Claude Opus 4 ranks first on 3 of the 138 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 | |
|---|---|---|---|---|
| MMLU-Pro | 87.3% | #12 of 212 | Official leaderboard | |
| AraGen | 81% | #4 of 42 | Official leaderboard | |
| IFEval | 91.8% | #7 of 60 | Official leaderboard | |
| SuperGPQA | 61.4% | #8 of 59 | Official leaderboard | |
| RewardBench 2 | 76.5% | #6 of 42 | Official leaderboard | |
| Meeseeks Utility Rate | 87.6% | #2 of 10 | Official leaderboard | |
| MMMLU | 88.8% | #17 of 71 | Lab model card | |
| FaithJudge | 14.1% | #9 of 37 | Official leaderboard | |
| ECLeKTic | 24% | #10 of 35 | Official leaderboard | |
| LMArena (Elo) | 1,426 | #96 of 335 | Official leaderboard | |
| SnorkelSpatial | 40.3% | #9 of 27 | Official leaderboard | |
| WritingBench | 75.1 | #16 of 45 | Official leaderboard | |
| EQ-Bench 3 | 1,440 | #30 of 78 | Official leaderboard | |
| BullshitBench | 32.7% | #45 of 107 | Official leaderboard | |
| MultiChallenge | 58.6% | #19 of 42 | Official leaderboard | |
| Creative Writing v3 | 70.0 | #55 of 120 | Official leaderboard | |
| LLM Confabulation Leaderboard | 15.9% | #27 of 58 | Official leaderboard | |
| MultiNRC | 33.9% | #19 of 38 | Official leaderboard | |
| FACTS Grounding | 77.7% | #17 of 34 | Official leaderboard | |
| BullshitBench V2 | 34% | #73 of 129 | Official leaderboard | |
| Fiction.LiveBench | 61.1% | #33 of 57 | Fiction.live leaderboard, via Epoch AI | |
| Vectara Hallucination Leaderboard (HHEM) | 4.80% | #91 of 148 | Official leaderboard | |
| Humanity's Last Exam | 10.7% | #29 of 47 | Official leaderboard | |
| HUMAINE | 25 | #32 of 50 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 12% | #78 of 104 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Misguided Attention | 64.1% | #4 of 45 | Official leaderboard | |
| GPQA Main | 70.8% | #9 of 60 | Official leaderboard | |
| SciArena | 1,072 | #6 of 34 | Official leaderboard | |
| MATH Level 5 | 85% | #22 of 94 | Epoch AI run | |
| SnorkelGraph | 64.5% | #8 of 24 | Official leaderboard | |
| AIME 2025 | 90% | #49 of 143 | Lab model card | |
| ForecastBench | 0.13 | #31 of 82 | Official leaderboard | |
| SnorkelSequences | 65.6% | #10 of 25 | Official leaderboard | |
| SimpleBench | 58.8% | #46 of 106 | Official leaderboard | |
| DTBench | 81.6% | #84 of 180 | Official leaderboard | |
| GPQA Diamond | 76.3% | #108 of 229 | Epoch AI run | |
| IneqMath | 5.50% | #17 of 34 | Official leaderboard | |
| FrontierMath Tier 4 | 4.17% | #28 of 55 | Epoch AI run | |
| OTIS Mock AIME 2024-2025 | 64.4% | #121 of 212 | Epoch AI run | |
| MATH-500 | 90.4% | #22 of 38 | Official leaderboard | |
| ARC-AGI-2 | 8.61% | #61 of 98 | Official leaderboard | |
| CMT-Benchmark | 10% | #7 of 11 | Official leaderboard | |
| HMMT 2025 | 60% | #37 of 53 | Official leaderboard | |
| FrontierMath | 4.48% | #42 of 55 | Official leaderboard | |
| ARC-AGI | 35.7% | #74 of 95 | Official leaderboard | |
| AMO-Bench | 10.6% | #17 of 21 | Official leaderboard | |
| BRUMO 2025 | 81.7% | #35 of 42 | Official leaderboard | |
| IMO-ProofBench | 2.90% | #20 of 20 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AutoCodeBench | 52.4% | #1 of 34 | Official leaderboard | |
| Aider Polyglot | 72% | #9 of 46 | Official leaderboard | |
| METR Time Horizon | 84 min | #8 of 31 | Official leaderboard | |
| METR data deduplication | 32.6% | #1 of 3 | Lab model card | |
| METR Time Horizon 1.1 | 100 min | #13 of 26 | Official leaderboard | |
| WeirdML | 43.7% | #64 of 112 | Official leaderboard | |
| DesignArena Game Dev | 1,160 | #82 of 143 | Official leaderboard | |
| Internal model use survey (Anthropic) | 0% | #3 of 5 | Lab model card | |
| DesignArena 3D Design | 1,121 | #84 of 136 | Official leaderboard | |
| GSO-Bench | 6.86% | #19 of 30 | Official leaderboard | |
| DesignArena UI Component | 1,144 | #90 of 139 | Official leaderboard | |
| DesignArena Website | 1,158 | #98 of 151 | Official leaderboard | |
| DesignArena Code | 1,151 | #95 of 143 | Official leaderboard | |
| LiveCodeBench | 76.6% | #14 of 21 | Official leaderboard | |
| LiveCodeBench v5 | 59.8% | #14 of 21 | Official leaderboard | |
| LiveCodeBench v6 | 56.6% | #14 of 21 | Official leaderboard | |
| SWE-bench Verified (hard subset) | 39.5% | #6 of 9 | Lab model card | |
| DesignArena SVG | 1,118 | #71 of 103 | Official leaderboard | |
| SWE-bench Verified | 70.7% | #24 of 32 | Epoch AI run | |
| Quadruped reinforcement learning task (Anthropic) | 1.25 | #6 of 8 | Lab model card | |
| DesignArena Data Visualization | 1,144 | #92 of 118 | Official leaderboard | |
| Text-based reinforcement learning task (Anthropic) | 0.63 | #6 of 7 | Lab model card | |
| Time series forecasting task (Anthropic) | 6.15 | #8 of 9 | Lab model card | |
| AlgoTune | 1.33× | #17 of 18 | Official leaderboard | |
| Internal AI research evaluation suite 2 | 0.35 | #6 of 6 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau-bench | 70.5% | #1 of 23 | Lab model card | |
| Terminal-Bench | 50% | #1 of 22 | Lab model card | |
| MCP-Universe | 28.1% | #9 of 30 | Official leaderboard | |
| lmgame-Bench | 34.6 | #8 of 23 | Official leaderboard | |
| Elimination Game | 4 | #23 of 56 | Official leaderboard | |
| TextQuests | 16.5 | #12 of 28 | Official leaderboard | |
| SnorkelWordle | 85.6% | #9 of 19 | Official leaderboard | |
| LisanBench | 598.0 | #64 of 133 | Official leaderboard | |
| Step Race | 2 | #29 of 59 | Official leaderboard | |
| LLM Chess | 29 | #64 of 123 | Official leaderboard | |
| DeepResearch Bench | 46.8% | #12 of 22 | evals.futuresearch.ai, via Epoch AI | |
| Deep Research Bench (FutureSearch) | 0.47 | #12 of 21 | Official leaderboard | |
| BrowseComp-Plus | 36.1% | #5 of 7 | Official leaderboard | |
| LMArena Search Arena | 1,152 | #33 of 37 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| FinQA (Snorkel OpenEnv environment) | 78.3% | #6 of 22 | Official leaderboard | |
| SnorkelFinance | 78.3% | #6 of 22 | Official leaderboard | |
| Finance Reasoning (Snorkel) | 48.1% | #6 of 21 | Official leaderboard | |
| Vals Finance Agent v1 | 43.6% | #14 of 47 | Official leaderboard | |
| SnorkelUnderwrite | 77% | #9 of 27 | Official leaderboard | |
| LegalBench | 83.1% | #60 of 138 | Official leaderboard | |
| TaxEval v2 | 71.9% | #68 of 129 | Official leaderboard | |
| MortgageTax | 58.6% | #65 of 87 | Official leaderboard | |
| Vending-Bench | $2,077 | #7 of 9 | Official leaderboard | |
| OccuBench | 61.3% | #11 of 14 | Official leaderboard | |
| TaxCalcBench | 27.4% | #14 of 17 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMMU | 76.5% | #17 of 136 | Official leaderboard | |
| VISTA | 47% | #19 of 54 | Official leaderboard | |
| Vision Arena | 1,208 | #74 of 136 | Official leaderboard | |
| ZeroBench | 6.00% | #47 of 76 | Official leaderboard | |
| VPCT | 38% | #19 of 27 | VPCT leaderboard, via Epoch AI | |
| Drone-Bench | 41% | #14 of 19 | Official leaderboard | |
| Blueprint-Bench | 0.35 | #6 of 8 | Official leaderboard | |
| GeoBench | 49% | #43 of 48 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BBQ (Bias Benchmark for QA) | 91.1% | #2 of 25 | Lab model card | |
| Molecular Biology Capabilities Test | 59.1% | #10 of 90 | Official leaderboard | |
| nocot-bench (NCRI) | 101.4 | #22 of 185 | Official leaderboard | |
| AIR-Bench 2024 | 85.7% | #12 of 82 | Official leaderboard | |
| HELM Safety | 96.9 | #12 of 75 | Official leaderboard | |
| SimpleSafetyTests | 100% | #14 of 75 | Official leaderboard | |
| Lakera B3 (Backbone Breaker Benchmark) | 72.6% | #9 of 35 | Official leaderboard | |
| Bioweapons acquisition uplift trial (Anthropic) | 2.53 | #1 of 3 | Lab model card | |
| Average in-eval reward hacking rate (Anthropic) | 15.2% | #2 of 5 | Lab model card | |
| Single-turn harmful request evaluations (Anthropic) | 98.4% | #8 of 19 | Lab model card | |
| Cybench | 56.4% | #8 of 18 | Lab model card | |
| ABC Bench Liquid-Handling Robot | 93.3% | #16 of 34 | Official leaderboard | |
| Single-turn benign request evaluations (Anthropic) | 0.07% | #9 of 19 | Lab model card | |
| Pwn CTF challenges (Anthropic) | 55.6% | #3 of 6 | Lab model card | |
| FORTRESS | 24.8% | #37 of 73 | Official leaderboard | |
| Malicious use of Claude Code (Anthropic) | 88% | #9 of 17 | Lab model card | |
| Anthropic Red Team | 98.9% | #41 of 75 | Official leaderboard | |
| World-Class Bio | 40.3% | #47 of 74 | Official leaderboard | |
| Impossible tasks (Anthropic) | 47% | #7 of 11 | Lab model card | |
| Network CTF challenges (Anthropic) | 50% | #4 of 6 | Lab model card | |
| Rev CTF challenges (Anthropic) | 50% | #4 of 6 | Lab model card | |
| Bioweapons knowledge questions | 51.5% | #2 of 3 | Lab model card | |
| Computer use prompt injection evaluation (Anthropic) | 71% | #4 of 5 | Lab model card | |
| Crypto CTF challenges (Anthropic) | 36.4% | #5 of 6 | Lab model card | |
| Web CTF challenges (Anthropic) | 80% | #5 of 6 | Lab model card | |
| Creative biology | 45% | #6 of 7 | Lab model card | |
| Reward-hack-prone coding tasks (Anthropic) | 9.00% | #7 of 8 | Lab model card | |
| ABC Bench Fragment Design | 71% | #25 of 27 | Official leaderboard | |
| SHADE-Arena | 0% | #16 of 16 | Lab model card | |
| Agentic misalignment suite (Anthropic) | 84% | #7 of 7 | Lab model card | |
| Many-shot jailbreaking evaluation (Anthropic) | 25% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| HELM Capabilities | 78.0 | #8 of 60 | Official leaderboard | |
| ACCoRD | 65.8 | #71 of 152 | Official leaderboard | |
| Conceptual Reasoning Index | 49.5 | #71 of 151 | Official leaderboard | |
| LMCA | 37.4% | #70 of 147 | Official leaderboard | |
| TutorBench | 49.7% | #16 of 20 | 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 142.4 to 149.0. 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 | What differs |
|---|---|---|---|---|
| AIME 2025 | 90% High school math competition AIME 2025 (parallel test-time compute) | 70% Leaderboard, thinking | +20 pts | The lab used parallel test-time compute, while MathArena averages separate individual runs. |
| GPQA Diamond | 83.3% Graduate-level reasoning GPQA Diamond (parallel test-time compute) | 76.3% Epoch AI run, 16K | +7.0 pts | |
| SWE-bench Verified | 79.4% Agentic coding SWE-bench Verified (parallel test-time compute) | 73.2% Leaderboard, Tools | +6.2 pts | |
| MMMU | 76.5% Visual reasoning MMMU (validation) | 76.5% 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. 48 of them aren't in the launch post.