1bc3faa17daebe263884f","modelId":"gemini-3.1-pro","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"Gemini CLI · gemini-3-1-pro-preview · High","configuration":"Gemini CLI · gemini-3-1-pro-preview · High. Full/Overall pass rate 16.4%. 25 passes, 142 runs, 142 tasks scored, 152 split tasks. Partial-credit score 32.7%. Estimated cost $2,018. Runtime 92h 21m. Pass rate is the published passRate field, not passes divided by split tasks.","score":16.4,"warning":null},{"id":"7e17e17174ee5e6776535383","modelId":"claude-opus-4-7","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · claude-opus-4-7 · High","configuration":"OpenClaw · claude-opus-4-7 · High. Full/Overall pass rate 15.8%. 24 passes, 142 runs, 141 tasks scored, 152 split tasks. Partial-credit score 35.2%. Estimated cost $1,689. Runtime 124h 10m. Pass rate is the published passRate field, not passes divided by split tasks.","score":15.8,"warning":null},{"id":"ff1f22b4d68310bc283968c3","modelId":"gemini-3.1-pro","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · gemini-3-1-pro-preview · High","configuration":"OpenClaw · gemini-3-1-pro-preview · High. Full/Overall pass rate 14.1%. 22 passes, 149 runs, 141 tasks scored, 152 split tasks. Partial-credit score 29.3%. Estimated cost $2,969. Runtime 147h 6m. Pass rate is the published passRate field, not passes divided by split tasks.","score":14.1,"warning":null},{"id":"3c177d8a521629c80a420cba","modelId":"claude-opus-4-7","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"Claude Code · claude-opus-4-7 · High","configuration":"Claude Code · claude-opus-4-7 · High. Full/Overall pass rate 13.8%. 21 passes, 137 runs, 137 tasks scored, 152 split tasks. Partial-credit score 35.8%. Estimated cost $1,793. Runtime 42h 36m. Pass rate is the published passRate field, not passes divided by split tasks.","score":13.8,"warning":null},{"id":"a776154c46f2cc0828eb39a0","modelId":"claude-opus-4-7","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"Droid · claude-opus-4-7 · High","configuration":"Droid · claude-opus-4-7 · High. Full/Overall pass rate 13.5%. 22 passes, 105 runs, 99 tasks scored, 152 split tasks. Partial-credit score 31.7%. Estimated cost $1,356. Runtime 28h 16m. Pass rate is the published passRate field, not passes divided by split tasks.","score":13.5,"warning":null},{"id":"d5fd02cd82ec9b3e9de42bff","modelId":"qwen3.7-max","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · qwen-qwen3-7-max · High","configuration":"OpenClaw · qwen-qwen3-7-max · High. Full/Overall pass rate 11.8%. 18 passes, 142 runs, 142 tasks scored, 152 split tasks. Partial-credit score 31.6%. Estimated cost $664. Runtime 101h 0m. Pass rate is the published passRate field, not passes divided by split tasks.","score":11.8,"warning":null},{"id":"f38f231a2fc30b1d26dc2c51","modelId":"gpt-5.4","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"ALE-Claw · gpt-5-4 · High","configuration":"ALE-Claw · gpt-5-4 · High. Full/Overall pass rate 11.8%. 18 passes, 143 runs, 143 tasks scored, 152 split tasks. Partial-credit score 28.6%. Estimated cost $335. Runtime 55h 28m. Pass rate is the published passRate field, not passes divided by split tasks.","score":11.8,"warning":null},{"id":"99c13e65612f655c0f444506","modelId":"glm-5.1","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · glm-5-1 · High","configuration":"OpenClaw · glm-5-1 · High. Full/Overall pass rate 11.5%. 35 passes, 260 runs, 140 tasks scored, 152 split tasks. Partial-credit score 28.1%. Estimated cost $440. Runtime 168h 39m. Pass rate is the published passRate field, not passes divided by split tasks.","score":11.5,"warning":null},{"id":"6a05810f60840d208e7d7a5c","modelId":"kimi-k2.6","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · kimi-k2-6 · High","configuration":"OpenClaw · kimi-k2-6 · High. Full/Overall pass rate 9.2%. 25 passes, 241 runs, 144 tasks scored, 152 split tasks. Partial-credit score 21.7%. Estimated cost $123. Runtime 183h 18m. Pass rate is the published passRate field, not passes divided by split tasks.","score":9.2,"warning":null},{"id":"ee1282ce96154165b172e969","modelId":"qwen3.6-plus","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · qwen3-6-plus · High","configuration":"OpenClaw · qwen3-6-plus · High. Full/Overall pass rate 8.6%. 25 passes, 272 runs, 146 tasks scored, 152 split tasks. Partial-credit score 24.6%. Estimated cost $254. Runtime 134h 43m. Pass rate is the published passRate field, not passes divided by split tasks.","score":8.6,"warning":null},{"id":"bd81d0ff1a33ba4825491483","modelId":"mimo-v2.5","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · mimo-v2-5 · High","configuration":"OpenClaw · mimo-v2-5 · High. Full/Overall pass rate 8.6%. 25 passes, 261 runs, 146 tasks scored, 152 split tasks. Partial-credit score 23.6%. Estimated cost $43. Runtime 107h 22m. Pass rate is the published passRate field, not passes divided by split tasks.","score":8.6,"warning":null},{"id":"c29cbf6cb5c6183084b940ba","modelId":"grok-4.3","benchmarkId":"agents-last-exam","level":"unspecified","sourceLabel":"Grok CLI · grok-4-3 · -","configuration":"Grok CLI · grok-4-3 · -. Full/Overall pass rate 7.2%. 11 passes, 149 runs, 142 tasks scored, 152 split tasks. Partial-credit score 21.3%. Estimated cost $285. Runtime 43h 7m. Pass rate is the published passRate field, not passes divided by split tasks.","score":7.2,"warning":null},{"id":"a2c07f57afd3fa445d3d2ddd","modelId":"minimax-m2.7","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · minimax-m2-7 · High","configuration":"OpenClaw · minimax-m2-7 · High. Full/Overall pass rate 5.9%. 14 passes, 223 runs, 146 tasks scored, 152 split tasks. Partial-credit score 14.2%. Estimated cost $27. Runtime 144h 37m. Pass rate is the published passRate field, not passes divided by split tasks.","score":5.9,"warning":null},{"id":"34f1a58bf158b01775e872a6","modelId":"grok-4.3","benchmarkId":"agents-last-exam","level":"high","sourceLabel":"OpenClaw · grok-4-3 · High","configuration":"OpenClaw · grok-4-3 · High. Full/Overall pass rate 4.3%. 13 passes, 262 runs, 145 tasks scored, 152 split tasks. Partial-credit score 15.8%. Estimated cost $80. Runtime 79h 4m. Pass rate is the published passRate field, not passes divided by split tasks.","score":4.3,"warning":null},{"id":"d0340ac19be6cfb68ecb513d","modelId":"claude-fable-5-1","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"Claude Fable 5.1 (Max)","configuration":"Claude Fable 5.1 (Max); Arena Agent Mode; IPS net improvement 14.31 points ±1.90. rank 1; observations 1683381; sessions 15130; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":14.31,"warning":null},{"id":"c4b3229a8f23f808571e12f1","modelId":"claude-opus-5-5","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Claude Opus 5.5 (High)","configuration":"Claude Opus 5.5 (High); Arena Agent Mode; IPS net improvement 13.82 points ±2.17. rank 2; observations 754031; sessions 5359; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":13.82,"warning":null},{"id":"0776624889a32af4fb08e2bd","modelId":"claude-sonnet-5-5","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"Claude Sonnet 5.5 (Max)","configuration":"Claude Sonnet 5.5 (Max); Arena Agent Mode; IPS net improvement 12.52 points ±3.09. rank 3; observations 483211; sessions 5219; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":12.52,"warning":null},{"id":"45c8d6cbfc9d9cd5ff4e4ad5","modelId":"gpt-6-astra","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"GPT 6 Astra (Max)","configuration":"GPT 6 Astra (Max); Arena Agent Mode; IPS net improvement 12.27 points ±2.23. rank 4; observations 888641; sessions 11717; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":12.27,"warning":null},{"id":"947bac6664a523899cdba3f9","modelId":"gpt-6.1-sol","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"GPT 6.1 Sol (Max)","configuration":"GPT 6.1 Sol (Max); Arena Agent Mode; IPS net improvement 11.23 points ±2.76. rank 5; observations 379372; sessions 6278; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":11.23,"warning":null},{"id":"a9a25cfcdd0ec17e5a09ebe7","modelId":"gpt-6-sol","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"GPT 6 Sol (Max)","configuration":"GPT 6 Sol (Max); Arena Agent Mode; IPS net improvement 9.71 points ±2.37. rank 6; observations 994367; sessions 6106; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":9.71,"warning":null},{"id":"3282f64a5da3e2dcd05537fb","modelId":"claude-opus-5","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Claude Opus 5 (High)","configuration":"Claude Opus 5 (High); Arena Agent Mode; IPS net improvement 8.67 points ±1.41. rank 7; observations 3531760; sessions 27560; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":8.67,"warning":null},{"id":"ab25c1100a249db08a0ccc50","modelId":"claude-fable-5","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Claude Fable 5 (High)","configuration":"Claude Fable 5 (High); Arena Agent Mode; IPS net improvement 8.21 points ±1.22. rank 8; observations 2603067; sessions 41832; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":8.21,"warning":null},{"id":"8c35a383b0dd6e6ab648d6f6","modelId":"claude-opus-5","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"Claude Opus 5 (Max)","configuration":"Claude Opus 5 (Max); Arena Agent Mode; IPS net improvement 7.92 points ±1.47. rank 9; observations 3006257; sessions 22781; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":7.92,"warning":null},{"id":"d636f326988768d1353e6613","modelId":"gemini-4-argon","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Gemini 4 Argon (High)","configuration":"Gemini 4 Argon (High); Arena Agent Mode; IPS net improvement 7.57 points ±1.93. rank 10; observations 587353; sessions 10030; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":7.57,"warning":null},{"id":"03ae20fc24f1f2bebe6a8f10","modelId":"claude-opus-4-8","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Claude Opus 4.8 (High)","configuration":"Claude Opus 4.8 (High); Arena Agent Mode; IPS net improvement 6.64 points ±1.34. rank 11; observations 2431185; sessions 41132; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":6.64,"warning":null},{"id":"7ee13b6baf57fa44f4ecf9f5","modelId":"gpt-5.6-sol","benchmarkId":"lmarena-agent","level":"xhigh","sourceLabel":"GPT 5.6 Sol (xHigh)","configuration":"GPT 5.6 Sol (xHigh); Arena Agent Mode; IPS net improvement 6.45 points ±1.28. rank 12; observations 4348715; sessions 35896; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":6.45,"warning":null},{"id":"f1407dd17f71174872700d81","modelId":"claude-sonnet-5","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Claude Sonnet 5 (High)","configuration":"Claude Sonnet 5 (High); Arena Agent Mode; IPS net improvement 4.40 points ±1.58. rank 13; observations 3612898; sessions 32467; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":4.4,"warning":null},{"id":"790e1a899cce18c5e7115e5e","modelId":"kimi-k3","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"Kimi K3 (Max)","configuration":"Kimi K3 (Max); Arena Agent Mode; IPS net improvement 4.18 points ±0.54. rank 14; observations 12305667; sessions 132505; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":4.18,"warning":null},{"id":"8f9242e8ec4831420dfa6cf8","modelId":"gpt-5.5","benchmarkId":"lmarena-agent","level":"xhigh","sourceLabel":"GPT 5.5 (xHigh)","configuration":"GPT 5.5 (xHigh); Arena Agent Mode; IPS net improvement 4.15 points ±1.03. rank 15; observations 3060353; sessions 55006; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":4.15,"warning":null},{"id":"eb38bdb02586fc5c7b56eb43","modelId":"grok-4.7","benchmarkId":"lmarena-agent","level":"xhigh","sourceLabel":"Grok 4.7 (xHigh)","configuration":"Grok 4.7 (xHigh); Arena Agent Mode; IPS net improvement 4.04 points ±1.60. rank 16; observations 1656641; sessions 11591; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":4.04,"warning":null},{"id":"a9c19f28b6b35f1cfecff3f7","modelId":"deepseek-v4.1-flash","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"Deepseek V4.1 Flash (Max)","configuration":"Deepseek V4.1 Flash (Max); Arena Agent Mode; IPS net improvement 4.02 points ±0.47. rank 17; observations 16932092; sessions 112938; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":4.02,"warning":null},{"id":"543bed84dd98726b9cda8ed0","modelId":"muse-spark-1.3","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"Muse Spark 1.3 (Max)","configuration":"Muse Spark 1.3 (Max); Arena Agent Mode; IPS net improvement 4.00 points ±0.60. rank 18; observations 6139426; sessions 64671; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":4,"warning":null},{"id":"279ad2ba50cd782c323458b5","modelId":"hy4-preview","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Hy4 preview","configuration":"Hy4 preview; Arena Agent Mode; IPS net improvement 3.95 points ±0.79. rank 19; observations 5108255; sessions 59388; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":3.95,"warning":null},{"id":"9bab6207f50373b71744538c","modelId":"glm-5.2","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"GLM 5.2 (Max)","configuration":"GLM 5.2 (Max); Arena Agent Mode; IPS net improvement 3.48 points ±0.72. rank 20; observations 6684620; sessions 86172; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":3.48,"warning":null},{"id":"4f27cdab99c08b380ea60eff","modelId":"gemini-3.8-flash","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Gemini 3.8 Flash (High)","configuration":"Gemini 3.8 Flash (High); Arena Agent Mode; IPS net improvement 2.97 points ±0.84. rank 21; observations 4279699; sessions 34396; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":2.97,"warning":null},{"id":"95296dde8063cef39635d0b9","modelId":"qwen-3.8-max","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Qwen3.8 Max","configuration":"Qwen3.8 Max; Arena Agent Mode; IPS net improvement 2.51 points ±0.62. rank 22; observations 4577924; sessions 52824; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":2.51,"warning":null},{"id":"e481a590d3ab672be6892c52","modelId":"glm-5.3","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"GLM 5.3 (Max)","configuration":"GLM 5.3 (Max); Arena Agent Mode; IPS net improvement 2.36 points ±0.68. rank 23; observations 9155315; sessions 86384; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":2.36,"warning":null},{"id":"4e865ae63f79c95b4a197ebc","modelId":"gpt-6-luna","benchmarkId":"lmarena-agent","level":"max","sourceLabel":"GPT 6 Luna (Max)","configuration":"GPT 6 Luna (Max); Arena Agent Mode; IPS net improvement 1.35 points ±1.07. rank 24; observations 3443780; sessions 29241; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":1.35,"warning":null},{"id":"dcc7c5758d6865db29bb27be","modelId":"grok-4.6","benchmarkId":"lmarena-agent","level":"xhigh","sourceLabel":"Grok 4.6 (xHigh)","configuration":"Grok 4.6 (xHigh); Arena Agent Mode; IPS net improvement 1.28 points ±1.00. rank 25; observations 3131725; sessions 27380; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":1.28,"warning":null},{"id":"54aed1470c504f06fe839893","modelId":"grok-4.5","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Grok 4.5","configuration":"Grok 4.5; Arena Agent Mode; IPS net improvement 1.21 points ±1.01. rank 26; observations 2418629; sessions 41920; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":1.21,"warning":null},{"id":"a4195a4f532e0d321ef0c94c","modelId":"deepseek-v4-pro","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"DeepSeek V4 Pro (High) (0813)","configuration":"DeepSeek V4 Pro (High) (0813); Arena Agent Mode; IPS net improvement 1.19 points ±1.66. rank 27; observations 797284; sessions 7090; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":1.19,"warning":null},{"id":"e0c64dea01582f2b42d1b4b8","modelId":"gpt-5.4","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"GPT 5.4 (High)","configuration":"GPT 5.4 (High); Arena Agent Mode; IPS net improvement 1.05 points ±0.92. rank 28; observations 4332904; sessions 81597; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":1.05,"warning":null},{"id":"312698f57dded23365430ed8","modelId":"gpt-5.5","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"GPT 5.5","configuration":"GPT 5.5; Arena Agent Mode; IPS net improvement 1.04 points ±0.98. rank 29; observations 2659460; sessions 82416; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":1.04,"warning":null},{"id":"510b1906e1532d94112acce2","modelId":"gpt-5.6-terra","benchmarkId":"lmarena-agent","level":"xhigh","sourceLabel":"GPT 5.6 Terra (xHigh)","configuration":"GPT 5.6 Terra (xHigh); Arena Agent Mode; IPS net improvement 0.40 points ±1.16. rank 31; observations 1686712; sessions 23457; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":0.4,"warning":null},{"id":"64d37dff5c0b2237fd2dc9fb","modelId":"glm-5.3-flash","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"GLM 5.3 Flash","configuration":"GLM 5.3 Flash; Arena Agent Mode; IPS net improvement -0.44 points ±0.47. rank 32; observations 9533773; sessions 80836; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-0.44,"warning":null},{"id":"166578670251d544d464c5f8","modelId":"qwen3.8-flash-next","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Qwen3.8 Flash Next","configuration":"Qwen3.8 Flash Next; Arena Agent Mode; IPS net improvement -0.73 points ±0.63. rank 34; observations 11249221; sessions 83198; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-0.73,"warning":null},{"id":"fb4286feccac496c1b6396cd","modelId":"gpt-5.6-luna","benchmarkId":"lmarena-agent","level":"xhigh","sourceLabel":"GPT 5.6 Luna (xHigh)","configuration":"GPT 5.6 Luna (xHigh); Arena Agent Mode; IPS net improvement -1.17 points ±0.67. rank 35; observations 4028029; sessions 42599; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-1.17,"warning":null},{"id":"3a21c505486d52bd5f114c15","modelId":"gemini-3.7-flash","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Gemini 3.7 Flash (High)","configuration":"Gemini 3.7 Flash (High); Arena Agent Mode; IPS net improvement -1.49 points ±0.74. rank 36; observations 5048920; sessions 54594; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-1.49,"warning":null},{"id":"f0693c6f295f758d738c52e6","modelId":"qwen3.8-27b","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Qwen 3.8 27B","configuration":"Qwen 3.8 27B; Arena Agent Mode; IPS net improvement -1.66 points ±0.60. rank 37; observations 6972838; sessions 57343; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-1.66,"warning":null},{"id":"c1360f99a929ad065c2ff5ec","modelId":"muse-spark-1.2","benchmarkId":"lmarena-agent","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xHigh)","configuration":"Muse Spark 1.2 (xHigh); Arena Agent Mode; IPS net improvement -3.27 points ±0.67. rank 38; observations 2884900; sessions 48150; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-3.27,"warning":null},{"id":"a197a7064aa5d129c4f73d78","modelId":"muse-spark-1.1","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Muse Spark 1.1","configuration":"Muse Spark 1.1; Arena Agent Mode; IPS net improvement -4.88 points ±0.45. rank 39; observations 7612456; sessions 142711; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-4.88,"warning":null},{"id":"aa3b48be077bc3d95033a274","modelId":"qwen3.7-max","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Qwen3.7 Max","configuration":"Qwen3.7 Max; Arena Agent Mode; IPS net improvement -5.13 points ±0.93. rank 40; observations 2207521; sessions 40438; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-5.13,"warning":null},{"id":"dad0090bcc37a9b2dc9499a8","modelId":"hy3","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Hy3","configuration":"Hy3; Arena Agent Mode; IPS net improvement -5.38 points ±1.21. rank 41; observations 1171568; sessions 29677; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-5.38,"warning":null},{"id":"929e18aebcc14c0d353763d0","modelId":"minimax-m3","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Minimax M3","configuration":"Minimax M3; Arena Agent Mode; IPS net improvement -6.87 points ±0.88. rank 42; observations 3338791; sessions 41267; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-6.87,"warning":null},{"id":"73ea539ceb5e9b232c4b9323","modelId":"qwen3.7-plus","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Qwen3.7 Plus","configuration":"Qwen3.7 Plus; Arena Agent Mode; IPS net improvement -7.15 points ±1.46. rank 43; observations 1144063; sessions 21722; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-7.15,"warning":null},{"id":"3415c8a3bffb156dd40d6a7d","modelId":"mimo-v2.5-pro","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Mimo V2.5 Pro","configuration":"Mimo V2.5 Pro; Arena Agent Mode; IPS net improvement -7.52 points ±0.89. rank 44; observations 2309389; sessions 45239; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-7.52,"warning":null},{"id":"9ce8a035d41740b238aab251","modelId":"gemini-3.6-flash","benchmarkId":"lmarena-agent","level":"high","sourceLabel":"Gemini 3.6 Flash (High)","configuration":"Gemini 3.6 Flash (High); Arena Agent Mode; IPS net improvement -7.68 points ±1.06. rank 45; observations 1523777; sessions 22913; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-7.68,"warning":null},{"id":"b4fc9a81e348fec4b20f2754","modelId":"gemini-3.1-pro","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Gemini 3.1 Pro Preview","configuration":"Gemini 3.1 Pro Preview; Arena Agent Mode; IPS net improvement -7.71 points ±0.98. rank 46; observations 3446977; sessions 86483; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-7.71,"warning":null},{"id":"f15486c1efb571ab847d2fba","modelId":"mistral-medium-3.5","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Mistral Medium 3.5","configuration":"Mistral Medium 3.5; Arena Agent Mode; IPS net improvement -12.38 points ±1.52. rank 49; observations 437621; sessions 11156; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-12.38,"warning":null},{"id":"5b7e5263e136f2def58c73e5","modelId":"solar-pro4-260806","benchmarkId":"lmarena-agent","level":"unspecified","sourceLabel":"Solar Pro 4","configuration":"Solar Pro 4; Arena Agent Mode; IPS net improvement -15.90 points ±1.85. rank 50; observations 427705; sessions 9506; published 2026-10-02. The board does not name a per-model scaffold. Display only. Excluded from weighted Overall.","score":-15.9,"warning":null},{"id":"81e9970add62c0c7b2531bfb","modelId":"gpt-6-astra","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"GPT-6 Astra","configuration":"GPT-6 Astra; Proximal FrontierSWE v2; Proximus; mean@5 65.5%. Best@5 73.0%; worst@5 55.1%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 169 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified. The task table records 169 trials of the nominal 170. Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=gpt-6-astra.","score":65.5,"warning":"The task table records 169 trials of the nominal 170."},{"id":"041dbc44ffd19c2b84b093db","modelId":"claude-opus-5-5","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Claude Opus 5.5","configuration":"Claude Opus 5.5; Proximal FrontierSWE v2; Proximus; mean@5 62.3%. Best@5 71.5%; worst@5 52.0%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=claude-opus-5.5.","score":62.3,"warning":null},{"id":"ac99b7569548ee908cfe0fb3","modelId":"claude-sonnet-5-5","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Claude Sonnet 5.5","configuration":"Claude Sonnet 5.5; Proximal FrontierSWE v2; Proximus; mean@5 61.9%. Best@5 71.7%; worst@5 50.9%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=claude-sonnet-5.5.","score":61.9,"warning":null},{"id":"44d1ed48497d24a4dd085aaa","modelId":"claude-fable-5-1","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Claude Fable 5.1","configuration":"Claude Fable 5.1; Proximal FrontierSWE v2; Proximus; mean@5 56.3%. Best@5 66.7%; worst@5 44.5%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified. The v2 methodology discloses Opus 5 fallback on tasks blocked by content filters. The task-summary mean 56.52830% differs from the published headline mean@5 56.29079%; the headline score is retained. Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=claude-fable-5.1.","score":56.3,"warning":"The v2 methodology discloses Opus 5 fallback on tasks blocked by content filters. The task-summary mean 56.52830% differs from the published headline mean@5 56.29079%; the headline score is retained."},{"id":"ff3051d2f04ecd2ad550e7a0","modelId":"gemini-4-argon","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Gemini 4 Argon","configuration":"Gemini 4 Argon; Proximal FrontierSWE v2; Proximus; mean@5 55.0%. Best@5 64.3%; worst@5 44.5%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified. Cache hit rate is much lower due to a non-production setting. Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=gemini-4-argon.","score":55,"warning":"Cache hit rate is much lower due to a non-production setting."},{"id":"e8f2358815c124dd0449c1ca","modelId":"claude-opus-5","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Claude Opus 5","configuration":"Claude Opus 5; Proximal FrontierSWE v2; Proximus; mean@5 52.0%. Best@5 60.7%; worst@5 39.1%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=claude-opus-5.","score":52,"warning":null},{"id":"f0129750ffae88478acefff0","modelId":"claude-fable-5","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Claude Fable 5","configuration":"Claude Fable 5; Proximal FrontierSWE v2; Proximus; mean@5 47.0%. Best@5 61.8%; worst@5 33.9%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified. The task-summary mean 47.20200% differs from the published headline mean@5 46.96449%; the headline score is retained. Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=claude-fable-5.","score":47,"warning":"The task-summary mean 47.20200% differs from the published headline mean@5 46.96449%; the headline score is retained."},{"id":"60d94bb8cee7478e8cff443b","modelId":"claude-haiku-5-5","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Claude Haiku 5.5","configuration":"Claude Haiku 5.5; Proximal FrontierSWE v2; Proximus; mean@5 43.8%. Best@5 52.8%; worst@5 34.5%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=claude-haiku-5.5.","score":43.8,"warning":null},{"id":"b7ebc08725286a2ecf054772","modelId":"gpt-5.6-sol","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"GPT-5.6 Sol","configuration":"GPT-5.6 Sol; Proximal FrontierSWE v2; Proximus; mean@5 32.2%. Best@5 44.2%; worst@5 22.2%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=gpt-5.6-sol.","score":32.2,"warning":null},{"id":"89528e3aa6e3997d416741b6","modelId":"glm-5.3","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"GLM-5.3","configuration":"GLM-5.3; Proximal FrontierSWE v2; Proximus; mean@5 30.2%. Best@5 40.6%; worst@5 17.7%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=glm-5.3.","score":30.2,"warning":null},{"id":"ded42586dbefafa89c12009f","modelId":"grok-4.7","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Grok 4.7","configuration":"Grok 4.7; Proximal FrontierSWE v2; Proximus; mean@5 29.5%. Best@5 41.1%; worst@5 15.9%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=grok-4.7.","score":29.5,"warning":null},{"id":"da50ef39dd8b9399b3ee8d7f","modelId":"kimi-k3","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Kimi K3","configuration":"Kimi K3; Proximal FrontierSWE v2; Proximus; mean@5 25.9%. Best@5 37.6%; worst@5 13.9%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 165 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified. The task table records 165 trials of the nominal 170. Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=kimi-k3.","score":25.9,"warning":"The task table records 165 trials of the nominal 170."},{"id":"00b5696cd7c263fe4470e081","modelId":"muse-spark-1.3","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Muse Spark 1.3","configuration":"Muse Spark 1.3; Proximal FrontierSWE v2; Proximus; mean@5 25.8%. Best@5 40.9%; worst@5 12.6%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=muse-spark-1.3.","score":25.8,"warning":null},{"id":"bb0115557a23b53b4fe10b76","modelId":"grok-4.6","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Grok 4.6","configuration":"Grok 4.6; Proximal FrontierSWE v2; Proximus; mean@5 25.3%. Best@5 37.3%; worst@5 12.9%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=grok-4.6.","score":25.3,"warning":null},{"id":"676afe6774d1ab6eaf6a7ccd","modelId":"gemini-3.7-flash","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Gemini 3.7 Flash","configuration":"Gemini 3.7 Flash; Proximal FrontierSWE v2; Proximus; mean@5 20.3%. Best@5 31.2%; worst@5 10.3%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=gemini-3.7-flash.","score":20.3,"warning":null},{"id":"0b258d6cd8b710583beb5bbb","modelId":"gemini-3.8-flash","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Gemini 3.8 Flash","configuration":"Gemini 3.8 Flash; Proximal FrontierSWE v2; Proximus; mean@5 19.6%. Best@5 31.6%; worst@5 9.5%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=gemini-3.8-flash.","score":19.6,"warning":null},{"id":"ccb8c9a318581c00f767d922","modelId":"glm-5.3-flash","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"GLM-5.3 Flash","configuration":"GLM-5.3 Flash; Proximal FrontierSWE v2; Proximus; mean@5 18.1%. Best@5 28.4%; worst@5 8.6%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=glm-5.3-flash.","score":18.1,"warning":null},{"id":"69de10246886f142905d7542","modelId":"qwen-3.8-max-0902","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Qwen3.8-Max-0902","configuration":"Qwen3.8-Max-0902; Proximal FrontierSWE v2; Proximus; mean@5 17.8%. Best@5 26.8%; worst@5 8.1%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=qwen3.8-max-0902.","score":17.8,"warning":null},{"id":"3228a6cefadba28448a44428","modelId":"qwen-3.8-max","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Qwen3.8-Max","configuration":"Qwen3.8-Max; Proximal FrontierSWE v2; Proximus; mean@5 15.8%. Best@5 24.3%; worst@5 8.7%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=qwen3.8-max.","score":15.8,"warning":null},{"id":"455c653df3d24ea2e0c4851f","modelId":"deepseek-v4-flash-vision-exp","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"DeepSeek V4 Flash Vision Exp","configuration":"DeepSeek V4 Flash Vision Exp; Proximal FrontierSWE v2; Proximus; mean@5 14.8%. Best@5 26.5%; worst@5 7.0%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 167 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified. The task table records 167 trials of the nominal 170. Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=deepseek-v4-flash-vision-exp.","score":14.8,"warning":"The task table records 167 trials of the nominal 170."},{"id":"060fac95feb6422df27091c3","modelId":"muse-spark-1.2","benchmarkId":"frontierswe-v2","level":"unspecified","sourceLabel":"Muse Spark 1.2","configuration":"Muse Spark 1.2; Proximal FrontierSWE v2; Proximus; mean@5 12.0%. Best@5 18.4%; worst@5 6.8%, contextual aggregates, not a confidence interval. 34 tasks, nominal five trials per task, 20-hour task budget. Task table records 170 trials. The methodology states maximum reasoning effort without a named per-model setting; effort unspecified.  Display only. Excluded from weighted Overall. Trace https://www.frontierswe.com/traces?model=muse-spark-1.2.","score":12,"warning":null},{"id":"a15e4bf3befd7f7ae09a6858","modelId":"gpt-6-astra","benchmarkId":"hss","level":"max","sourceLabel":"GPT-6-Astra (max)","configuration":"GPT-6-Astra (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 53.6% ±4.1 percentage points; rank 1; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":53.6,"warning":null},{"id":"a4ada587a84348711490c89b","modelId":"gpt-6.1-sol","benchmarkId":"hss","level":"max","sourceLabel":"GPT-6.1-Sol (max)","configuration":"GPT-6.1-Sol (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 46.6% ±3.8 percentage points; rank 2; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":46.6,"warning":null},{"id":"4dbe3c30fa34d2b33ef22677","modelId":"claude-opus-5-5","benchmarkId":"hss","level":"xhigh","sourceLabel":"Claude-Opus-5.5 (xhigh)","configuration":"Claude-Opus-5.5 (xhigh); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 44.6% ±3.8 percentage points; rank 3; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":44.6,"warning":null},{"id":"c7a097bc2e305075826d2a59","modelId":"gemini-3.8-flash","benchmarkId":"hss","level":"high","sourceLabel":"Gemini-3.8-Flash (high)","configuration":"Gemini-3.8-Flash (high); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 41.6% ±3.7 percentage points; rank 4; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":41.6,"warning":null},{"id":"0311f7bd1351961cee7a9240","modelId":"claude-fable-5-1","benchmarkId":"hss","level":"unspecified","sourceLabel":"Claude-Fable-5.1","configuration":"Claude-Fable-5.1; Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 40.8% ±3.6 percentage points; rank 5; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":40.8,"warning":"The board does not publish an effort setting for this row."},{"id":"830ca9b63ce7eb52b8dce017","modelId":"gemini-3.7-flash","benchmarkId":"hss","level":"high","sourceLabel":"Gemini-3.7-Flash (high)","configuration":"Gemini-3.7-Flash (high); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 39.8% ±3.7 percentage points; rank 6; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":39.8,"warning":null},{"id":"b399eabdcbf649daa5218245","modelId":"muse-spark-1.3","benchmarkId":"hss","level":"max","sourceLabel":"Muse Spark 1.3 (max)","configuration":"Muse Spark 1.3 (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 37.4% ±3.5 percentage points; rank 7; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":37.4,"warning":null},{"id":"7d021e24a2dc6e213f69d17a","modelId":"claude-fable-5","benchmarkId":"hss","level":"max","sourceLabel":"Claude-Fable-5 (max)","configuration":"Claude-Fable-5 (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 34.5% ±3.7 percentage points; rank 8; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":34.5,"warning":null},{"id":"2131b16c32557cee50a60f92","modelId":"qwen-3.8-max","benchmarkId":"hss","level":"max","sourceLabel":"Qwen3.8-Max (max)","configuration":"Qwen3.8-Max (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 33% ±3.6 percentage points; rank 9; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":33,"warning":null},{"id":"795aea8aba98f63ce3479cc8","modelId":"gemini-3.5-flash","benchmarkId":"hss","level":"high","sourceLabel":"Gemini-3.5-Flash (high)","configuration":"Gemini-3.5-Flash (high); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 32.6% ±3.4 percentage points; rank 10; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":32.6,"warning":null},{"id":"bd64e697822e6c9e153a5ef8","modelId":"gemini-3.6-flash","benchmarkId":"hss","level":"high","sourceLabel":"Gemini-3.6-Flash (high)","configuration":"Gemini-3.6-Flash (high); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 31.9% ±3.4 percentage points; rank 11; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":31.9,"warning":null},{"id":"4d0badf71b8d3d8c61801909","modelId":"gpt-6-sol","benchmarkId":"hss","level":"max","sourceLabel":"GPT-6-Sol (max)","configuration":"GPT-6-Sol (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 31.2% ±3.4 percentage points; rank 12; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":31.2,"warning":null},{"id":"2a2fe835120323f80463fd8b","modelId":"qwen3.8-flash","benchmarkId":"hss","level":"max","sourceLabel":"Qwen3.8-Flash (max)","configuration":"Qwen3.8-Flash (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 30.9% ±3.3 percentage points; rank 13; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":30.9,"warning":null},{"id":"884750c61c8ee37b44c56c69","modelId":"claude-opus-5","benchmarkId":"hss","level":"max","sourceLabel":"Claude-Opus-5 (max)","configuration":"Claude-Opus-5 (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 30.5% ±3.5 percentage points; rank 14; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":30.5,"warning":null},{"id":"46659376ecba1fe13fffb611","modelId":"gpt-5.6-sol","benchmarkId":"hss","level":"max","sourceLabel":"GPT-5.6-Sol (max)","configuration":"GPT-5.6-Sol (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 30% ±3.3 percentage points; rank 15; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":30,"warning":null},{"id":"e7a8767f88bd7aea8058ac83","modelId":"glm-5.3-flash","benchmarkId":"hss","level":"max","sourceLabel":"GLM-5.3-Flash (max)","configuration":"GLM-5.3-Flash (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 26.8% ±3 percentage points; rank 16; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":26.8,"warning":null},{"id":"ded2e2416b8e56ec60ecee9b","modelId":"kimi-k3","benchmarkId":"hss","level":"max","sourceLabel":"Kimi-K3 (max)","configuration":"Kimi-K3 (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 25.5% ±3.1 percentage points; rank 17; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":25.5,"warning":null},{"id":"49cd86f41454a9710c05d448","modelId":"mimo-v2.6-flash","benchmarkId":"hss","level":"high","sourceLabel":"MiMo-v2.6-Flash (high)","configuration":"MiMo-v2.6-Flash (high); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 25.4% ±2.9 percentage points; rank 18; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":25.4,"warning":null},{"id":"21471cb42e2a96c3549bf8ee","modelId":"mimo-v2.6-pro","benchmarkId":"hss","level":"high","sourceLabel":"MiMo-v2.6-Pro (high)","configuration":"MiMo-v2.6-Pro (high); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 25.2% ±2.9 percentage points; rank 19; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":25.2,"warning":null},{"id":"7ab4b62bef03e4fc16ac041d","modelId":"claude-sonnet-5","benchmarkId":"hss","level":"max","sourceLabel":"Claude-Sonnet-5 (max)","configuration":"Claude-Sonnet-5 (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 24.8% ±3.2 percentage points; rank 20; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":24.8,"warning":null},{"id":"e4fa929703a4d06882d14187","modelId":"muse-spark-1.2","benchmarkId":"hss","level":"xhigh","sourceLabel":"Muse Spark 1.2 (xhigh)","configuration":"Muse Spark 1.2 (xhigh); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 24.5% ±3.1 percentage points; rank 21; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":24.5,"warning":null},{"id":"fbb82323532284442cdd12f7","modelId":"claude-opus-4-8","benchmarkId":"hss","level":"max","sourceLabel":"Claude-Opus-4.8 (max)","configuration":"Claude-Opus-4.8 (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 23.5% ±3.2 percentage points; rank 22; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":23.5,"warning":null},{"id":"695ded2356d3f25584dd210d","modelId":"gpt-5.6-terra","benchmarkId":"hss","level":"max","sourceLabel":"GPT-5.6-Terra (max)","configuration":"GPT-5.6-Terra (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 22.7% ±3.2 percentage points; rank 23; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":22.7,"warning":null},{"id":"f884228e6423471428032523","modelId":"gpt-5.6-luna","benchmarkId":"hss","level":"max","sourceLabel":"GPT-5.6-Luna (max)","configuration":"GPT-5.6-Luna (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 21.6% ±3.2 percentage points; rank 24; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":21.6,"warning":null},{"id":"0e78b9edbdbf167341a3174a","modelId":"gpt-6-luna","benchmarkId":"hss","level":"max","sourceLabel":"GPT-6-Luna (max)","configuration":"GPT-6-Luna (max); Full 522-task pass@1, task mean of three attempts; every rubric criterion must pass. 95% task-cluster bootstrap CI. Claude-Opus-5 judge. Scale AI + Elorian. Unversioned board; source hashes identify captures. Display only. Excluded from weighted Overall. Score 21% ±3 percentage points; rank 25; board record time 2026-10-06T23:03:39.000Z, not a proven evaluation date. Temperature and top-p use vendor defaults. Public-web media contamination cannot be excluded.","score":21,"warning":null}],"skipped":[{"source":"aa-gdpval-aa","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdpval-aa","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-3d","label":"/evaluations/aa-3d","reason":"Unreviewed AA metric"},{"source":"aa-aa-analyst-agent","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-analyst-agent","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-analyst-agent","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-analyst-agent","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-aa-briefcase","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Claude 4 Opus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-aime-2025","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"o1-preview","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-aime-2025","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-aime-2025","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-aime-2025","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-apex-agents-aa","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-apex-agents-aa","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-apex-agents-aa","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-apex-agents-aa","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"G9v3-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-artificial-analysis-long-context-reasoning","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-automationbench-aa","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-critpt","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-critpt","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-critpt","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-critpt","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"G9v3-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-critpt","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-critpt","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Gemini 3 Deep Think","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-critpt","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-enterprise-ops-gym-aa","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-enterprise-ops-gym-aa","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-enterprise-ops-gym-aa","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-enterprise-ops-gym-aa","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-frontiercode","label":"/evaluations/frontiercode","reason":"Unreviewed AA metric"},{"source":"aa-gdp-pdf","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-gdp-pdf","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-global-mmlu-lite","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-global-mmlu-lite","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-global-mmlu-lite","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-gpqa-diamond","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-gpqa-diamond","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-gpqa-diamond","label":"Claude 4 Opus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Pro Preview (Mar' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Pro Preview (May' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-gpqa-diamond","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Flash Preview (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"o1-preview","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"G9v3-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"o1-mini","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.0 Flash Thinking Experimental (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Solar Pro 2 (Preview) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.0 Pro Experimental (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3 Opus","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Sonar Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.5 Flash Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.0 Flash (experimental)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen2.5 Max","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 1.5 Pro (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Solar Pro 2 (Preview) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3.5 Sonnet (Oct '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Sonar","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Sonar Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"QwQ 32B-Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-gpqa-diamond","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.0 Flash-Lite (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 2.0 Flash-Lite (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3.5 Sonnet (June '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.1 Tulu3 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok 2 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 1.5 Flash (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-gpqa-diamond","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Grok Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen2.5 Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen2.5 Coder Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek-V2.5 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Saba","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 1.5 Pro (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen2.5 Turbo","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.2 Instruct 90B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen2 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 1.5 Flash-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepHermes 3 - Mistral 24B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Jamba 1.5 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Hermes 3 - Llama-3.1 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Jamba 1.6 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 1.5 Flash (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3 Sonnet","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Small (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemma 3n E4B Instruct Preview (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen2.5 Coder Instruct 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Large (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mixtral 8x22B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 2 Chat 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Qwen1.5 Chat 110B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 2.1","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Small (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM 40B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Gemini 1.0 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepSeek Coder V2 Lite Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 2 Chat 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 2 Chat 13B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"OpenChat 3.5 (1210)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DBRX Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Jamba 1.5 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Jamba 1.6 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mixtral 8x7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Claude Instant","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-gpqa-diamond","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-harvey-lab-aa-v1.1","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-harvey-lab-aa-v1.1","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-humanitys-last-exam","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-humanitys-last-exam","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-humanitys-last-exam","label":"Claude 4 Opus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Pro Preview (Mar' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Pro Preview (May' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-humanitys-last-exam","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Flash Preview (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"G9v3-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"o1-mini","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.0 Flash Thinking Experimental (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Solar Pro 2 (Preview) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.0 Pro Experimental (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3 Opus","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.5 Flash Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.0 Flash (experimental)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen2.5 Max","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 1.5 Pro (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Solar Pro 2 (Preview) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3.5 Sonnet (Oct '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Sonar","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Sonar Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"QwQ 32B-Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-humanitys-last-exam","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.0 Flash-Lite (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 2.0 Flash-Lite (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3.5 Sonnet (June '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.1 Tulu3 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok 2 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 1.5 Flash (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-humanitys-last-exam","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Grok Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen2.5 Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen2.5 Coder Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Saba","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 1.5 Pro (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen2.5 Turbo","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.2 Instruct 90B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen2 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 1.5 Flash-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepHermes 3 - Mistral 24B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Jamba 1.5 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Hermes 3 - Llama-3.1 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Jamba 1.6 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 1.5 Flash (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3 Sonnet","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Small (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemma 3n E4B Instruct Preview (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Qwen2.5 Coder Instruct 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Large (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mixtral 8x22B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 2 Chat 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 2.1","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Small (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM 40B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Gemini 1.0 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepSeek Coder V2 Lite Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 2 Chat 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 2 Chat 13B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"OpenChat 3.5 (1210)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DBRX Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Jamba 1.5 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Jamba 1.6 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mixtral 8x7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Claude Instant","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-humanitys-last-exam","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-ifbench","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-ifbench","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-ifbench","label":"Claude 4 Opus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-ifbench","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-ifbench","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-ifbench","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-ifbench","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-itbench-aa","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-itbench-aa","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 4 Opus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Pro Preview (Mar' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Pro Preview (May' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-livecodebench","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Flash Preview (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"o1-mini","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.0 Flash Thinking Experimental (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Solar Pro 2 (Preview) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.0 Pro Experimental (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 3 Opus","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.5 Flash Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.0 Flash (experimental)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen2.5 Max","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 1.5 Pro (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Solar Pro 2 (Preview) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 3.5 Sonnet (Oct '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Sonar","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Sonar Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"QwQ 32B-Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-livecodebench","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.0 Flash-Lite (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 2.0 Flash-Lite (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.1 Tulu3 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok 2 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 1.5 Flash (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-livecodebench","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Grok Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen2.5 Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen2.5 Coder Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 1.5 Pro (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen2.5 Turbo","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.2 Instruct 90B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen2 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 1.5 Flash-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepHermes 3 - Mistral 24B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Jamba 1.5 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Hermes 3 - Llama-3.1 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Jamba 1.6 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 1.5 Flash (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 3 Sonnet","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral Small (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemma 3n E4B Instruct Preview (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Qwen2.5 Coder Instruct 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral Large (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mixtral 8x22B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 2 Chat 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 2.1","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral Small (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"LFM 40B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Gemini 1.0 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepSeek Coder V2 Lite Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 2 Chat 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 2 Chat 13B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"OpenChat 3.5 (1210)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DBRX Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Jamba 1.5 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Jamba 1.6 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mixtral 8x7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Claude Instant","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-livecodebench","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 4 Opus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.5 Pro Preview (Mar' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.5 Pro Preview (May' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Sonar Reasoning Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.5 Flash Preview (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"o1-preview","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"o1-mini","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.0 Flash Thinking Experimental (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Solar Pro 2 (Preview) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.0 Pro Experimental (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3 Opus","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Sonar Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.5 Flash Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.0 Flash (experimental)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Qwen2.5 Max","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 1.5 Pro (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Solar Pro 2 (Preview) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3.5 Sonnet (Oct '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Sonar","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Sonar Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"QwQ 32B-Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-math-500","label":"Gemini 2.0 Flash-Lite (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.0 Flash-Lite (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3.5 Sonnet (June '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.1 Tulu3 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Grok 2 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 1.5 Flash (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-math-500","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Grok Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Qwen2.5 Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Qwen2.5 Coder Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DeepSeek-V2.5 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 2.0 Flash Thinking Experimental (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Saba","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 1.5 Pro (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"R1 1776","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Qwen2.5 Turbo","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Reka Flash (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.2 Instruct 90B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Solar Mini","reason":"2024 evaluator model must not map to current April 2025 API revision"},{"source":"aa-math-500","label":"Qwen2 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 1.5 Flash-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DeepHermes 3 - Mistral 24B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Jamba 1.5 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Hermes 3 - Llama-3.1 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DeepSeek-Coder-V2","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Jamba 1.6 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 1.5 Flash (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3 Sonnet","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Small (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemma 3n E4B Instruct Preview (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Qwen2.5 Coder Instruct 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Large (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mixtral 8x22B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 2 Chat 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 2.1","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Small (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"LFM 40B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Gemini 1.0 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 2 Chat 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 2 Chat 13B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"OpenChat 3.5 (1210)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DBRX Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Jamba 1.5 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Jamba 1.6 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mixtral 8x7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Claude Instant","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-math-500","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-mlcr-aa","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-mlcr-aa","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-mlcr-aa","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 4 Opus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Pro Preview (Mar' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Pro Preview (May' 25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-mmlu-pro","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Flash Preview (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"o1-preview","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"o1-mini","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"QwQ 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.0 Flash Thinking Experimental (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Solar Pro 2 (Preview) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.0 Pro Experimental (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 3 Opus","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.5 Flash Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.0 Flash (experimental)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen2.5 Max","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 1.5 Pro (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Solar Pro 2 (Preview) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 3.5 Sonnet (Oct '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Sonar","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Sonar Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"QwQ 32B-Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-mmlu-pro","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 2.0 Flash-Lite (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 3.5 Sonnet (June '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.1 Tulu3 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok 2 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 1.5 Flash (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-mmlu-pro","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Grok Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen2.5 Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen2.5 Coder Instruct 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek-V2.5 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Saba","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 1.5 Pro (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen2.5 Turbo","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.2 Instruct 90B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen2 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 1.5 Flash-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepHermes 3 - Mistral 24B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Jamba 1.5 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Hermes 3 - Llama-3.1 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Jamba 1.6 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 1.5 Flash (May '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 3 Sonnet","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Small (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemma 3n E4B Instruct Preview (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Qwen2.5 Coder Instruct 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Large (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mixtral 8x22B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 2 Chat 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 2.1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Small (Feb '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"LFM 40B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Gemini 1.0 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepSeek Coder V2 Lite Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 2 Chat 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 2 Chat 13B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"OpenChat 3.5 (1210)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DBRX Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Jamba 1.5 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Jamba 1.6 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mixtral 8x7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"DeepHermes 3 - Llama-3.1 8B Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Claude Instant","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-mmlu-pro","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-mmmu-pro","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 2.5 Flash Preview (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 1.5 Pro (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 1.5 Flash (Sep '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Llama 3.2 Instruct 90B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Gemini 1.5 Flash-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-mmmu-pro","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-omniscience","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-omniscience","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-omniscience","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-omniscience","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"G9v3-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-omniscience","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-omniscience","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-omniscience-accuracy","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-omniscience-accuracy","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-omniscience-accuracy","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-omniscience-accuracy","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"G9v3-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-omniscience-accuracy","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-omniscience-accuracy","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-omniscience-accuracy","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-scicode","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-tau2-bench","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-tau2-bench","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-tau2-bench","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-tau2-bench","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-tau2-bench","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.1 Nemotron Nano 4B v1.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-tau2-bench","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Mistral Small 3","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Solar Mini","reason":"2024 evaluator model must not map to current April 2025 API revision"},{"source":"aa-tau2-bench","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.2 Instruct 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Mistral 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-tau2-bench","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-tau3-banking","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminal-bench-science","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminal-bench-science","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-terminal-bench-science","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminal-bench-science","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Agnes 2.5 Pro Beta","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Motif 3","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Motif 3 (Beta)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Nex-N2-Pro (Based on Qwen3.5-397B-A17B)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"JT-4.1 Flash 236B A21B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Grok Build 0.1 0616","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"G9v3-39A5B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"G9v3-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"LFM2.5-2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v2-1","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Ling 3.1 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Muse Spark Open (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"DeepSeek V4 Flash Vision (Max)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"JT-4.1 Flash 236B A21B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"K2 Horizon 375B A23B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Beam 1.0","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Quasar 438B (Max, Based on GLM-5.2)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Nemotron 3.5 Super VL","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Apodex 1.1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"GPT-5.5 Instant (June 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Inkling Small","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"K2 Horizon MoVA 36B A4B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Inkling (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Ling-3.0-flash-VL","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Ling-3.0-flash-Fin","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"A.X-K2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Nemotron 3.5 Bolt 74B-A9B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"K2 Horizon 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Ling 3.0 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"LongCat 2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Muse Glimmer (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"K2 Horizon 3.7B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Granite 4.2 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Mercury 2.5","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Granite 4.2 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Ling 3.0 Tiny","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Granite 4.2 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"Celeris-1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-v4-0","label":"K2 Horizon 0.9B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Muse Spark","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"MiMo-V2-Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"GPT-5.2 Codex (Xhigh)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Qwen3.6 Max Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Gemini 3 Pro Preview (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4.20 0309 v2 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4.20 0309 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"MiMo-V2-Omni-0327","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"GPT-5 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"MiMo-V2-Omni","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"GPT-5.1 Codex (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Kimi K2.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Claude 4.1 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"GPT-5.5 Instant (May 2026)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"MiMo-V2-Flash (Feb 2026)","reason":"February API revision is distinct from the original MiMo V2 Flash weights"},{"source":"aa-terminalbench-hard","label":"Gemini 3 Pro Preview (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Kimi K2 Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"KAT Coder Pro V2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Qwen3 Max Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Claude 4 Opus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"GPT-5.1 Codex mini (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4.1 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Kimi K2.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Claude 4 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"JT-35B-Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"KAT-Coder-Pro V1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4 Fast (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Claude 3.7 Sonnet (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Ling-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Step 3.5 Flash 2603","reason":"Source explicitly labels 2603; exact match to current provider weights needs checkpoint review"},{"source":"aa-terminalbench-hard","label":"Ring-2.6-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Step 3.5 Flash","reason":"February checkpoint must not map to the current rolling Step 3.5 Flash model"},{"source":"aa-terminalbench-hard","label":"Claude 4 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3.2 Exp (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Qwen3 Max Thinking (Preview)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Gemini 2.5 Flash Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Kimi K2 0905","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Claude 3.7 Sonnet (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3.1 Terminus (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 3 mini Reasoning (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4.20 0309 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4.20 0309 v2 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok Code Fast 1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3.1 Terminus (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3.2 Exp (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Apriel-v1.5-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3.1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Nova 2.0 Omni (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Qwen3 VL 235B A22B (Reasoning)","reason":"Evaluator reasoning label conflicts with its linked Instruct checkpoint. Hold out until the evaluator identifies the actual Thinking or Instruct weights."},{"source":"aa-terminalbench-hard","label":"Apriel-v1.6-15B-Thinker","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Kimi K2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Gemini 2.5 Flash Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"JT-MINI","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 3","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"HyperNova 60B 2605 (High, Based on gpt-oss-120b)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Nemotron Cascade 2 30B A3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"K2 Think V2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LongCat Flash Lite","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek R1 (Jan '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"HyperCLOVA X SEED Think (32B)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4.1 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Nova 2.0 Omni (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Grok 4 Fast (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Mi:dm K 2.5 Pro","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Ring-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Trinity Large Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Mercury 2","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"INTELLECT-3 (Based on GLM-4.5-Air)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"North Mini Code","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"K2-V2 (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3 0324","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Ling 2.6 Flash","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Tri-21B-think Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Gemini 2.5 Flash-Lite Preview (Sep '25) (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Motif-2-12.7B-Reasoning","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Ling-1T","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Magistral Medium 1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"K2-V2 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama Nemotron Super 49B v1.5 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Devstral Medium","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Tri-21B-Think","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Gemini 2.0 Flash (Feb '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Claude 3.5 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"MiniCPM5-1B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Sarvam 105B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Devstral Small (May '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"MiniCPM5-1B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"DeepSeek V3 (Dec '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Nanbeige4.1-3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Magistral Small 1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Nova 2.0 Omni (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Hermes 4 - Llama-3.1 70B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Falcon-H1R-7B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Ling-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Qwen3 Omni 30B A3B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Qwen2.5 Instruct 72B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Devstral Small (Jul '25)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Mistral Large 2 (Nov '24)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Hermes 4 - Llama-3.1 405B (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Solar Pro 2 (Reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-terminalbench-hard","label":"Granite 4.1 30B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Hermes 4 - Llama-3.1 405B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama Nemotron Super 49B v1.5 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"K2-V2 (Low)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.1 Instruct 405B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2.5-8B-A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Ring-flash-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Olmo 3.1 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Mistral Small 3.1","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Solar Pro 2 (Non-reasoning)","reason":"July 2025 evaluator release must not map to current December 2025 API revision"},{"source":"aa-terminalbench-hard","label":"Command A","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.1 Nemotron Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.1 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Hermes 4 - Llama-3.1 70B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.1 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.1 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Sarvam 30B (High)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Olmo 3.1 32B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Olmo 3 32B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Jamba 1.7 Large","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.0 H Small","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"OLMo 2 32B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2 24B A2B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.1 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Phi-3 Mini Instruct 3.8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"MiniCPM-V 4.6 1.3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Jamba Reasoning 3B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Reka Flash 3","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Olmo 3 7B Think","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Claude 3 Haiku","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"OLMo 2 7B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Molmo 7B-D","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Ling-mini-2.0","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3 Instruct 70B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.2 Instruct 11B (Vision)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Sarvam M (Reasoning, Based on Mistral Small 3.1)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Olmo 3 7B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2.5-1.2B-Thinking","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Jamba 1.7 Mini","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2 2.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2.5-1.2B-Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.0 H 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Apertus 70B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.0 Micro","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.0 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Molmo2-8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2 8B A1B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 3.3 8B (Non-reasoning)","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2 1.2B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"LFM2.5-VL-1.6B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.0 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3.2 Instruct 1B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Llama 3 Instruct 8B","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Apertus 8B Instruct","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Granite 4.0 H 350M","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Mi:dm K 2.5 Pro Preview","reason":"Unresolved exact catalog identity"},{"source":"aa-terminalbench-hard","label":"Cogito v2.1 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude 4.7 (High)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude 4.7 (Low)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude 4.7 (Max)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude 4.7 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude Opus 4 (Thinking 16K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude Opus 4 (Thinking 1K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude Opus 4 (Thinking 8K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Opus 4.5 (Thinking, 16K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Opus 4.5 (Thinking, 64K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Opus 4.5 (Thinking, 8K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Opus 4.5 (Thinking, None)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude Sonnet 4 (Thinking 16K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude Sonnet 4 (Thinking 1K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Claude Sonnet 4 (Thinking 8K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Dots3-Note Preview (Max)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Gemini 2.5 Flash (Preview) (Thinking 16K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Gemini 2.5 Flash (Preview) (Thinking 1K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Gemini 2.5 Flash (Preview) (Thinking 24K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Gemini 2.5 Flash (Preview) (Thinking 8K)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Grok 3 Mini (Low)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Grok 4 (Thinking)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Grok 4 (Fast Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Grok 4.20 (Reasoning)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Magistral Medium (Thinking)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"GPT-5.6 Luna 2026-07-30 (High)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"GPT-5.6 Luna 2026-07-30 (Low)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"GPT-5.6 Luna 2026-07-30 (Max)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"GPT-5.6 Luna 2026-07-30 (Medium)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"GPT-5.6 Luna 2026-07-30 (None)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"GPT-5.6 Luna 2026-07-30 (XHigh)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Inkling Small (High)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Inkling Small (Low)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Inkling Small (Medium)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Inkling Small (Minimal)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Inkling Small (None)","reason":"Unresolved exact catalog identity"},{"source":"arc-v2semiprivate","label":"Inkling Small (XHigh)","reason":"Unresolved exact catalog identity"},{"source":"terminal-terminus2","label":"Gemini 3 Pro","reason":"Unresolved exact catalog identity"},{"source":"terminal-geminicli","label":"Gemini 3 Pro","reason":"Unresolved exact catalog identity"},{"source":"livecode-generationpass1","label":"Claude-Opus-4 (Thinking)","reason":"Unresolved exact catalog identity"},{"source":"livecode-generationpass1","label":"Claude-Sonnet-4 (Thinking)","reason":"Unresolved exact catalog identity"},{"source":"livecode-generationpass1","label":"Grok-3-Mini (High)","reason":"Unresolved exact catalog identity"},{"source":"livecode-generationpass1","label":"O3-Mini-2025-01-31 (High)","reason":"Unresolved exact catalog identity"},{"source":"livecode-generationpass1","label":"O3-Mini-2025-01-31 (Low)","reason":"Unresolved exact catalog identity"},{"source":"vulcanbench-v3-13-deepseek-v4-pro","label":"DeepSeek V4 Pro","reason":"No exact catalog identity. Unversioned DeepSeek report predates current Pro revision; do not assume equivalence."},{"source":"vulcanbench-v3-13-deepseek-v4-pro","label":"DeepSeek V4-Flash","reason":"No exact catalog identity. Unversioned DeepSeek report predates current Pro revision; do not assume equivalence."},{"source":"Cognition","label":"SWE-2","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-2","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-2","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-2","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-2","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-2","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-1.7","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-1.7","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"Kimi K2.7","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"Kimi K2.7","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"Composer 2.5","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"Composer 2.5","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"Inkling","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"Inkling","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-1.6","reason":"No unambiguous catalog identity"},{"source":"Cognition","label":"SWE-1.6","reason":"No unambiguous catalog identity"},{"source":"ARC Prize","label":"GPT-5.6 Luna extra-high","reason":"Unreviewed effort label"},{"source":"ARC Prize","label":"GPT-5.6 Sol extra-high","reason":"Unreviewed effort label"},{"source":"ARC Prize","label":"GPT-5.6 Terra extra-high","reason":"Unreviewed effort label"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"GPT 6 Astra","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Gemini 3.8 Flash","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gpt-5.4-pro-2026-03-05","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Muse Spark","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gemini-3-pro-preview","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gpt-5.4-2026-03-05 (xhigh thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"claude-opus-4-7","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gpt-5-pro-2025-10-06","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Claude Haiku 5.5","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gpt-5.2-2025-12-11","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gpt-5-2025-08-07","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"kimi-k2.5","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gemini-2.5-pro-preview-06-05","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"o3 (high) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gpt-5-mini-2025-08-07","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"o3 (medium) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Gemini 2.5 Pro Experimental (March 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"o4-mini (high) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Gemini 2.5 Pro Preview (May 06 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"o4-mini (medium) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"claude-opus-4-5-20251101","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Gemini 2.5 Flash (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"claude-opus-4-1-20250805-thinking","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Gemini 2.5 Flash Preview (May 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Claude Opus 4 (Thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gemini-3.1-flash-lite-preview","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"glm-4p5","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"glm-4p5-air","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"o1 Pro","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Claude 3.7 Sonnet (Thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"o1 (December 2024)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"claude-opus-4-1-20250805","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Claude Sonnet 4 (Thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"claude-sonnet-4-5-20250929","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"gpt-5.1-instant","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Claude Opus 4","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Gemini 2.0 Flash Thinking (January 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Llama 4 Maverick","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Claude Sonnet 4","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"GPT 4.5 Preview","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"GPT-4.1","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Gemini-1.5-Pro-002","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Mistral Medium 3","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Nova Pro","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Claude 3.5 Sonnet (October 2024)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"Nova Lite","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam","label":"GPT-4o (November 2024)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"GPT 6 Astra","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-5.4-pro-2026-03-05","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Gemini 3.8 Flash","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Muse Spark","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gemini-3-pro-preview","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-5.4-2026-03-05 (xhigh thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-5-pro-2025-10-06","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Claude Haiku 5.5","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-5.2-2025-12-11","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-5-2025-08-07","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gemini-2.5-pro-preview-06-05","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"o3 (high) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"o3 (medium) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-5-mini-2025-08-07","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"o4-mini (high) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Gemini 2.5 Pro Experimental (March 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Gemini 2.5 Pro Preview (May 06 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-oss-120b","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Qwen3-235B-A22B-Thinking-2507","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"o4-mini (medium) (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"DeepSeek-R1-0528","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"claude-opus-4-5-20251101","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"DeepSeek-V3.1","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Gemini 2.5 Flash (April 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Qwen3-235B-A22B","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"claude-opus-4-1-20250805-thinking","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Claude Opus 4 (Thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Gemini 2.5 Flash Preview (May 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-oss-20b","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"glm-4p5","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"glm-4p5-air","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"DeepSeek R1","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gemini-3.1-flash-lite-preview","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Claude 3.7 Sonnet (Thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"o1 (December 2024)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"o1 Pro","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"claude-sonnet-4-5-20250929","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Claude Sonnet 4 (Thinking)","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"claude-opus-4-1-20250805","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Gemini 2.0 Flash Thinking (January 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"gpt-5.1-instant","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Claude Opus 4","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"GPT 4.5 Preview","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Claude Sonnet 4","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Llama 4 Maverick","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"GPT-4.1","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"kimi-k2-instruct","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"DeepSeek V3 (March 2025)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Gemini-1.5-Pro-002","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Nova Micro","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Mistral Medium 3","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Claude 3.5 Sonnet (October 2024)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Nova Pro","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"Nova Lite","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/humanitys_last_exam_text_only","label":"GPT-4o (November 2024)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"GLM 5.2 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Gemini 3.8 Flash (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Opus 4.7 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Gpt 5.4 xHigh (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Opus 4.6 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"GPT 5.3 (Codex) xHigh","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Sonnet 4.6 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Opus 4.6 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"DeepSeek V4 Pro (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Muse Spark","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Glm 5 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Gemini 3.1 Pro (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Kimi K2.5 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Minimax M2.5 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-qna","label":"Gemini 3 Flash (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Opus-4.7 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Opus 4.8 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Gemini 3.8 Flash (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"GLM 5.2 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"GPT 5.3 (Codex) xHigh","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Opus-4.6 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Gemini-3.1-Pro (Gemini CLI)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Sonnet-4.6 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Glm-5 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Kimi-K2.5 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Minimax-M2.5 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-refactoring","label":"Gemini-3-Flash (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Gemini 3.8 Flash (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"GLM 5.2 (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"GPT 5.3 (Codex) xHigh","reason":"No unambiguous exact catalog checkpoint"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Opus 4.7 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Opus-4.6 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Opus-4.6 (Mini-SWE)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Sonnet-4.6 (Claude Code)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Muse Spark","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Gemini-3-Flash (Mini-SWE)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Gemini-3.1-Pro (Mini-SWE)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Glm-5 (Mini-SWE)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Deepseek V4 Pro (Mini-SWE-Agent)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Kimi-K2.5 (Mini-SWE)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"https://labs.scale.com/leaderboard/sweatlas-tw","label":"Minimax-M2.5 (Mini-SWE)","reason":"Exact effort or harness not identified; no default inferred"},{"source":"LiveBench","label":"deepseek-v4-flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"deepseek-v4-pro","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"glm-5.2","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"gpt-5.2-codex","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"grok-4.3","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"grok-4.5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"grok-build-0.1","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"kimi-k2.6-thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"kimi-k2.7-code","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"minimax-m3","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"qwen3.6-27b","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"qwen3.6-plus","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"qwen3.7-max","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"kimi-k3","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"inkling-xhigh","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"deepseek-v4-flash-0731","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"qwen3.8-max","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"smaug-agentic","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"smaug-mini","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"deepseek-v4-pro-0813","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"grok-4.6","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"qwen3.8-27b","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"ox-alpha-max","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"deepseek-v4-flash-vision-exp","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"glm-5.3","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"glm-5.3-flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"qwen3.8-flash-next","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"nemotron-3-ultra-550b-a55b","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"smaug-flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"deepseek-v4.1-flash-max","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"claude-opus-5-5-max-effort","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"LiveBench","label":"claude-opus-5-5-xhigh-effort","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-4.1-2025-04-14","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"DeepSeek-V3-0324","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"DeepSeek-V3","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-235B-A22B no-thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-235B-A22B thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-32B no-thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-32B thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Llama-4-Maverick-17B-128E-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen2.5-72B-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Llama-3.3-70B-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Llama-4-Scout-17B-16E-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gemma-3-27b-it","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen2.5-Coder-32B-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-4.1-mini-2025-04-14","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-4.1-nano-2025-04-14","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gemini-2.0-flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gemini-2.5-flash-preview-05-20 no-thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Devstral-Small-2505","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"DeepSeek-V3-0324","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Claude Sonnet 3.5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Claude Sonnet 4","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gemini-2.5-flash-preview-05-20 no-thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-4.1-2025-04-14","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"o3-2025-04-16","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gemini-2.5-flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gemini-2.0-flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gemini-2.5-pro","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-Coder-480B-A35B-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-235B-A22B-Thinking-2507","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-235B-A22B-Instruct-2507","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"o4-mini-2025-04-16","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"DeepSeek-R1-0528","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"horizon-alpha","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-Coder-30B-A3B-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"horizon-beta","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-4.5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-4.5 Air","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Grok Code Fast 1","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-oss-120b","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-oss-20b","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-4.1-mini-2025-04-14","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Kimi K2","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-235B-A22B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-32B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Kimi K2 Instruct 0905","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-Next-80B-A3B-Instruct","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"DeepSeek-V3.1","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Grok 4","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Claude Sonnet 4.5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Claude Opus 4.1","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-5-codex","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-30B-A3B-Thinking-2507","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-30B-A3B-Instruct-2507","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-4.6","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"MiniMax M2","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Claude Opus 4.5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Gemini 3 Pro Preview","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-5.1-codex-max","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Devstral-2-123B-Instruct-2512","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Devstral-Small-2-24B-Instruct-2512","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"DeepSeek-V3.2","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Claude Code","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"MiniMax M2.1","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-5.1-codex","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Kimi K2 Thinking","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-4.7","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Gemini 3 Flash Preview","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-4.7 Flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Kimi K2.5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-5.2-codex","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3-Coder-Next","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Codex","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"MiniMax M2.5","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Junie","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Gemini 3.1 Pro Preview","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Claude Sonnet 4.6","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"gpt-5.3-codex","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3.5-35B-A3B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3.5-397B-A17B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Step-3.5-Flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-5.1","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"MiniMax M2.7","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3.5-27B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Gemma 4 31B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Cursor","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"GLM-5.1","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Kimi K2.6","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Gemini 3.5 Flash","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"MiniMax M3","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"MiMo V2.5 Pro","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3.6-27B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"Qwen3.6-35B-A3B","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"SWE-rebench","label":"DeepSeek-V4 Flash [high]","reason":"Model checkpoint or effort has no explicit reviewed mapping"},{"source":"Epoch AI · Chess Puzzles","label":"mistral-large-4-0_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Furniture Assembly","label":"muse-wonder-1.0_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"Muse-Spark-Open-816B-A42B_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"Muse-Spark-Open-816B-A42B_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Muse-Spark-Open-816B-A42B_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"Muse-Spark-Open-816B-A42B_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"Muse-Spark-Open-816B-A42B_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-6-astra_promax","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-nano-2026-03-17_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"muse-wonder-1.0_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.2-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-pro-2025-10-06_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.2-pro-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-pro-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.6-sol_promax","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Furniture Assembly","label":"mistral-large-4-0_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Mystery Game Puzzles","label":"mistral-large-4-0_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"gemini-4-argon_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gemini-4-argon_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Chess Puzzles","label":"gemini-4-argon_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gemini-4-argon_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-4-argon_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-4-argon_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gemini-4-argon_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gemini-4-argon_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gemini-4-argon_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-4o-mini-2024-07-18","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o1-2024-12-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o1-2024-12-17_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-4.1-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-4o-2024-08-06","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o1-2024-12-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o3-mini-2025-01-31_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-4.1-nano-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o3-mini-2025-01-31_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o3-mini-2025-01-31_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o4-mini-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-4.1-mini-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o4-mini-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o4-mini-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-mini-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o3-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-mini-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o3-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"o3-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-nano-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-nano-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-nano-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-mini-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.1-2025-11-13_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.2-2025-12-11_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.1-2025-11-13_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.1-2025-11-13_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.2-2025-12-11_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.2-2025-12-11_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.2-2025-12-11_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-2026-03-05_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-2026-03-05_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-2026-03-05_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-2026-03-05_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-nano-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-nano-2026-03-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-nano-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-mini-2026-03-17_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-mini-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-5.4-mini-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"claude-3-opus-20240229","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"claude-haiku-4-5-20251001_59K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"claude-opus-4-5-20251101_59K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"deepseek-v4-pro_max","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · Chess Puzzles","label":"deepseek-v4-pro_high","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · Chess Puzzles","label":"deepseek-v4-pro_none","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · Chess Puzzles","label":"muse-spark-1.1_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"muse-spark-1.3_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"deepseek-v4.1-flash_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"Inkling_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"claude-haiku-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"deepseek-v4.1-flash_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"muse-spark-1.3_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"grok-4.7_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Chess Puzzles","label":"claude-haiku-5-5_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"muse-wonder-1.0_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-large-4-0_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"mistral-large-4-0_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"mistral-large-4-0_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"mistral-large-4-0_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"muse-wonder-1.0_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"muse-wonder-1.0_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"muse-wonder-1.0_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"deepseek-v4.1-flash_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"claude-haiku-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-haiku-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"mistral-large-4-0_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"claude-haiku-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"deepseek-v4.1-flash_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"deepseek-v4.1-flash_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"deepseek-v4.1-flash_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"deepseek-v4.1-flash_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Furniture Assembly","label":"deepseek-v4.1-flash_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Erdos","label":"claude-sonnet-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-haiku-5-5_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-haiku-5-5_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"claude-haiku-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"claude-haiku-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"claude-haiku-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Furniture Assembly","label":"claude-haiku-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"gpt-6.1-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"gpt-6.1-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-6.1-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gpt-6.1-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Mystery Game Puzzles","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"claude-sonnet-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"claude-sonnet-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-sonnet-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"gpt-6-luna_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"gpt-6-luna_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-6-luna_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"claude-fable-5-1_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"claude-fable-5-1_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"claude-opus-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"grok-4.5_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"grok-4.6_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"grok-4.7_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Erdos","label":"claude-opus-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"gpt-6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"claude-opus-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"gpt-6-astra_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"grok-4.7_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"grok-4.7_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"grok-4.7_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"grok-4.7_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"grok-4.7_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"grok-4.7_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"claude-opus-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gpt-6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"gpt-6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Furniture Assembly","label":"claude-opus-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MirrorCode","label":"claude-opus-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"gpt-6-astra_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Mystery Game Puzzles","label":"muse-spark-1.3_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"muse-spark-1.3_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"muse-spark-1.3_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"muse-spark-1.3_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"muse-spark-1.3_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"muse-spark-1.3_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"muse-spark-1.3_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Furniture Assembly","label":"kimi-k3_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"kimi-k2.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"qwen3.8-max-0902_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gemini-3.7-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gemini-3.6-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gemini-3.8-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-fable-5-1_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gemini-3.1-pro-preview_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-fable-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-opus-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-opus-4-8_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-opus-4-7_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-opus-4-6_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"claude-opus-4-5-20251101_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-5.6-luna_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-6-astra_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-5.6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-5.6-terra_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-5.5_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-5.2-2025-12-11_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Furniture Assembly","label":"gpt-5.4-2026-03-05_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MirrorCode","label":"claude-fable-5-1_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"claude-fable-5-1_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"claude-opus-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gpt-5.6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Erdos","label":"claude-fable-5-1_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-4o-mini-2024-07-18","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-4o-2024-08-06","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-4.1-nano-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-4.1-mini-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-4.1-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"o3-mini-2025-01-31_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"o1-2024-12-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"claude-3-opus-20240229","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"granite-4.0-micro","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"granite-4.0-micro","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"granite-4.0-1b","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen2.5-7b-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"qwen2.5-7b-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-small-3.2-2506","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-small-3.1-2503","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"mistral-small-3.1-2503","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-small-3-2501","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"mistral-small-3-2501","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Mistral-7B-Instruct-v0.3","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Meta-Llama-3-8B-Instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"qwen3-max-2025-09-23","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-2026-03-05_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MirrorCode","label":"gpt-6-astra_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gpt-6-astra_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-6-astra_none","reason":"Source reports none, but provider only documents Low through Max; hold out unsupported setting."},{"source":"Epoch AI · GPQA diamond","label":"granite-4.0-1b","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"deepseek-llm-67b-chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Erdos","label":"gpt-5.5_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Erdos","label":"gpt-6-astra_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Erdos","label":"claude-fable-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"mistral-small-3.2-2506","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Erdos","label":"gpt-5.6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"granite-4.0-350m","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"granite-4.0-350m","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-nano-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-mini-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-nano-2026-03-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-mini-2026-03-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"QwQ-32B","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"QwQ-32B","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.2-2025-12-11_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-2026-03-05_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o1-2024-12-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-4-0613","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o1-2024-12-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o1-2024-12-17_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o3-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o3-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-4.1-mini-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-3.5-turbo-0125","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-4o-mini-2024-07-18","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"Inkling_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"inkling-small_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"kimi-k2.5","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"qwen3.6-max-preview","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"qwen3-max-2025-09-23","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"deepseek-v4-pro_max","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · SimpleQA Verified","label":"o4-mini-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.2-2025-12-11_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.2-2025-12-11_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"o3-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.4-nano-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"o4-mini-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.4-mini-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.2-2025-12-11_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.4-pro-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.1-2025-11-13_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.4-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5.2-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"o3-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"o4-mini-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"o3-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-2026-03-05_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"qwen3.6-max-preview_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"o3-mini-2025-01-31_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.1-2025-11-13_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.2-2025-12-11_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-mini-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-nano-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.1-2025-11-13_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-mini-2026-03-17_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-mini-2026-03-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-mini-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-nano-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-nano-2026-03-17_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"qwen3-max-2025-09-23","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.2-2025-12-11_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"deepseek-v4-pro_none","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-nano-2026-03-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-nano-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.1-2025-11-13_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-4.1-mini-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-mini-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-3.5-turbo-0125","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-mini-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-nano-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-nano-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-nano-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-4o-mini-2024-07-18","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o3-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-4.1-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-4-turbo-2024-04-09","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-4o-2024-08-06","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o4-mini-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o4-mini-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o3-mini-2025-01-31_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-mini-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-mini-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-nano-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o3-mini-2025-01-31_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-nano-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"grok-4.6_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Mystery Game Puzzles","label":"inkling-small_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"inkling-small_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"inkling-small_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"inkling-small_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"inkling-small_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MirrorCode","label":"claude-opus-4-7_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open Lite","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open Lite","label":"gpt-5.5_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open Lite","label":"claude-opus-4-8_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open Lite","label":"gpt-5.6-sol_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open Lite","label":"claude-fable-5_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open","label":"gpt-5.5_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OEIS Open","label":"claude-opus-4-8_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5-mini-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SimpleQA Verified","label":"gpt-5-nano-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MirrorCode","label":"gemini-3.1-pro-preview_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MirrorCode","label":"gpt-5.5_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MirrorCode","label":"gpt-5.4-2026-03-05_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MirrorCode","label":"gpt-5.6-sol_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MirrorCode","label":"claude-fable-5_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"claude-opus-4-8_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-mini-2026-03-17_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-mini-2026-03-17_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-mini-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"qwen3.6-max-preview","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-mini-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen3.6-max-preview","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o3-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.1-2025-11-13_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o3-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.1-2025-11-13_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o4-mini-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o4-mini-2025-04-16_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-nano-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-nano-2026-03-17_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-3.5-turbo-0125","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-mini-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-nano-2026-03-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-mini-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-nano-2026-03-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"deepseek-v4-pro_none","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"deepseek-v4-pro_high","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · GPQA diamond","label":"deepseek-v4-pro_none","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · GPQA diamond","label":"deepseek-v4-pro_high","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"magistral-small-2506","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"magistral-small-2506","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"Inkling_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"Inkling_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.2-2025-12-11_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"Inkling_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Inkling_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.5-instant","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.5-instant","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.5-instant","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.5-instant","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"gpt-5.5_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · Mystery Game Puzzles","label":"claude-opus-4-1-20250805_24K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · Mystery Game Puzzles","label":"gpt-5.4-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o1-2024-12-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o3-mini-2025-01-31_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o3-mini-2025-01-31_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"claude-fable-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"deepseek-chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"deepseek-chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o1-2024-12-17_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o1-2024-12-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o3-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o3-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-2026-03-05_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-2026-03-05_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-2026-03-05_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-2026-03-05_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-2026-03-05_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-2026-03-05_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-2026-03-05_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-2026-03-05_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-nano-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-nano-2025-08-07_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-nano-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-nano-2025-08-07_minimal","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.2-2025-12-11_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.2-2025-12-11_none","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o4-mini-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o4-mini-2025-04-16_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.6-sol_promax","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · EBR-bench","label":"claude-opus-4-7_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"claude-opus-4-6_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gpt-5-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"glm-5.2_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gpt-5.2-2025-12-11_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"glm-5.2_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"qwen3.7-max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gpt-5.4-2026-03-05_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"claude-opus-4-5-20251101_128K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"kimi-k2.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"claude-sonnet-4-5-20250929","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"claude-opus-4-1-20250805","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · EBR-bench","label":"gemini-3.1-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"qwen3.7-max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"deepseek-v4-pro_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"deepseek-v4-pro_max","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"deepseek-v4-pro_max","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"deepseek-v4-pro_max","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · GPQA diamond","label":"deepseek-v4-pro_max","reason":"Checkpoint is unspecified; April and August V4 Pro must not be combined."},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.4-pro-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-pro-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.2-pro-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.2-pro-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5-pro-2025-10-06_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-pro-2025-10-06_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-mini-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5-mini-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.4-nano-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-nano-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-nano-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5-nano-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.4-mini-2026-03-17_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-mini-2026-03-17_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gdm-ai-co-mathematician","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"claude-opus-4-1-20250805_32K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"claude-opus-4-1-20250805_32K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.4-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.4-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5.2-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o4-mini-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"o3-mini-2025-01-31_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"o3-mini-2025-01-31_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"o4-mini-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5.2-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-v2-Private","label":"gpt-5-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-fable-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-fable-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tiers-1-3-v2-Private","label":"gpt-5-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-8_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-8_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"qwen3.6-max-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"qwen3.6-max-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"qwen3.6-max-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"qwen3.6-max-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen3.6-max-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-3.5-flash_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"glm-5.1","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"qwen3.5-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"qwen3.6-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"qwen3.5-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen3.5-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"qwen3.6-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"qwen3.6-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"qwen3.6-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"qwen3.6-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"qwen3.6-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen3.6-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen3.6-plus","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"qwen3.5-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"qwen3.5-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen3.5-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"glm-5.1","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"glm-5.1","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"glm-5.1","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gdm-ai-co-mathematician","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"kimi-k2.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"kimi-k2.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"kimi-k2.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"glm-5.1","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"kimi-k2.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.5-pro-pre-release_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.5-pro-pre-release_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.5-pre-release_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5.5-pre-release_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.5-pre-release_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.5-pro-pre-release_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.5-pro-pre-release_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.5-pro-pre-release_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.5-pro-pre-release_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.5-pre-release_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.5-pre-release_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-opus-4-7_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-opus-4-7_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-7_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-7_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-7_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-mini-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-mini-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5.4-mini-2026-03-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.4-mini-2026-03-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.4-mini-2026-03-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.4-nano-2026-03-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.4-nano-2026-03-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5.4-nano-2026-03-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-nano-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-nano-2026-03-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"muse-spark","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"muse-spark","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"muse-spark","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"muse-spark","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5.4-2026-03-05_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5.4-2026-03-05_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5.4-2026-03-05_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-pro-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5.4-2026-03-05_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.4-2026-03-05_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.4-2026-03-05_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.4-pro-2026-03-05-web-app","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.4-pro-2026-03-05_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.4-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.4-2026-03-05_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5.3-codex_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gemini-3.1-pro-preview-customtools","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-sonnet-4-6_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-sonnet-4-6_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-sonnet-4-6_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-sonnet-4-6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-sonnet-4-6_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"glm-5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-3.1-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-3.1-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-3.1-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gemini-3.1-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"glm-5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gemini-3-flash-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-opus-4-6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5.1-2025-11-13_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"kimi-k2.5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5.1-2025-11-13_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"glm-5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-6_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gemini-2.5-pro","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gemini-3-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-opus-4-6_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"glm-5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-6_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"glm-5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-6_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5.2-2025-12-11_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"o3-2025-04-16_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-opus-4-1-20250805","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-4o-2024-11-20","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-opus-4-6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-4.1-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-opus-4-6_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-6_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-6_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-opus-4-6_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-6_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-6_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-6_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-6_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-opus-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-opus-4-5-20251101","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-sonnet-4-5-20250929","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SWE-Bench verified","label":"claude-3-7-sonnet-20250219","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"fireworks/kimi-k2p5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"fireworks/kimi-k2p5","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"fireworks/kimi-k2p5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"fireworks/kimi-k2p5","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SWE-Bench verified","label":"gpt-5-mini-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"zai-org/GLM-4.7","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"zai-org/GLM-4.7","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.2-pro-2025-12-11-webapp","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"fireworks/deepseek-v3p2","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-2.5-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-2.5-flash","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-3-flash-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-3-flash-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-3-flash-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gemini-3-flash-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"fireworks/deepseek-v3p2","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"deepseek-reasoner","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"deepseek-reasoner","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.2-2025-12-11_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.2-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.2-2025-12-11_xhigh","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.2-2025-12-11_xhigh","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"openai/gpt-oss-120b_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"openai/gpt-oss-120b_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.2-2025-12-11_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.2-2025-12-11_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5.2-2025-12-11_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5.2-2025-12-11_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.2-2025-12-11_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.2-2025-12-11_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.2-2025-12-11_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.2-2025-12-11_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.2-2025-12-11_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5.2-2025-12-11_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.2-2025-12-11_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5.2-2025-12-11_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.2-2025-12-11_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.2-2025-12-11_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5.2-2025-12-11_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.2-2025-12-11_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.2-2025-12-11_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5.2-2025-12-11_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"Qwen/Qwen3-235B-A22B-Thinking-2507","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"Qwen/Qwen3-235B-A22B-Thinking-2507","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"Qwen/Qwen3-235B-A22B-Thinking-2507","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"Qwen/Qwen3-235B-A22B-Thinking-2507","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"moonshotai/Kimi-K2-Thinking","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"zai-org/GLM-4.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"zai-org/GLM-4.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"zai-org/GLM-4.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"zai-org/GLM-4.6","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"moonshotai/Kimi-K2-Thinking","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"moonshotai/Kimi-K2-Thinking","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"moonshotai/Kimi-K2-Thinking","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.1-2025-11-13_none","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.1-2025-11-13_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.1-2025-11-13_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-5-20251101","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-5-20251101","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-5-20251101_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-5-20251101_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-5-20251101_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-5-20251101_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-5-20251101_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-5-20251101","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-2.5-pro","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-3-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-3-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-3-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-3-pro-preview","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-3-pro-preview","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gemini-3-pro-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.1-2025-11-13_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.1-2025-11-13_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.1-2025-11-13_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.1-2025-11-13_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o3-2025-04-16_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o3-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o3-2025-04-16_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-2.5-pro","reason":"Mixed or unavailable scorer summary; do not select the best scorer"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-sonnet-4-5-20250929","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-sonnet-4-5-20250929_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o4-mini-2025-04-16_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o3-mini-2025-01-31_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"kimi-k2-thinking-turbo","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5.1-2025-11-13_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5.1-2025-11-13_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5.1-2025-11-13_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5.1-2025-11-13_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-sonnet-4-5-20250929_59K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5-mini-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o4-mini-2025-04-16_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o4-mini-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5-mini-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"grok-4-0709","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"kimi-k2-thinking-turbo","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"kimi-k2-thinking-turbo","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-nano-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5-mini-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5-mini-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5-nano-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5-nano-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5-mini-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5-nano-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-nano-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5-nano-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-5-mini-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-mini-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-mini-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-5-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-2025-08-07_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-4-5-20250929_59K","reason":"Mixed or unavailable scorer summary; do not select the best scorer"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4-0613","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4-0314","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4-0314","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-haiku-4-5-20251001_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-sonnet-4-5-20250929_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-haiku-4-5-20251001_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-haiku-4-5-20251001_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-haiku-4-5-20251001_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-sonnet-4-5-20250929_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-4-5-20250929_32K","reason":"Mixed or unavailable scorer summary; do not select the best scorer"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-haiku-4-5-20251001","reason":"Mixed or unavailable scorer summary; do not select the best scorer"},{"source":"Epoch AI · MATH level 5","label":"claude-haiku-4-5-20251001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-haiku-4-5-20251001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"grok-4-heavy-web-app","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"qwen3-max-2025-09-23","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5-pro-2025-10-06_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5-pro-2025-10-06_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"qwen3-max-2025-09-23","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen3-max-2025-09-23","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-sonnet-4-5-20250929","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-4-5-20250929","reason":"Mixed or unavailable scorer summary; do not select the best scorer"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-sonnet-4-5-20250929","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-sonnet-4-5-20250929","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"glm-4.5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"glm-4.5","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-5-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-5-nano-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-5-nano-2025-08-07_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-5-mini-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"grok-4-0709","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"grok-4-0709","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"o4-mini-2025-04-16_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"o4-mini-2025-04-16_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-mini-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-mini-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-5-nano-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-5-nano-2025-08-07_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5-mini-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5-nano-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5-mini-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5-mini-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-5-nano-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5-nano-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-5-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-5-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-5-nano-2025-08-07_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-20250514_27K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-1-20250805_27K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-1-20250805_27K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-1-20250805","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-1-20250805_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-1-20250805_27K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-opus-4-1-20250805_27K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-opus-4-1-20250805_27K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-opus-4-1-20250805_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-opus-4-1-20250805_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-1-20250805","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-opus-4-1-20250805","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-opus-4-1-20250805","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4o-mini-2024-07-18","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-opus-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-sonnet-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gemini-2.5-pro","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-2.5-pro","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-2.5-pro","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-2.5-pro-preview-06-05","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-2.5-pro-preview-06-05","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-2.5-pro-preview-06-05","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gemini-2.5-pro-preview-06-05","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gpt-4.1-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-sonnet-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"o3-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"grok-3-mini-beta_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-3-5-sonnet-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-3-5-sonnet-20240620","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-opus-4-20250514_27K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"o3-mini-2025-01-31_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-opus-4-20250514_27K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-3-5-sonnet-20240620","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-sonnet-4-20250514_59K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-3-5-sonnet-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-3-7-sonnet-20250219_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"gpt-4.1-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"o4-mini-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"o3-mini-2025-01-31_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"DeepSeek-R1-0528","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"grok-3-beta","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"grok-3-beta","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"claude-sonnet-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"claude-opus-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"o4-mini-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Public","label":"o3-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"magistral-small-2506","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gemini-2.5-pro-preview-06-05","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"qwen3-235b-a22b","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen3-235b-a22b","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gemini-2.5-pro-preview-05-06","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"DeepSeek-R1-0528","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-opus-4-20250514_27K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-7-sonnet-20250219_64K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"grok-3-mini-beta_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"DeepSeek-R1","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-sonnet-4-20250514_59K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"grok-3-beta","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-4-20250514_59K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-sonnet-4-20250514_32K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-4-20250514_32K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-opus-4-20250514_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-opus-4-20250514_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-4-20250514_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-sonnet-4-20250514_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-sonnet-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-sonnet-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-opus-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-opus-4-20250514","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-opus-4-20250514","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-sonnet-4-20250514","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-opus-4-20250514","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-sonnet-4-20250514","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-2.5-flash-preview-05-20","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen-plus-2025-04-28","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"qwen-plus-2025-04-28","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"DeepSeek-V3-0324","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemini-2.5-pro-preview-05-06","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemini-2.5-pro-preview-03-25","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"mistral-medium-2505","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"mistral-medium-2505","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"mistral-medium-2505","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o4-mini-2025-04-16_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o4-mini-2025-04-16_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-2.5-flash-preview-04-17","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o3-2025-04-16_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o3-2025-04-16_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o3-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o4-mini-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"o4-mini-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o3-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"o4-mini-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o4-mini-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"o3-2025-04-16_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o3-2025-04-16_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-4.1-nano-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-4.1-nano-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4.1-mini-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4.1-nano-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4.1-mini-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4.1-mini-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gpt-4.1-nano-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4.1-nano-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-4.1-mini-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-4.1-mini-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-4.1-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-4.1-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4.1-2025-04-14","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4.1-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4.1-2025-04-14","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"grok-3-mini-beta_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"grok-3-mini-beta_low","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"grok-3-mini-beta_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"grok-3-mini-beta_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"grok-3-mini-beta_low","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"grok-3-mini-beta_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"grok-3-mini-beta_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"grok-3-mini-beta_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"grok-3-mini-beta_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"grok-3-beta","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"grok-3-beta","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"grok-3-beta","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"grok-3-beta","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"Llama-4-Maverick-17B-128E-Instruct-FP8","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"Llama-4-Maverick-17B-128E-Instruct-FP8","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"Llama-4-Scout-17B-16E-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"Llama-4-Scout-17B-16E-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Llama-4-Maverick-17B-128E-Instruct-FP8","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"Llama-4-Maverick-17B-128E-Instruct-FP8","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Llama-4-Scout-17B-16E-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"Llama-4-Maverick-17B-128E-Instruct-FP8","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"qwen-turbo-2024-11-01","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"qwen-turbo-2024-11-01","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"qwen-plus-2025-01-25","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen-turbo-2024-11-01","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen-plus-2025-01-25","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"qwen-plus-2025-01-25","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"qwen-max-2025-01-25","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"qwen-max-2025-01-25","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"qwen-max-2025-01-25","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen-max-2025-01-25","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"qwen-max-2025-01-25","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Hermes-2-Theta-Llama-3-70B","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"DeepSeek-V3-0324","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"DeepSeek-V3-0324","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"DeepSeek-V3-0324","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-2.5-pro-exp-03-25","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-small-2501","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"mistral-small-2503","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"mistral-small-2503","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-small-2503","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"claude-3-7-sonnet-20250219_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-7-sonnet-20250219_64K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-3-7-sonnet-20250219_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-3-7-sonnet-20250219_64K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-3-7-sonnet-20250219_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemma-3-27b-it","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-3-7-sonnet-20250219_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-3-5-haiku-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-5-haiku-20241022","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-1.5-flash-8b-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"claude-3-7-sonnet-20250219_32K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-7-sonnet-20250219_32K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"DeepSeek-R1-Distill-Qwen-14B","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"DeepSeek-R1-Distill-Llama-70B","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-7-sonnet-20250219_32K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-2.0-flash-001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"DeepSeek-V3","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"DeepSeek-V3","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Llama-3.1-Tulu-3-70B-DPO","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-2.1","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-2.0","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-2.0-flash-thinking-exp-01-21","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o1-preview-2024-09-12","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-2.0-pro-exp-02-05","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o1-2024-12-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o1-2024-12-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-4o-2024-08-06","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-2.0-flash-001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-3-5-sonnet-20240620","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-4o-2024-08-06","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-3-5-sonnet-20240620","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-1.5-flash-002","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gemini-1.5-flash-002","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-3-5-haiku-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-3-5-haiku-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-3-5-sonnet-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-3-5-sonnet-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o1-mini-2024-09-12_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o1-mini-2024-09-12_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o1-mini-2024-09-12_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o1-mini-2024-09-12_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o1-mini-2024-09-12_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o1-mini-2024-09-12_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-3-7-sonnet-20250219_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-3-7-sonnet-20250219_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o3-mini-2025-01-31_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"o3-mini-2025-01-31_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"grok-2-1212","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"gpt-4o-2024-11-20","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"mistral-large-2411","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"claude-3-7-sonnet-20250219","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"o3-mini-2025-01-31_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Public","label":"grok-2-1212","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"claude-3-7-sonnet-20250219","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"mistral-large-2411","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gpt-4o-2024-11-20","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4.5-preview-2025-02-27","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4.5-preview-2025-02-27","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4.5-preview-2025-02-27","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gpt-4-turbo-2024-04-09","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4-turbo-2024-04-09","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o3-mini-2025-01-31_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4-turbo-2024-04-09","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o1-2024-12-17_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-7-sonnet-20250219_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"DeepSeek-R1","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-7-sonnet-20250219_16K","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"claude-3-7-sonnet-20250219_16K","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Meta-Llama-3-70B-Instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-1.5-pro-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"grok-2-1212","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"Llama-2-70b-chat-hf","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-5-sonnet-20240620","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-1.5-pro-002","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4o-2024-08-06","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-1.5-flash-002","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen2.5-32b-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-large-2402","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-1.5-flash-8b-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-sonnet-20240229","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"DeepSeek-V3","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-1.0-pro-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-7-sonnet-20250219","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-5-sonnet-20241022","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"qwen2.5-72b-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4o-2024-05-13","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-5-haiku-20241022","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-1.5-flash-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-haiku-20240307","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"claude-3-opus-20240229","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"o3-mini-2025-01-31_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gpt-4o-2024-11-20","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"mistral-large-2411","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"gemini-2.0-flash-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"mistral-large-2407","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-7-sonnet-20250219","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"claude-3-7-sonnet-20250219","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"grok-2-1212","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"o1-2024-12-17_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"o1-2024-12-17_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"o1-mini-2024-09-12_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"o1-mini-2024-09-12_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o3-mini-2025-01-31_high","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"o3-mini-2025-01-31_high","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gemini-2.0-pro-exp-02-05","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-2.0-flash-thinking-exp-01-21","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gemini-2.0-pro-exp-02-05","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gemini-2.0-flash-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gemini-2.0-flash-001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4o-2024-11-20","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4o-2024-11-20","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"Phi-3-medium-128k-instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"DeepSeek-R1","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"phi-4","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"o3-mini-2025-01-31_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"o3-mini-2025-01-31_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"mistral-small-2501","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"qwen2.5-32b-instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"qwen2.5-32b-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"mistral-small-2501","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gemini-1.5-flash-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-2.0","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-opus-20240229","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o1-mini-2024-09-12_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4o-2024-08-06","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"o1-preview-2024-09-12","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-sonnet-20240229","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-5-sonnet-20241022","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4-0125-preview","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-2.1","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-haiku-20240307","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"claude-3-5-sonnet-20240620","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4o-2024-05-13","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-1.5-flash-002","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4-0613","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-1.5-pro-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4-1106-preview","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-4o-mini-2024-07-18","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gemini-1.5-pro-002","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"gpt-3.5-turbo-0125","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gpt-4-0613","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemini-1.5-flash-001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemini-1.5-pro-002","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-3.5-turbo-0125","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemini-1.5-pro-001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4-1106-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-3.5-turbo-1106","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemini-1.5-flash-002","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-2.0","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-3-5-sonnet-20241022","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-3-5-sonnet-20240620","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-3-opus-20240229","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-3-haiku-20240307","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"claude-3-sonnet-20240229","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"Yi-1.5-34B-Chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"Yi-34B-Chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"Llama-3.1-8B-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Llama-3.1-405B-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Llama-3.1-70B-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"grok-2-1212","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gpt-4o-2024-05-13","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4o-mini-2024-07-18","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4o-2024-08-06","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"qwen2-72b-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"qwen1.5-72b-chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"qwen1.5-32b-chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"o1-mini-2024-09-12_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"Hermes-2-Theta-Llama-3-70B","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"Llama-2-70b-chat-hf","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"Meta-Llama-3-70B-Instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"Meta-Llama-3-8B-Instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"Hermes-2-Theta-Llama-3-70B","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Meta-Llama-3-8B-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Meta-Llama-3-70B-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Llama-2-70b-chat-hf","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Yi-1.5-34B-Chat","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Yi-34B-Chat","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gpt-4-0125-preview","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"Mistral-7B-Instruct-v0.3","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"deepseek-llm-67b-chat","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"Mixtral-8x7B-Instruct-v0.1","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gemma-2-9b-it","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"gemma-2-27b-it","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"qwen2.5-72b-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"qwen2-72b-instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Mistral-7B-Instruct-v0.3","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Mixtral-8x7B-Instruct-v0.1","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"deepseek-llm-67b-chat","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"WizardLM-2-8x22B","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"dbrx-instruct","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"WizardLM-2-8x22B","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"dbrx-instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"gemini-1.0-pro-001","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"gemini-1.0-pro-001","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"mistral-large-2402","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"open-mixtral-8x22b","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"mistral-large-2407","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"open-mixtral-8x7b","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"open-mistral-7b","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"ministral-3b-2410","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"open-mistral-7b","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"ministral-8b-2410","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"mistral-large-2407","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"open-mistral-nemo-2407","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"open-mixtral-8x7b","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"open-mixtral-8x22b","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"mistral-large-2402","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Llama-3.2-90B-Vision-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"Llama-3.1-Tulu-3-70B-DPO","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · GPQA diamond","label":"Eurus-2-7B-PRIME","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"Llama-3.1-Tulu-3-70B-DPO","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"Llama-3.3-70B-Instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"qwen2.5-72b-instruct","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · MATH level 5","label":"o1-preview-2024-09-12","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"o1-2024-12-17_medium","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"o1-2024-12-17_medium","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"DeepSeek-V3","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · MATH level 5","label":"DeepSeek-V3","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · GPQA diamond","label":"grok-4-0709","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · OTIS Mock AIME 2024-2025","label":"grok-4-0709","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · FrontierMath-2025-02-28-Private","label":"gemini-2.5-deep-think-2025-08-01-webapp","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · FrontierMath-Tier-4-2025-07-01-Private","label":"gemini-2.5-deep-think-2025-08-01-webapp","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb","label":"gpt-6.1-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb","label":"claude-sonnet-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb","label":"gpt-6-luna_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb","label":"claude-fable-5-1_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb","label":"claude-opus-5-5_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb","label":"gpt-6-sol_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Epoch AI · SpireClimb","label":"gpt-6-astra_max","reason":"Task is outside this reviewed protocol set; external/reprinted and composite results are not admitted"},{"source":"Vals AI","label":"vals-finance-agent-v2: meta/muse-spark-open","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: anthropic/claude-haiku-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: fireworks/ember-1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: stepfun/step-5-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: poolside/laguna-m.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: poolside/laguna-xs.2","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-finance-agent-v2: cohere/command-a-plus-05-2026","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: anthropic/claude-haiku-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: stepfun/step-5-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: fireworks/ember-1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: meta/muse-spark-open","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-code-migration-v1: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: fireworks/ember-1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: anthropic/claude-haiku-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: zai/glm-legal","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: stepfun/step-5-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: meta/muse-spark-open","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: poolside/laguna-m.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: poolside/laguna-xs.2","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-research-v1: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: fireworks/ember-1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: anthropic/claude-haiku-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: stepfun/step-5-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: meta/muse-spark-open","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-emb-v1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-3-pro-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-opus-4-6-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-opus-4-1-20250805-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-sonnet-4-5-20250929-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: meta/muse_spark","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-opus-4-5-20251101-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-opus-4-1-20250805","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: alibaba/qwen3.5-plus-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: minimax/MiniMax-M2.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-5-2025-08-07","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-5.1-2025-11-13","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-5.2-2025-12-11","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-opus-4-20250514","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: zai/glm-5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/o3-2025-04-16","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-opus-4-5-20251101","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-4-0709","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: alibaba/qwen3-max-2026-01-23","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/deepseek-v3p2-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: alibaba/qwen3-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-4-1-fast-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: alibaba/qwen3.5-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-2.5-pro-exp-03-25","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-sonnet-4-20250514-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-2.5-flash-preview-09-2025","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-2.5-flash-preview-09-2025-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: alibaba/qwen3-max-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/o1-2024-12-17","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/deepseek-r1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/deepseek-v3p2","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: zai/glm-4.7","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-3-7-sonnet-20250219-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-5-mini-2025-08-07","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: zai/glm-4.6","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-3-mini-fast-high-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/qwen3-235b-a22b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: zai/glm-4.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: kimi/kimi-k2-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-3-7-sonnet-20250219","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/o4-mini-2025-04-16","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-4.1-2025-04-14","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: minimax/MiniMax-M2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-3-mini-fast-low-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: mistralai/mistral-large-2512","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-4-fast-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/deepseek-v3-0324","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-sonnet-4-20250514","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/llama4-maverick-instruct-basic","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: together/moonshotai/Kimi-K2-Instruct","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/gpt-oss-120b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-2.5-flash-lite-preview-09-2025-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/o3-mini-2025-01-31","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-2.5-flash-lite-preview-09-2025","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-3-5-sonnet-20241022","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-2.0-flash-001","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-4.1-mini-2025-04-14","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-5-nano-2025-08-07","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-2-1212","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-1.5-pro-002","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: mistralai/mistral-medium-2505","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-4-1-fast-non-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-4o-2024-08-06","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/deepseek-v3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-4o-2024-11-20","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: fireworks/gpt-oss-20b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: together/langston/nim/nvidia/llama-3.3-nemotron-super-49b-v1-42e84561","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: grok/grok-4-fast-non-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: together/meta-llama/Llama-3.3-70B-Instruct-Turbo","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: mistralai/mistral-large-2411","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: together/meta-llama/Llama-4-Scout-17B-16E-Instruct","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: together/langston/nim/nvidia/llama-3.3-nemotron-super-49b-v1-42e84561-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: cohere/command-a-03-2025","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: poolside/laguna-xs.2","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: poolside/laguna-m.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: mistralai/magistral-medium-2509","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: mistralai/mistral-small-2503","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: google/gemini-1.5-flash-002","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: mistralai/mistral-small-2402","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: anthropic/claude-3-5-haiku-20241022","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-4.1-nano-2025-04-14","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: openai/gpt-4o-mini-2024-07-18","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: mistralai/magistral-small-2509","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: ai21labs/jamba-large-1.6","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: cohere/command-r-plus","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-mmlu-pro-v1: ai21labs/jamba-mini-1.6","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: anthropic/claude-opus-4-8-claude-code","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: cursor/composer-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: google/barium-bb","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: openai/gpt-5.5-codex","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: openai/gpt-5.5-factory","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: poolside/laguna-m.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: poolside/laguna-xs.2","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: cohere/command-a-plus-05-2026","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-terminal-bench-2-1-v2.1: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-haiku-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: fireworks/ember-1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-opus-4-8-claude-code","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: meta/muse-spark-open","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: stepfun/step-5-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.5-factory","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.3-codex","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.5-codex","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-opus-4-6","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-sonnet-4-6-claude-code","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.2-2025-12-11","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-opus-4-6-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: cursor/composer-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.4","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.2-codex","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.1-2025-11-13","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: zai/glm-5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-sonnet-4-5-20250929-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.1-codex-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: anthropic/claude-opus-4-5-20251101-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5-2025-08-07","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: meta/muse_spark","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: alibaba/qwen3.5-plus-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: minimax/MiniMax-M2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: google/gemini-3-pro-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5-mini-2025-08-07","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: grok/grok-build-0.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: openai/gpt-5.1-codex","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: alibaba/qwen3.6-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: poolside/laguna-m.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: devin/swe-1-6-fast","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: poolside/laguna-xs.2","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: fireworks/deepseek-v3p2-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: alibaba/qwen3-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: zai/glm-4.6","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: grok/grok-4-1-fast-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: google/gemini-2.5-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: grok/grok-4-fast-reasoning","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-vibe-code-bench-v1.1: mistralai/mistral-small-2603","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: zai/glm-legal","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: meta/muse-spark-open","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: stepfun/step-5-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: fireworks/ember-1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: openai/gpt-6.1-sol","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: openai/gpt-6-luna","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: openai/gpt-6-sol","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: ant/ling-3.0-flash-2607","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: anthropic/claude-haiku-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: nvidia/nemotron-3-ultra-550b-a55b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: poolside/laguna-m.1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: poolside/laguna-xs.2","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: kimi/kimi-k2.5-thinking","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-legal-agent-benchmark-v1: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: meta/muse_spark_1_3_max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: anthropic/claude-opus-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: fireworks/ember-1","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: alibaba/qwen3.8-max","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: xiaomi/mimo-v2.6-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: tencent/hy4-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: deepseek/deepseek-v4.1-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: anthropic/claude-haiku-5-5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: alibaba/qwen3.8-27b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: xiaomi/mimo-v2.6-flash","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: stepfun/step-5-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: deepseek/deepseek-v4-pro","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: meta/muse-spark-open","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: ant/ling-3.0-flash-af-rc3","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: thinkingmachines/inkling","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: thinkingmachines/inkling-small","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: google/gemini-3.1-flash-lite-preview","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: fireworks/nemotron-lightning-3p5-30b-a3b","reason":"No reviewed catalog checkpoint mapping"},{"source":"Vals AI","label":"vals-tax-agent-bench-v1: inception/mercury-2.5","reason":"No reviewed catalog checkpoint mapping"},{"source":"Artificial Analysis · Cyber Index","label":"GPT-6 Sol (Daybreak Blue, max)","reason":"Trusted-access unmitigated variant has no distinct catalog identity. Never fold into public GPT-6 Sol."},{"source":"Artificial Analysis · Cyber Index","label":"Qwen3.8 2.4T A95B","reason":"No reviewed exact catalog identity"},{"source":"bughunt-bench","label":"Union Alpha (stealth; free) - mean of 3","reason":"No exact catalog identity. Stealth or composite label is not a catalog model."},{"source":"bughunt-bench","label":"MiMo-V2.6-Pro - mean of 3","reason":"No exact catalog identity. MiMo-V2.6-Pro is not in the catalog."},{"source":"bughunt-bench","label":"DeepSeek V4.1 Flash (max effort) - mean of 3","reason":"No exact catalog identity. DeepSeek V4.1 Flash is not in the catalog."},{"source":"dayjob-healthcare","label":"DeepSeek V4 Pro","reason":"No exact catalog identity. Unversioned DeepSeek V4 Pro does not match a dated catalog checkpoint."},{"source":"dayjob-healthcare","label":"Inkling","reason":"No exact catalog identity. Inkling is not in the catalog."},{"source":"dayjob-healthcare","label":"DeepSeek V4 Flash","reason":"No exact catalog identity. Unversioned DeepSeek V4 Flash does not match a dated catalog checkpoint."},{"source":"dayjob-finance","label":"DeepSeek V4 Pro","reason":"No exact catalog identity. Unversioned DeepSeek V4 Pro does not match a dated catalog checkpoint."},{"source":"dayjob-finance","label":"DeepSeek V4 Flash","reason":"No exact catalog identity. Unversioned DeepSeek V4 Flash does not match a dated catalog checkpoint."},{"source":"dayjob-finance","label":"Inkling","reason":"No exact catalog identity. Inkling is not in the catalog."},{"source":"toolathlon-verified","label":"DeepSeek V4 Pro (max)","reason":"No exact catalog identity. Unversioned DeepSeek V4 Pro does not match the dated 0813 or 0424 checkpoint."},{"source":"toolathlon-verified","label":"Inkling-Small (xhigh)","reason":"No exact catalog identity. Inkling-Small is not in the catalog."},{"source":"toolathlon-verified","label":"DeepSeek V4 Flash (max)","reason":"No exact catalog identity. Unversioned DeepSeek V4 Flash does not match the dated 0731 checkpoint."},{"source":"toolathlon-verified","label":"Inkling (xhigh)","reason":"No exact catalog identity. Inkling is not in the catalog."},{"source":"toolathlon-verified","label":"Kimi K2.5","reason":"No exact catalog identity. Kimi K2.5 is not in the catalog."},{"source":"thinkingbox-bench","label":"DeepSeek-V4-Pro","reason":"No exact catalog identity. Unversioned DeepSeek-V4-Pro does not name the April or August checkpoint."},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"gpt-6.1-sol_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"claude-sonnet-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SpireClimb (Slay the Spire 1, Ascension 20 Heart)","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"Epoch AI · SpireClimb (Slay the Spire 2, Ascension 10)","label":"claude-opus-5-5_max","reason":"No reviewed exact checkpoint and setting mapping"},{"source":"swe-sweep","label":"Astra 6","reason":"SWE-sweep fraction is not an integer bug count over 4068"},{"source":"swe-sweep","label":"Fable 5.1","reason":"SWE-sweep slug fable-max names effort max, and the public label does not. Unreviewed configuration."},{"source":"agents-last-exam","label":"Agnes Harness · Agnes-2.5-Pro-Beta · Not reported","reason":"No exact catalog identity. Agnes-2.5-Pro-Beta is not Agnes 2.5 Pro Alpha."},{"source":"agents-last-exam","label":"Cursor CLI · composer-2-5 · Adaptive","reason":"No exact catalog identity. Composer 2.5 is not in the catalog."},{"source":"agents-last-exam","label":"Claude Code · ark-0614c · -","reason":"codename ark-0614c is not a catalog id; board label seed-2.1-pro was not auto-bound to doubao-seed-2.1-pro"},{"source":"agents-last-exam","label":"OpenClaw CLI · ark-0614c · High","reason":"codename ark-0614c is not a catalog id; board label seed-2.1-pro was not auto-bound to doubao-seed-2.1-pro"},{"source":"agents-last-exam","label":"OpenClaw · deepseek-v4-pro · High","reason":"No exact catalog identity. Unversioned deepseek-v4-pro does not identify the 0813 checkpoint versus 0424."},{"source":"lmarena-agent","label":"Step 5 Preview","reason":"No exact catalog identity. Step 5 Preview is not in the catalog."},{"source":"lmarena-agent","label":"MiMo V2.6 Flash","reason":"No exact catalog identity. MiMo V2.6 Flash is not in the catalog."},{"source":"lmarena-agent","label":"Inkling Small","reason":"No exact catalog identity. Inkling Small is not a catalog model."},{"source":"lmarena-agent","label":"Inkling","reason":"No exact catalog identity. Inkling is not a catalog model."},{"source":"frontierswe-v2","label":"Inkling","reason":"No exact catalog identity. The source row remains in the raw capture."}]},"collections":[{"collection":"evaluators","datasetId":"sha256-49305f3d640f7387a764d95d62f10bbce44aeff4a4012ddba973eddf1e57a707","collectedOn":"2026-10-18"},{"collection":"vulcan","datasetId":"sha256-443256461e035ccdee585c01fdbd68b4b26545d45e2bd6b2bd93e1e913a2d41a","collectedOn":"2026-10-18"},{"collection":"frontier","datasetId":"sha256-32c6374f8a0cdb9cd64f17244f1a697dd35d31cff071002599f3e1226d1e027d","collectedOn":"2026-10-18"},{"collection":"scale","datasetId":"sha256-9b7bfe52f0cba33df0362a330cb12551ea2f5853bf843931b08dad98e5d8d15c","collectedOn":"2026-10-18"},{"collection":"public-panels","datasetId":"sha256-c4f243913106bee769c57849d3cb4662fee3097206d30d3c8d4d37d687f8e556","collectedOn":"2026-10-18"},{"collection":"epoch","datasetId":"sha256-c6f9cb8b67b4289e79e621b65d529e5f57eee55d89d86915f397859092051dad","collectedOn":"2026-10-18"},{"collection":"vals","datasetId":"sha256-78cb5ce46bffe96a698a9a7117551eafb90b2c3d05066722f4a703be2680aa55","collectedOn":"2026-10-18"},{"collection":"realswe","datasetId":"sha256-e34a2de55476481fa69ea734e4027620fbc14df27f896c3bcec6c07275b0da42","collectedOn":"2026-10-18"}]}