RankingEffort benchmarks
Step 5 Preview
20 measured results · 2 effort settings · 20 benchmark/harness combinations
Change model
Capability profile
One configuration, seven capabilities
- AStep 5 Preview · Reasoning2 of 7 capabilities supported
Gaps are unknown, not zero. Hollow points are preliminary.
Agentic
A55.2SupportedHard reasoning
A54.6SupportedCoding
A71.4PreliminaryHuman pref
UnknownKnowledge
A42.4PreliminaryMultimodal
A29.2PreliminaryLong context
A65.3Preliminary
Each profile belongs to the named configuration. Capability scores use a shared panel for all axes, distinct from the overall ranking panel; they are not accuracy percentages or direct head-to-head wins.
Capability score evidence
Step 5 Preview · Reasoning: 47.3 overall comparison score. Meets the overall evidence requirements.
Which sources carry the weight?
- Artificial Analysis · 100.0% direct comparison weight · independent of the model provider
Which benchmark families carry the weight?
- Humanity’s Last Exam · 8.1% direct comparison weight
- Critpt · 8.1% direct comparison weight
- Mmmu Pro · 8.0% direct comparison weight
- GDPval · 7.5% direct comparison weight
- Aa Lcr · 7.4% direct comparison weight
- Other families (10) · 60.9% direct comparison weight
These shares describe direct comparison weight, not fractions of the final score or confidence. Source balancing does not give every family equal weight or remove correlations between overlapping sources.
How sensitive is this score?
The audit removes one source at a time and refits against the same overall reference panel. These published results use the matching dataset audited 2026-11-22.
Both ranks use the same surviving configurations for each removal. The cohort starts with 602 configurations supported in the original audit, including separate effort settings. These are diagnostic positions, not the default leaderboard ranks. Eligibility is not rechecked after removal.
| Removed source | Rank before → after | Rank movement | Cohort remaining / lost support | Score after (change) |
|---|---|---|---|---|
| Artificial Analysis | Comparison support lost; no rank | 230 remaining / 372 lost | Unavailable | |
| ARC Prize | 110 → 108 | Up 2 | 602 remaining / 0 lost | 52.7 |
| Datacurve | 110 → 112 | Down 2 | 602 remaining / 0 lost | 43.1 |
| LiveCodeBench | 110 → 110 | No movement | 602 remaining / 0 lost | 47.3 |
| Scale | 110 → 110 | No movement | 602 remaining / 0 lost | 49.0 |
| Terminal-Bench | 110 → 109 | Up 1 | 602 remaining / 0 lost | 47.5 |
| Cognition | 110 → 113 | Down 3 | 602 remaining / 0 lost | 45.5 |
| LiveBench | 110 → 111 | Down 1 | 601 remaining / 1 lost | 48.5 |
| SWE-rebench | 110 → 110 | No movement | 602 remaining / 0 lost | 47.5 |
| Epoch AI | 109 → 110 | Down 1 | 588 remaining / 14 lost | 47.8 |
| Vals AI | 110 → 111 | Down 1 | 601 remaining / 1 lost | 46.6 |
| Provider reports · Anthropic | 110 → 110 | No movement | 602 remaining / 0 lost | 47.3 |
| Provider reports · Moonshot AI | 110 → 110 | No movement | 602 remaining / 0 lost | 46.8 |
| Provider reports · Meta | 110 → 111 | Down 1 | 602 remaining / 0 lost | 47.3 |
| Provider reports · Mistral | 110 → 110 | No movement | 602 remaining / 0 lost | 47.3 |
| Provider reports · OpenAI | 110 → 110 | No movement | 602 remaining / 0 lost | 47.5 |
| All provider reports | 110 → 111 | Down 1 | 602 remaining / 0 lost | 46.9 |
Lost support means a configuration no longer has a comparison path to the complete reference panel. Its rank is unavailable. A surviving path alone does not establish leaderboard eligibility.
Overlapping sources may remain correlated even with source balancing. Leave-one-source-out movement is a sensitivity diagnostic, not a spread of source-specific ranks or a confidence measure. Scoring methodology · Dated configuration audit
Every benchmark family
| Removed evidence | Recomputed score | Change |
|---|---|---|
| arc-agi | 52.0 | +4.7 |
| deepswe | 43.2 | -4.2 |
| mmmu-pro | 50.6 | +3.2 |
| terminal-bench-science | 50.5 | +3.2 |
| itbench | 45.3 | -2.0 |
| frontiercode | 45.5 | -1.8 |
| enterprise-ops | 45.6 | -1.8 |
| gdp-pdf | 49.0 | +1.7 |
| omniscience | 48.7 | +1.3 |
| aa-lcr | 46.3 | -1.1 |
| gdpval | 46.4 | -0.9 |
| apex-agents | 46.5 | -0.8 |
| simpleqa | 48.2 | +0.8 |
| analyst-agent | 48.1 | +0.8 |
| epoch-game-puzzles | 46.6 | -0.7 |
| ifbench | 46.7 | -0.7 |
| briefcase | 46.7 | -0.7 |
| arc-agi-3 | 46.8 | -0.6 |
| terminal-bench | 46.8 | -0.5 |
| enigma-eval | 47.8 | +0.5 |
| mmlu-pro | 47.8 | +0.4 |
| multi-swe-bench | 47.0 | -0.4 |
| livebench-instructions | 47.7 | +0.4 |
| gpqa | 47.0 | -0.3 |
| livebench-language | 47.6 | +0.3 |
| hle | 47.6 | +0.3 |
| swe-atlas-qna | 47.1 | -0.3 |
| livebench-coding | 47.2 | -0.2 |
| automationbench | 47.5 | +0.2 |
| vibe-code | 47.2 | -0.2 |
| aime | 47.2 | -0.2 |
| tau-bench | 47.2 | -0.2 |
| swe-rebench | 47.5 | +0.1 |
| swe-atlas-test-writing | 47.2 | -0.1 |
| vals-legal-research | 47.2 | -0.1 |
| vals-code-migration | 47.2 | -0.1 |
| livebench-reasoning | 47.2 | -0.1 |
| vals-tax-agent | 47.2 | -0.1 |
| vals-finance-agent | 47.3 | -0.1 |
| critpt | 47.4 | +0.1 |
| harvey | 47.3 | -0.1 |
| vals-excel-modeling | 47.3 | -0.1 |
| swe-atlas-refactoring | 47.3 | -0.1 |
| swe-bench-pro | 47.4 | +0.1 |
| gmmlu | 47.4 | +0.0 |
| frontiermath | 47.3 | -0.0 |
| livecodebench | 47.3 | -0.0 |
| livebench-data | 47.4 | +0.0 |
| livebench-math | 47.3 | -0.0 |
| math500 | 47.4 | +0.0 |
What evidence is missing from the fit?
Across all collected settings for this model: 26 observations, 16 contributing, 0 matched without graph weight and 10 excluded. Counts do not establish rank eligibility.
- 1 · Benchmark family or source protocol has not been reviewed
- 1 · Original SciCode grading defects are under review; retained as supporting evidence, excluded from aggregate and capability scores. Corrected SciCode-Verified requires a separate protocol review. Sources: https://arxiv.org/abs/2608.04975 and https://artificialanalysis.ai/evaluations/scicode
- 2 · Effort is not identified
- 4 · Supporting evidence outside the reviewed capability core
- 2 · No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately
Benchmark results
Reviewed retained observations are in use. October 2026 Real-SWE board has no 95% confidence whisker, so these scores stay at the 2026-10-18 retrieval. Original source retrieval 2026-10-18; source updated 2026-10-18.
| Benchmark / source | Reasoning | Unspecified |
|---|---|---|
AA-AnalystAgentBenchmark detailsShare 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 Unit: percent. Source collected 2026-11-22. | 35% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
AA-Briefcase EloBenchmark detailsAA-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 Unit: elo. Source collected 2026-11-22. | 1,423.95 Elo Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
APEX-Agents-AABenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 37.98% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
AA-LCR v1.1Benchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 88.33% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
AutomationBench-AABenchmark detailsShare 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 Unit: percent. Source collected 2026-11-22. | 50.96% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
CritPtBenchmark detailsBenchmark developed by Argonne and UIUC, with contributions from 60+ researchers globally · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 20.86% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
EnterpriseOps-Gym-AABenchmark detailsOverall 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 Unit: percent. Source collected 2026-11-22. | 47.18% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
GDP.pdf: All-pass RateBenchmark detailsShare of attempts where every atomic criterion passed · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 14.8% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
GDPval-AA v2.1 LeaderboardBenchmark detailsElo 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 Unit: elo. Source collected 2026-11-22. | 1,590.7 Elo Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
Humanity's Last ExamBenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 46.48% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
ITBench-AABenchmark detailsCombined public and private Site Reliability Engineering (SRE) tasks · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 55.56% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
MLCR-AABenchmark detailsShare of responses judged complete with no claim contradicting the record · Expert + compound (hard) sets · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 18.33% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
MMMU-ProBenchmark detailsIndependently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 76.36% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
AA-Omniscience IndexBenchmark detailsAA-Omniscience v1.1 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 Unit: index. Source collected 2026-11-22. | 16.43 Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
AA-Omniscience AccuracyBenchmark detailsAA-Omniscience v1.1 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 Unit: percent. Source collected 2026-11-22. | 36.37% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
SciCodeBenchmark detailsBenchmark 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 Unit: percent. Source collected 2026-11-22. | 58.91% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
Terminal-Bench-Science 0.1Benchmark detailsShare of tasks passed (pass@1), averaged over three repeats · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 2.38% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
Terminal-Bench v4.0Benchmark detailsBenchmark developed by the Laude Institute, Stanford, and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis Unit: percent. Source collected 2026-11-22. | 33.33% Reported settingsStep 5 Preview; Reported effort: reasoning; Model source: https://artificialanalysis.ai/models/step-5; Mode from evaluator isReasoning metadata; effort budget not specified | Not reported |
Artificial Analysis Cyber IndexBenchmark detailsv1.0. 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 Unit: index. Source collected 2026-11-22. | Not reported | 47.71 Reported settingsArtificial 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. |
FrontierCS 2Benchmark detailsPass rate on 48 tasks. 1 run per task. Each row uses that model's vendor harness. Kaggle verification is unfinished. UC Berkeley and Princeton. Display only. Excluded from weighted Overall. https://arxiv.org/abs/2610.29932 https://www.kaggle.com/benchmarks/frontiercs/frontiercs-2 Unit: percent. Source collected 2026-11-22. | Not reported | 8.3% Reported settingsStep 5 Preview; Step Code; effort unspecified; 4 of 48 tasks passed; 1 run per task. Each row uses that model's vendor harness. Kaggle verification is unfinished. UC Berkeley and Princeton. Display only. Excluded from weighted Overall. |
What these comparisons mean
Each source reports its own effort labels. Matching labels do not establish equal compute budgets or identical fallback behavior. Open Reported settings to inspect the exact model name, agent and task coverage. Multiple results at one level remain visible rather than selecting the best. Source-reported None or Non-reasoning labels are not treated as evidence that an API supports those settings.
These measurements use their own dated source collection. The leaderboard fits identified configurations jointly; it contains no pooled model ratings. Explicit thinking budgets are separate configurations. Read the methodology.
Download all effort benchmark results · 10,184 results across 381 models · Collected 2026-11-22