{"version":"configuration-source-5","evidencePolicy":"documented","snapshotId":"sha256-df858793dcb753b0f77beb223fce04e1de99a2c08f2c7b5f14534efcafa8c7c5","effortDatasetId":"sha256-1f186994047883a45f76e96fbce2f07320baf4f104506414be2bba69a334a02f+sha256-2bc644d4e2fb34bd4fecd1ef0a2a50302a9d18f3b8cd3aa0e6dfc487669f5146+sha256-b6d2e58f71798704ee9f3e6b341e3627e6b0dc1caef890fc1e08b62a226e3f36+sha256-7a7d95c40c608cc28483fb5cef9007da8c102f1e7278882ccbcc02a16078607d+sha256-99e5ccb2eac4abb63cc6d9823b591f2ffac992cdee03ac725f66bd8a5ec086a3+sha256-9fbe9729fecc9cf9232f3292684307ba1a85d69beff3920a2a7ed9e2c77bf7ef+sha256-7e00d90a437b5a33e79607a04b57b10dc14322bbaa63ce8bbe9ce1338eaf26f5","scope":"All collected observations, not a claim that every public result has been collected. Contributing means positive graph weight, not sufficient evidence for a rank.","models":[{"id":"gpt-5.4","collected":110,"contributing":73,"admittedWithoutWeight":0,"excluded":37}],"observations":[{"id":"b70ff21cadeade5250a4cdd4","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/apex-agents-aa","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"3f309b1f6f08733ebeaa0ef9","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"95d78edd094271c8788986f1","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"1f2febde52d8d84fd8ad145e","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"6c4e1bd0b148f6dd51d255ec","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/critpt","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"2d984aba50e03049fd01ac4a","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/critpt","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"d37b112d29dad129f8c2272c","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/critpt","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"46bbf2825e58cadd1a201fc1","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/gdpval-aa","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"800c2b0a0a70be8dcb0b707c","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/gpqa-diamond","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"39ae139dd203eb5f9760d5fe","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/gpqa-diamond","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"2344e5d6dcea1f546cabda9b","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/gpqa-diamond","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"1c96c49dbfd0aa0da502f5d7","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/humanitys-last-exam","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"b486a1fb6c5b341875d5225f","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/humanitys-last-exam","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"1f5b6d8b0cbb04daf83a73aa","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/humanitys-last-exam","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"7877f48e230cc4baeef262b1","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/ifbench","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"df4e5119a87760b18053846e","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/ifbench","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"0e811f1b201ee72834eab6f8","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/ifbench","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"de2513acb4a89ca7531cc863","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/mmmu-pro","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"3aea57a33fca8a2445fb7b9b","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/mmmu-pro","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"a6f6ccba2bca3ef7bf7d0890","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/mmmu-pro","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"d716b0e63c7c102a4901c761","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","status":"excluded","reason":"Benchmark family or source protocol has not been reviewed"},{"id":"0f3e47626675965f1c173e90","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","status":"excluded","reason":"Benchmark family or source protocol has not been reviewed"},{"id":"c500501b1945eea9d57c2159","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","status":"excluded","reason":"Benchmark family or source protocol has not been reviewed"},{"id":"5baf6141fd2ae3e320d68dc7","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"a69fa647e71e7ea817df5065","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"f673516a17545c4423d632a5","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/omniscience","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"bb7fd95206fa5df765473dad","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/tau2-bench","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"d29f2bc805408ffd2df1fcd8","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/tau2-bench","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"e6cf2569a32871892e4f8b8d","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/tau2-bench","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"7a637545b18b22b592f62617","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/tau3-banking","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"e041f09eacf7d4fd39fd06fc","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://artificialanalysis.ai/evaluations/terminalbench-hard","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"9c16de1fecb482726fe8e501","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-non-reasoning","sourceUrl":"https://artificialanalysis.ai/evaluations/terminalbench-hard","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"75d0d7792bc9a3baa87c3bb0","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/terminalbench-hard","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"be1c7cf00872b8462a1131fe","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://artificialanalysis.ai/evaluations/terminalbench-v2-1","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"b0856c22135884a80af840b3","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-high","sourceUrl":"https://arcprize.org/media/data/leaderboard/v2.json","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"061ce8db3e09b74846381514","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-low","sourceUrl":"https://arcprize.org/media/data/leaderboard/v2.json","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"f3c4874853691ebdb76d0676","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-medium","sourceUrl":"https://arcprize.org/media/data/leaderboard/v2.json","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"5a6430a8d8f773ee0e4b369f","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://arcprize.org/media/data/leaderboard/v2.json","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"38295b12f6b510281d185208","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://deepswe.datacurve.ai/artifacts/v1.1/leaderboard-live.json","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"62c54d2819ec816195f7efee","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://labs.scale.com/leaderboard/swe_bench_pro_public","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"07b714836f4d50d2545a5455","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-tw","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"34bc44371787dfc935087f65","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-tw","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"595d203626f42c7eaf274a4d","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-refactoring","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"79db12d910dc9b17468fc07c","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://labs.scale.com/leaderboard/sweatlas-qna","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-277c823db47aca935fcfe21d","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-32aa6b03fa6b1e2b8df86c7b","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-e93f929cb0853d27c95d04b3","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-424fc5747345bdb81336e06a","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-686153cb0b44a71199a0619d","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-ead1203dc409c40234d5fee4","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-53e2e2b8cbdb216a093034bf","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-ce06ee813be216293e7004cf","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-14b8986f6045509b1190f001","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-387ee41f45c34973406f6a99","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-6212f1adcca2531c22bd97dc","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-6fda03e93661f6c58ea5e7e4","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-cd52404d183578937e22deac","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-6f5160d5e1a3874da330f091","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-52f73ccb84c0dad0faf4db74","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-930496933f2794492ab365b2","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-4844de203c7156b6d863ace0","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-eca935ff02296892416ebe55","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-dd7582c72728fb5b3755e281","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-dd23f7bf3419e3011e2563f8","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-1a73cedae57299fa448f67d8","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-699ecb5439422ac056781a58","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-52b264b698e93d9c6004f43d","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://livebench.ai/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-ef0926b5f5d004f8a5348da5","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-medium","sourceUrl":"https://swe-rebench.com/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-91308a65440d721514f94949","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-medium","sourceUrl":"https://swe-rebench.com/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"panel-f0febcfcf5b14d5f6f8f1b5e","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-medium","sourceUrl":"https://swe-rebench.com/","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"vals-code-migration-v1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://www.vals.ai/benchmarks/code-migration","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"vals-mmlu-pro-v1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://www.vals.ai/benchmarks/mmlu_pro","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"vals-vibe-code-bench-v1.1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://www.vals.ai/benchmarks/vibe-code","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"vals-legal-agent-benchmark-v1:openai/gpt-5.4-2026-03-05","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://www.vals.ai/benchmarks/hlab","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"launch-kimi-k26-card-2107","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2111","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2115","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2119","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2123","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2128","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2132","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2136","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2140","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2144","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2148","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2151","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"launch-kimi-k26-card-2155","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2165","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2175","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2179","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2183","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2187","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2191","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2195","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Cited comparator requires original evaluation provenance; not a new matched run."},{"id":"launch-kimi-k26-card-2199","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2203","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"contributing","reason":"Positive comparison weight in the overall connected graph"},{"id":"launch-kimi-k26-card-2207","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2211","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2215","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2219","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2223","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-kimi-k26-card-2227","modelId":"gpt-5.4","configurationId":"gpt-5.4--effort-xhigh","sourceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md","status":"excluded","reason":"Supporting evidence outside the reviewed capability core"},{"id":"launch-microsoft-1387","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://microsoft.ai/pdf/mai-thinking-1.pdf","status":"excluded","reason":"No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately"},{"id":"launch-microsoft-1395","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://microsoft.ai/pdf/mai-thinking-1.pdf","status":"excluded","reason":"Microsoft report Table 11 and section 4.1: MAI Terminal-Bench removes timeouts and uses a minimal ReAct harness; comparator values are cited from other model releases. A common evaluation protocol is not established."},{"id":"launch-microsoft-1404","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://microsoft.ai/pdf/mai-thinking-1.pdf","status":"excluded","reason":"No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately"},{"id":"s-arc2-gpt-5.4-efebf61be277","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://arcprize.org/leaderboard","status":"excluded","reason":"No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately"},{"id":"s-deepswe-gpt-5.4-eb88d1c756c0","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://deepswe.datacurve.ai/","status":"excluded","reason":"No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately"},{"id":"s-lmarena-gpt-5.4-465fe8c26c46","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://lmarena.ai/leaderboard","status":"excluded","reason":"No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately"},{"id":"s-mmlu-pro-gpt-5.4-df03094a9acf","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://huggingface.co/spaces/TIGER-Lab/MMLU-Pro","status":"excluded","reason":"No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately"},{"id":"s-swepro-gpt-5.4-6112d01393e8","modelId":"gpt-5.4","configurationId":null,"sourceUrl":"https://scale.com/leaderboard/swe_bench_pro_public","status":"excluded","reason":"No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately"}]}