Claude Opus 4.1 ranks first on 2 of the 128 benchmarks with five or more models scored.
Closed weights Flagship
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Phare | 76.9% | #5 of 62 | Official leaderboard | |
| MMMLU | 89.5% | #10 of 71 | Lab model card | |
| LMArena (Elo) | 1,450 | #63 of 335 | Official leaderboard | |
| Global MMLU Lite | 94.3% | #9 of 43 | Official leaderboard | |
| SnorkelSpatial | 45.1% | #7 of 27 | Official leaderboard | |
| WritingBench | 76.4 | #15 of 45 | Official leaderboard | |
| BullshitBench | 40% | #36 of 107 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 70.5 | #42 of 119 | Official leaderboard | |
| IndicGenBench | 55.0 | #5 of 13 | Official leaderboard | |
| MultiNRC | 38.4% | #16 of 38 | Official leaderboard | |
| BullshitBench V2 | 43% | #61 of 129 | Official leaderboard | |
| MultiChallenge | 57.2% | #20 of 42 | Official leaderboard | |
| DesignArena ASCII Art | 1,162 | #37 of 77 | Official leaderboard | |
| Vectara Hallucination Leaderboard (HHEM) | 4.20% | #72 of 148 | Official leaderboard | |
| LLM Confabulation Leaderboard | 17.1% | #30 of 58 | Official leaderboard | |
| SimpleQA Verified | 31% | #29 of 56 | Official leaderboard | |
| Humanity's Last Exam | 11.5% | #27 of 47 | Official leaderboard | |
| FACTS Parametric | 33.2% | #26 of 45 | Official leaderboard | |
| FACTS Search | 65% | #21 of 31 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 11.8% | #76 of 104 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SciArena | 1,126 | #2 of 34 | Official leaderboard | |
| MATH-500 | 95.4% | #4 of 38 | Official leaderboard | |
| CORE-Bench | 51.1% | #3 of 17 | Official leaderboard | |
| ForecastBench | 0.13 | #26 of 82 | Official leaderboard | |
| SciPredict | 22.2% | #4 of 12 | Official leaderboard | |
| SimpleBench | 60% | #44 of 106 | Official leaderboard | |
| GPQA Diamond | 77.3% | #103 of 229 | Epoch AI run | |
| DTBench | 80% | #92 of 180 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 68.9% | #113 of 212 | Epoch AI run | |
| SGI-Bench | 30.4% | #8 of 15 | Official leaderboard | |
| FrontierMath Tier 4 | 4.17% | #31 of 55 | Epoch AI run | |
| FrontierMath | 7.24% | #35 of 55 | Official leaderboard | |
| AIME 2025 | 78% | #96 of 143 | Lab model card | |
| CMT-Benchmark | 8.00% | #8 of 11 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 2.44% | #61 of 68 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 12.6% | #81 of 86 | Official leaderboard | |
| IMProofBench | 3.38% | #9 of 9 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ArtifactsBench | 59.8% | #2 of 20 | Official leaderboard | |
| METR Time Horizon | 109 min | #5 of 31 | Official leaderboard | |
| METR Time Horizon 1.1 | 100 min | #12 of 26 | Official leaderboard | |
| SWE-fficiency | 9.84% | #8 of 17 | Official leaderboard | |
| WeirdML | 45.9% | #58 of 112 | Official leaderboard | |
| SWE-bench Verified (hard subset) | 43.8% | #5 of 9 | Lab model card | |
| DesignArena Game Dev | 1,161 | #80 of 143 | Official leaderboard | |
| IOI (Vals) | 12.5% | #31 of 55 | Official leaderboard | |
| DesignArena SVG | 1,143 | #59 of 103 | Official leaderboard | |
| DesignArena 3D Design | 1,136 | #79 of 136 | Official leaderboard | |
| DesignArena Website | 1,170 | #90 of 151 | Official leaderboard | |
| DesignArena Code | 1,161 | #87 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,156 | #86 of 139 | Official leaderboard | |
| SWE-bench Verified | 73.3% | #20 of 32 | Epoch AI run | |
| WebDev Arena | 1,389 | #71 of 105 | arena.ai, via Epoch AI | |
| ALE-Bench | 1,183 | #87 of 126 | Official leaderboard | |
| DesignArena Data Visualization | 1,164 | #83 of 118 | Official leaderboard | |
| Code Arena | 1,389 | #91 of 128 | Official leaderboard | |
| SWE-Bench Pro Private | 17.8% | #10 of 12 | Official leaderboard | |
| Quadruped reinforcement learning task (Anthropic) | 1.18 | #7 of 8 | Lab model card | |
| AlgoTune | 1.34× | #16 of 18 | Official leaderboard | |
| Time series forecasting task (Anthropic) | 6.54 | #9 of 9 | Lab model card | |
| Text-based reinforcement learning task (Anthropic) | 0.42 | #7 of 7 | Lab model card | |
| FlashInfer-Bench | 0.46× | #2 of 2 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CompileBench | 100% | #1 of 20 | Official leaderboard | |
| Elimination Game | 5 | #10 of 56 | Official leaderboard | |
| tau-bench | 69.2% | #5 of 23 | Lab model card | |
| Terminal-Bench | 38% | #4 of 18 | Terminal-Bench v2 Leaderboard, via Epoch AI | |
| MCP-Universe | 29.4% | #8 of 30 | Official leaderboard | |
| C3-Bench | 9.38% | #9 of 30 | Official leaderboard | |
| MCPMark | 29.9% | #9 of 28 | Official leaderboard | |
| Step Race | 3 | #21 of 59 | Official leaderboard | |
| DeepResearch Bench | 48.3% | #8 of 22 | evals.futuresearch.ai, via Epoch AI | |
| Deep Research Bench (FutureSearch) | 0.48 | #8 of 21 | Official leaderboard | |
| LisanBench | 394.7 | #72 of 133 | Official leaderboard | |
| Mystery Game Puzzles | 21% | #46 of 84 | Epoch AI run | |
| LLM Chess | -38 | #68 of 123 | Official leaderboard | |
| Terminal-Bench 2.0 | 38% | #25 of 43 | Official leaderboard | |
| TextQuests | 19.1 | #19 of 28 | Official leaderboard | |
| LMArena Search Arena | 1,167 | #30 of 37 | Official leaderboard | |
| LiveResearchBench | 61.8 | #6 of 7 | Official leaderboard | |
| EBR-bench | 7.94% | #28 of 31 | Epoch AI run | |
| DeR2 Bench | 72.4% | #5 of 5 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SnorkelUnderwrite | 86.3% | #2 of 27 | Official leaderboard | |
| Vals Finance Agent v1 | 50.9% | #6 of 47 | Official leaderboard | |
| FinQA (Snorkel OpenEnv environment) | 80.3% | #4 of 22 | Official leaderboard | |
| SnorkelFinance | 80.3% | #4 of 22 | Official leaderboard | |
| GDPval | 47.5% | #1 of 5 | Official leaderboard | |
| MedCode | 47.2% | #34 of 96 | Official leaderboard | |
| TaxEval v2 | 73.7% | #46 of 129 | Official leaderboard | |
| LegalBench | 83.5% | #54 of 138 | Official leaderboard | |
| Finance Reasoning (Snorkel) | 45.6% | #10 of 21 | Official leaderboard | |
| MortgageTax | 61.1% | #59 of 87 | Official leaderboard | |
| TaxCalcBench | 28.4% | #13 of 17 | Official leaderboard | |
| MedScribe | 73.9% | #76 of 97 | Official leaderboard | |
| PRBench | 34% | #46 of 48 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU | 77.1% | #16 of 136 | Official leaderboard | |
| Butter-Bench | 37% | #1 of 4 | Official leaderboard | |
| MATH-Vision (MathVision) | 66% | #28 of 105 | Official leaderboard | |
| VISTA | 48.4% | #15 of 54 | Official leaderboard | |
| Math-VR | 54.3% | #8 of 26 | Official leaderboard | |
| MeasureBench Synthetic | 13.3% | #8 of 19 | Official leaderboard | |
| ChartMuseum | 59.1% | #11 of 21 | Official leaderboard | |
| MeasureBench Real-world | 14.3% | #10 of 19 | Official leaderboard | |
| ZeroBench | 8.00% | #42 of 76 | Official leaderboard | |
| VPCT | 35% | #22 of 27 | VPCT leaderboard, via Epoch AI | |
| ClockBench | 8.90% | #31 of 37 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 102.6 | #19 of 185 | Official leaderboard | |
| BBQ (Bias Benchmark for QA) | 90.7% | #4 of 25 | Lab model card | |
| FORTRESS | 14.8% | #14 of 73 | Official leaderboard | |
| Molecular Biology Capabilities Test | 55.1% | #20 of 90 | Official leaderboard | |
| Human Pathogen Capabilities Test | 60.1% | #24 of 86 | Official leaderboard | |
| Single-turn harmful request evaluations (Anthropic) | 98.8% | #7 of 19 | Lab model card | |
| Anthropic political even-handedness eval | 95% | #2 of 5 | Official leaderboard | |
| LAB-Bench CloningScenarios | 56% | #5 of 12 | Lab model card | |
| SCONE-bench | $410,200,000 | #3 of 7 | Official leaderboard | |
| BABE | 37.4% | #3 of 6 | Official leaderboard | |
| Single-turn benign request evaluations (Anthropic) | 0.08% | #10 of 19 | Lab model card | |
| World-Class Bio | 42.8% | #45 of 74 | Official leaderboard | |
| Cybench | 51.4% | #11 of 18 | Lab model card | |
| CyberGym | 25% | #24 of 34 | Official leaderboard | |
| Creative biology | 48% | #5 of 7 | Lab model card | |
| LAB-Bench SeqQA | 71% | #9 of 12 | Lab model card | |
| LAB-Bench ProtocolQA | 69% | #13 of 16 | Lab model card | |
| Impossible tasks (Anthropic) | 52% | #9 of 11 | Lab model card | |
| LAB-Bench FigQA | 46% | #10 of 11 | Lab model card | |
| Reward-hack-prone coding tasks (Anthropic) | 12% | #8 of 8 | Lab model card | |
| Average in-eval reward hacking rate (Anthropic) | 18.2% | #5 of 5 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LMCA | 37.1% | #72 of 147 | Official leaderboard | |
| Conceptual Reasoning Index | 48.3 | #77 of 151 | Official leaderboard | |
| ACCoRD | 63.5 | #80 of 152 | Official leaderboard | |
| TutorBench | 50.8% | #15 of 20 | Official leaderboard | |
| SAGE (Vals) | 31.4% | #72 of 81 | Official leaderboard |
The Self-Reported Index fits one score to the 8 capability results Anthropic reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 139.5 to 146.2. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| GPQA Diamond | 80.9% Graduate-level reasoning GPQA Diamond | 77.3% Epoch AI run, 16K | +3.6 pts |
| SWE-bench Verified | 74.5% Agentic coding SWE-bench Verified | 73.3% Epoch AI run | +1.2 pts |
| MMMU | 77.1% Visual reasoning MMMU (validation) | 77.1% Leaderboard | Same |
Anthropic's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
The benchmarks the lab chose to headline, with the scores it gave.
Every result in the longer documents the lab published with it. 29 of them aren't in the launch post.