- SWE-bench Verified 88.6
- Agentic coding SWE-Bench Pro 69.2
- SWE-bench Multilingual 84.4
- SWE-bench Multimodal 38.4
- BrowseComp (multi-agent) 88.5
- Agentic terminal coding Terminal-Bench 2.1 74.6
- Multidisciplinary reasoning Humanity's Last Exam (with tools) 57.9
- ChartQAPro (with tools) 72.3
- Agentic computer use OSWorld-Verified 83.4
- GPQA Diamond 93.6
- ScreenSpot-Pro (with tools) 87.9
- Agentic financial analysis Finance Agent v2 53.9
- Knowledge work GDPval-AA 1890
- MCP-Atlas 82.2
- Automation Bench 15.5
- Best of GraphWalks BFS 256K / GraphWalks Parents 256K 99.3
- Online-Mind2Web, 84% 84
Claude Opus 4.8
Claude Opus 4.8 ranks first on 24 of the 367 benchmarks with five or more models scored.
386
Benchmarks with a score
24
Ranked first, out of five or more models
69
In the top three
Anthropic May 2026
Claude Opus 4.8
Closed weights Flagship
159.9
#21 of 256
Epoch Capabilities Index
Self-Reported Index 159.1 #37 of 404
- DeepSWE 59% #15 /28
- AutomationBench 16.9% #22 /42
- Agents' Last Exam 27% #11 /31
Where it ranks
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.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BullshitBench V2 | 95% | #1 of 129 | Official leaderboard | |
| BullshitBench | 96.4% | #1 of 107 | Official leaderboard | |
| EQ-Bench 3 | 2,058 | #2 of 78 | Official leaderboard | |
| INCLUDE | 87.6% | #2 of 31 | Lab model card | |
| LMArena (Elo) | 1,482 | #25 of 335 | Official leaderboard | |
| Humanity's Last Exam | 57.9% | #17 of 154 | Lab model card | |
| GraphWalks | 99.3% | #3 of 27 | Lab model card | |
| Creative Writing v3 | 79.7 | #14 of 120 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 80.8 | #14 of 119 | Official leaderboard | |
| FACTS Search | 80.3% | #4 of 31 | Official leaderboard | |
| SEA-HELM | 75.6 | #11 of 71 | Official leaderboard | |
| DesignArena ASCII Art | 1,271 | #12 of 77 | Official leaderboard | |
| Global MMLU Lite | 94.8% | #7 of 43 | Official leaderboard | |
| Document Arena | 1,485 | #7 of 39 | Official leaderboard | |
| OpenAI MRCR v2 | 61.8% | #10 of 47 | Official leaderboard | |
| Extended NYT Connections | 91.1% | #22 of 103 | Official leaderboard | |
| Judgemark v4 | 78.0 | #9 of 40 | Official leaderboard | |
| HLE Extended | 49.4% | #8 of 31 | Official leaderboard | |
| EQ-Bench 4 | 1,268 | #9 of 32 | Official leaderboard | |
| Lech Mazur Creative Writing | 0.72 | #17 of 50 | Official leaderboard | |
| SimpleQA Verified | 44.5% | #20 of 56 | Official leaderboard | |
| Hemingway-bench | 1,044 | #16 of 42 | Official leaderboard | |
| Global MMLU | 90.4% | #6 of 14 | Lab model card | |
| Antidote | 1,030 | #16 of 36 | Official leaderboard | |
| AA-LCR Extended | 63.2% | #24 of 49 | Official leaderboard | |
| False-premise questions (Anthropic) | 77% | #3 of 6 | Lab model card | |
| HUMAINE | 28 | #26 of 50 | Official leaderboard | |
| MILU | 90.3% | #6 of 11 | Lab model card | |
| CompassArena | 995 | #6 of 10 | Official leaderboard | |
| FACTS Search V2 | 79.6% | #13 of 21 | Official leaderboard | |
| FACTS Parametric | 25.1% | #30 of 45 | Official leaderboard | |
| 100Q-Hard (Anthropic) | 20% | #6 of 7 | Lab model card |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MathArena (Apex, IMO, USAMO) | 81.3% | #1 of 47 | Official leaderboard | |
| xbench-ScienceQA | 76.4% | #1 of 29 | Official leaderboard | |
| ForecastBench | 0.12 | #4 of 82 | Official leaderboard | |
| FrontierMath | 47.2% | #3 of 55 | Official leaderboard | |
| AIME 2026 | 100% | #2 of 31 | Official leaderboard | |
| ChemBench | 67.3% | #2 of 24 | Official leaderboard | |
| ProofBench | 69% | #6 of 59 | Official leaderboard | |
| MathArena Apex Shortlist | 90.4% | #4 of 38 | Official leaderboard | |
| FrontierMath Tier 4 | 31.3% | #6 of 55 | Epoch AI run | |
| Into the Unknown | 0% | #1 of 9 | Official leaderboard | |
| DTBench | 94.9% | #25 of 180 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 98.3% | #30 of 212 | Epoch AI run | |
| GPQA Diamond | 91% | #35 of 229 | Epoch AI run | |
| Surface Evolver Bench | 87.5% | #5 of 25 | Official leaderboard | |
| Organic Chemistry (Anthropic) | 86.2% | #1 of 5 | Lab model card | |
| ResearchClawBench | 22.8 | #5 of 23 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 80% | #19 of 86 | Official leaderboard | |
| HMMT 2026 | 95.5% | #7 of 31 | Official leaderboard | |
| BTF-3 | 0.13 | #2 of 8 | Official leaderboard | |
| ARC-AGI-2 | 72.1% | #27 of 98 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 56.1% | #19 of 68 | Official leaderboard | |
| SimpleBench | 64.8% | #30 of 106 | Official leaderboard | |
| SpatialBench (LatchBio) | 55.4% | #4 of 14 | Official leaderboard | |
| ARC-AGI | 92.5% | #29 of 95 | Official leaderboard | |
| BioPipelineBench (Anthropic) | 87.7% | #1 of 3 | Lab model card | |
| LatchBio Capabilities v1.0 | 45.3% | #11 of 27 | Official leaderboard | |
| Verifying the Verifiers (Project APE CRED benchmark) | 85% | #10 of 23 | Official leaderboard | |
| ASI-Bench | 37.5 | #5 of 11 | Official leaderboard | |
| BixBench3 | 46.0 | #10 of 20 | Official leaderboard | |
| USAMO 2026 | 96.7% | #3 of 6 | Lab model card | |
| Structural Biology (Anthropic) | 79% | #1 of 2 | Lab model card | |
| scBench (LatchBio SingleCellBench) | 57.1% | #14 of 26 | Official leaderboard | |
| Terminal-Bench-Science | 10.5% | #10 of 17 | Official leaderboard | |
| ArXivMath | 71.8% | #6 of 9 | Lab model card | |
| EurekaBench | 18.4 | #5 of 7 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 88.6% | #4 of 155 | Lab model card | |
| OpenHands Index | 71.9% | #2 of 33 | Official leaderboard | |
| EvoCode-Bench v2 | 59.1% | #1 of 13 | Official leaderboard | |
| WeirdML | 82.9% | #9 of 112 | Official leaderboard | |
| FrontierCode | 34.3% | #1 of 12 | Official leaderboard | |
| PostTrainBench | 34.1% | #2 of 23 | Official leaderboard | |
| SWE-Bench Pro | 69.2% | #8 of 81 | Lab model card | |
| SWE-bench Multilingual | 84.4% | #8 of 73 | Lab model card | |
| ALE-Bench | 2,102 | #18 of 126 | Official leaderboard | |
| GBAEval | 70.9% | #3 of 21 | Official leaderboard | |
| FrontierCS | 74.5 | #2 of 14 | Official leaderboard | |
| GSO-Bench | 47.1% | #5 of 30 | Official leaderboard | |
| Monthly-SWEBench | 69% | #2 of 12 | Official leaderboard | |
| SWE-Marathon | 26% | #2 of 12 | Official leaderboard | |
| HarnessDev | 67.8% | #1 of 6 | Official leaderboard | |
| SWE Atlas | 57.3% | #4 of 22 | Official leaderboard | |
| Code Migration | 47.3% | #16 of 78 | Official leaderboard | |
| MirrorCode | 65.6% | #8 of 39 | Epoch AI run | |
| Vibe Code Bench | 82.7% | #26 of 120 | Official leaderboard | |
| DesignArena Android (Kotlin) | 1,221 | #10 of 46 | Official leaderboard | |
| DesignArena Agentic Game Dev | 1,184 | #6 of 27 | Official leaderboard | |
| DesignArena Full Stack | 1,227 | #12 of 49 | Official leaderboard | |
| TasteVal | 84.2× | #5 of 20 | Official leaderboard | |
| FrontierCode 1.1 | 46.5% | #2 of 8 | Official leaderboard | |
| SWE Atlas Test Writing | 49.6% | #6 of 22 | Official leaderboard | |
| WebDev Arena | 1,559 | #29 of 105 | arena.ai, via Epoch AI | |
| ProgramBench tests passed | 70.9% | #7 of 25 | Official leaderboard | |
| Vision2Web S3-2026 | 71.4% | #2 of 7 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 24% | #11 of 38 | Official leaderboard | |
| FrogsGame post-training task | 3.00% | #5 of 17 | Official leaderboard | |
| SWE Atlas Refactoring | 46.7% | #5 of 17 | Official leaderboard | |
| Senior SWE-Bench | 30.5% | #7 of 23 | Official leaderboard | |
| DesignArena Agentic Overall | 1,218 | #23 of 74 | Official leaderboard | |
| DesignArena Game Dev | 1,252 | #45 of 143 | Official leaderboard | |
| SWE-bench Verified Extended | 63.3% | #14 of 44 | Official leaderboard | |
| ProgramBench almost solved | 16.5% | #8 of 25 | Official leaderboard | |
| DesignArena React Native | 1,210 | #18 of 56 | Official leaderboard | |
| APEX-SWE | 47.3% | #21 of 65 | Official leaderboard | |
| Code Arena | 1,555 | #44 of 128 | Official leaderboard | |
| FrontierSWE v1 | 66.5% | #6 of 17 | Official leaderboard | |
| DesignArena Website | 1,251 | #54 of 151 | Official leaderboard | |
| PostTrainBench 1.1 | 33.8% | #4 of 11 | Official leaderboard | |
| DesignArena Code | 1,244 | #53 of 143 | Official leaderboard | |
| DesignArena Frontend Web App | 1,216 | #20 of 53 | Official leaderboard | |
| DesignArena UI Component | 1,249 | #54 of 139 | Official leaderboard | |
| QwenQoderBench | 62.7% | #2 of 5 | Official leaderboard | |
| QwenSWEBench | 84% | #2 of 5 | Official leaderboard | |
| AndroidBench | 72.4% | #13 of 32 | Official leaderboard | |
| DesignArena Data Visualization | 1,235 | #48 of 118 | Official leaderboard | |
| DesignArena 3D Design | 1,205 | #57 of 136 | Official leaderboard | |
| Data Agent Benchmark | 80.4% | #8 of 18 | Official leaderboard | |
| DesignArena SVG | 1,183 | #47 of 103 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,214 | #11 of 24 | Official leaderboard | |
| MLS-Bench | 42.8% | #7 of 15 | Official leaderboard | |
| GameDevBench | 55.9% | #8 of 17 | Official leaderboard | |
| Vals RSI Index | 0.24 | #13 of 27 | Official leaderboard | |
| DesignArena Image to Frontend | 1,196 | #15 of 31 | Official leaderboard | |
| SWE-bench-Live | 34% | #6 of 12 | Official leaderboard | |
| DeepSWE | 59% | #15 of 28 | Official leaderboard | |
| ReactBench | 32.3% | #15 of 28 | Official leaderboard | |
| SWE-Together | 61.5% | #12 of 22 | Official leaderboard | |
| Vision2Web | 76.1% | #6 of 10 | Official leaderboard | |
| QwenReactBench | 1,694 | #3 of 5 | Official leaderboard | |
| Z.ai Code Bench | 29.5% | #3 of 5 | Official leaderboard | |
| Doc2Harness | 38% | #12 of 19 | Official leaderboard | |
| PostTrainBench 1.2 | 31% | #14 of 21 | Official leaderboard | |
| LoopArena | 33.8% | #4 of 6 | Official leaderboard | |
| QwenBlenderBench | 62.4% | #4 of 6 | Official leaderboard | |
| DesignArena Image to HTML | 1,177 | #36 of 51 | Official leaderboard | |
| SWE-bench Multimodal | 38.4% | #9 of 12 | Lab model card | |
| QwenSVGBench | 1,648 | #4 of 5 | Official leaderboard | |
| ProgramBench | 0% | #21 of 25 | Official leaderboard | |
| FrontierCode 2.0 | 15.3% | #27 of 32 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PinchBench 2.0 | 93.5% | #1 of 57 | Official leaderboard | |
| SnakeBench | 35 | #8 of 188 | Official leaderboard | |
| GAIA | 88.7% | #3 of 64 | Official leaderboard | |
| ScreenSpot-Pro | 87.9% | #2 of 32 | Lab model card | |
| WildClawBench | 64.7% | #2 of 31 | Official leaderboard | |
| ScienceBoard | 68% | #1 of 15 | Official leaderboard | |
| Toolathlon | 59.9% | #2 of 29 | Lab model card | |
| TUA-Bench | 65.8% | #1 of 12 | Official leaderboard | |
| UniClawBench | 47.5% | #1 of 10 | Official leaderboard | |
| Era by Eon | 76.8% | #1 of 9 | Official leaderboard | |
| MyPCBench | 62% | #1 of 9 | Official leaderboard | |
| LisanBench | 9463.0 | #15 of 133 | Official leaderboard | |
| Toolathlon-Verified | 76.2% | #3 of 25 | Official leaderboard | |
| OSWorld-Verified | 83.4% | #5 of 41 | Lab model card | |
| DeepSearchQA | 93.1% | #3 of 23 | Lab model card | |
| OSWorld 2.0 | 20.6% | #1 of 7 | Official leaderboard | |
| BrowseComp | 88.5% | #14 of 81 | Lab model card | |
| Chess Puzzles | 52% | #17 of 93 | Epoch AI run | |
| PACT | 1,562 | #5 of 26 | Official leaderboard | |
| EdgeBench | 51.3 | #1 of 5 | Official leaderboard | |
| CEO-Bench | $2,399,209 | #4 of 19 | Official leaderboard | |
| Harbor-Index | 20.7% | #2 of 9 | Official leaderboard | |
| DeepResearch Bench | 50.2% | #5 of 22 | evals.futuresearch.ai, via Epoch AI | |
| Game Arena Nine Men's Morris | 819 | #9 of 39 | Official leaderboard | |
| Deep Research Bench (FutureSearch) | 0.50 | #5 of 21 | Official leaderboard | |
| Long-Horizon-Terminal-Bench (LHTB) | 49.2% | #6 of 25 | Official leaderboard | |
| Game Arena Dots and Boxes | 1,005 | #10 of 40 | Official leaderboard | |
| OSWorld 2.1 | 44.4% | #1 of 4 | Official leaderboard | |
| BabaIsBench | 96% | #9 of 32 | Official leaderboard | |
| Mystery Game Puzzles | 36% | #24 of 84 | Epoch AI run | |
| CLASHarness | 23.8% | #4 of 14 | Official leaderboard | |
| Game Arena Checkers | 1,081 | #12 of 41 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 421 | #12 of 39 | Official leaderboard | |
| SkillsBench | 54.1% | #7 of 22 | Official leaderboard | |
| τ³-bench Banking | 39.7% | #8 of 25 | Official leaderboard | |
| clembench | 79.1 | #2 of 6 | Official leaderboard | |
| OSWorld 2.0 (Aug 2026) | 37.6% | #1 of 3 | Official leaderboard | |
| LLM Chess | 450 | #42 of 123 | Official leaderboard | |
| MCP Atlas | 82.2% | #13 of 37 | Official leaderboard | |
| Game Arena Go | 1,116 | #14 of 39 | Official leaderboard | |
| Terminal-Bench 2.1 | 78.9% | #5 of 13 | Official leaderboard | |
| Game Arena Four in a Row | 436 | #17 of 43 | Official leaderboard | |
| Online-Mind2Web | 84% | #4 of 10 | Lab model card | |
| BrowseComp Extended | 30% | #17 of 42 | Official leaderboard | |
| RuneBench | 5.08 | #32 of 76 | Official leaderboard | |
| Kaggle Game Arena | 328 | #18 of 42 | Official leaderboard | |
| Game Arena Lines of Action | 405 | #19 of 39 | Official leaderboard | |
| Game Arena Dark Hex | 265 | #20 of 41 | Official leaderboard | |
| Game Arena Clobber | 304 | #20 of 39 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 32.7% | #23 of 44 | Official leaderboard | |
| AutomationBench | 16.9% | #22 of 42 | Official leaderboard | |
| LiveClawBench | 68.8% | #9 of 17 | Official leaderboard | |
| Game Arena Coin Game | 349 | #21 of 39 | Official leaderboard | |
| Terminal-Bench 3.0 | 21.1% | #6 of 11 | Official leaderboard | |
| EBR-bench | 28.6% | #17 of 31 | Epoch AI run | |
| Terminal-Bench 4.0 | 23.6% | #22 of 40 | Official leaderboard | |
| Game Arena Five in a Row | 452 | #11 of 20 | Official leaderboard | |
| Terminal-Bench 4.1 | 23.6% | #24 of 43 | Official leaderboard | |
| DuelLab GameBench 2 | 62.5 | #32 of 57 | Official leaderboard | |
| LMArena Search Arena | 1,194 | #21 of 37 | Official leaderboard | |
| CAR-bench | 53.3% | #7 of 12 | Official leaderboard | |
| Game Arena Word Art | 370 | #23 of 39 | Official leaderboard | |
| ARC-AGI-3 | 1.52% | #16 of 27 | Official leaderboard | |
| Game Arena Reversi | 375 | #24 of 39 | Official leaderboard | |
| Time Horizon Index: KSP | 11.8% | #8 of 13 | Official leaderboard | |
| GameHorizon | 64.8 | #23 of 37 | Official leaderboard | |
| Game Arena Bargaining | 172 | #25 of 39 | Official leaderboard | |
| Game Arena Word Association | 160 | #27 of 40 | Official leaderboard | |
| Game Arena Poker | -6.35 | #21 of 31 | Official leaderboard | |
| RecreationBench | 31.1% | #7 of 10 | Official leaderboard | |
| Game Arena Chess | 621 | #30 of 42 | Official leaderboard | |
| TextQuests | 21.0 | #21 of 28 | Official leaderboard | |
| DRACO | 80.4% | #9 of 12 | Lab model card | |
| MCPMark Verified | 76.4% | #6 of 8 | Official leaderboard | |
| Clocktower Bench | 1,198 | #11 of 11 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MortgageTax | 69.9% | #5 of 87 | Official leaderboard | |
| CHI-Bench | 37.3% | #2 of 23 | Official leaderboard | |
| MedCode | 53.2% | #10 of 96 | Official leaderboard | |
| TaxEval v2 | 75.6% | #14 of 129 | Official leaderboard | |
| GDPval-AA | 1,890 | #2 of 16 | Lab model card | |
| PatientAgentBench | 4.25 | #1 of 8 | Official leaderboard | |
| FORTE | 77.2% | #3 of 21 | Official leaderboard | |
| CorpFin v2 | 66.7% | #18 of 119 | Official leaderboard | |
| CommerceAgentBench | 55.1% | #2 of 13 | Official leaderboard | |
| Vals Index v2 | 60.9% | #11 of 65 | Official leaderboard | |
| ProfBench | 56.8% | #13 of 72 | Official leaderboard | |
| Excel Modeling Benchmark | 69.4% | #15 of 76 | Official leaderboard | |
| SimuVerity | 42.9 | #1 of 5 | Official leaderboard | |
| E-Commerce Bench | 4.98 | #4 of 19 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 9.58% | #20 of 80 | Official leaderboard | |
| GDPevo | 67.1% | #1 of 4 | Official leaderboard | |
| YC-Bench | $1,711,389 | #15 of 59 | Official leaderboard | |
| GDPval Extended | 2.30% | #11 of 43 | Official leaderboard | |
| MedScribe | 85.8% | #25 of 97 | Official leaderboard | |
| Public Benefits Bench | 68.1% | #13 of 50 | Official leaderboard | |
| Legal Research Bench | 43.8% | #27 of 93 | Official leaderboard | |
| BankerToolBench v1.1 | 20.7% | #3 of 10 | Official leaderboard | |
| Vals Index v2.1 | 55.1% | #16 of 49 | Official leaderboard | |
| HW-Bench | 30% | #5 of 15 | Official leaderboard | |
| EnterpriseBench: CoreCraft | 52.3% | #18 of 51 | Official leaderboard | |
| Agents' Last Exam | 27% | #11 of 31 | Official leaderboard | |
| DesignArena HTML Slides | 1,116 | #10 of 28 | Official leaderboard | |
| 3DCodeArena | 1,091 | #5 of 14 | Official leaderboard | |
| Vals Finance Agent v2 | 53.9% | #29 of 80 | Official leaderboard | |
| Vending-Bench Arena | $6,196 | #12 of 32 | Official leaderboard | |
| JobBench | 48.4% | #6 of 16 | Official leaderboard | |
| LegalBench | 83.6% | #53 of 138 | Official leaderboard | |
| Vending-Bench 2 | $5,787 | #27 of 67 | Official leaderboard | |
| TRLawBench | 9.23 | #14 of 34 | Official leaderboard | |
| Remote Labor Index | 8.33% | #8 of 19 | Official leaderboard | |
| Tax Agent Bench | 30% | #36 of 84 | Official leaderboard | |
| OfficeQA | 77.6% | #6 of 14 | Lab model card | |
| DesignArena Python-PPTX Slides | 1,185 | #16 of 36 | Official leaderboard | |
| CADGenBench | 34.5% | #5 of 11 | Official leaderboard | |
| DesignArena Spreadsheet | 1,153 | #16 of 35 | Official leaderboard | |
| prinzbench | 42.0 | #12 of 26 | Official leaderboard | |
| OfficeQA Pro | 48.9% | #6 of 12 | Official leaderboard | |
| MerchantBench | 35.6 | #4 of 8 | Official leaderboard | |
| Kimi Claw 24/7 Bench | 50.4% | #2 of 4 | Official leaderboard | |
| BlenderWorld | 57.0 | #7 of 13 | Official leaderboard | |
| APEX-1 | 62.7% | #40 of 71 | Official leaderboard | |
| EEBench | 51.4% | #17 of 30 | Official leaderboard | |
| APEX-Agents 1.1 | 64.4% | #37 of 65 | Official leaderboard | |
| APEX-Agents | 48.9% | #35 of 61 | Collected by Epoch AI, source not named | |
| TaxCalcBench v2 | 38% | #19 of 32 | Official leaderboard | |
| CoWorkBench | 72.3% | #3 of 5 | Official leaderboard | |
| Tuesday Work Index | 44.5 | #27 of 44 | Official leaderboard | |
| MedXpertQA MM Extended | 51.6% | #26 of 41 | Official leaderboard | |
| DesignArena PDF | 1,156 | #25 of 37 | Official leaderboard | |
| FrontierFinance | 45% | #12 of 17 | Official leaderboard | |
| Management Consulting Tasks (OpenAI) | 31.6% | #5 of 7 | Official leaderboard | |
| OfficeQA Pro V2 | 32.2% | #5 of 7 | Official leaderboard | |
| HealthBench Professional | 55.8% | #22 of 26 | Lab model card | |
| PolicyBench | 75.8% | #43 of 50 | Official leaderboard | |
| LifeSci Diamond | 0% | #16 of 18 | Official leaderboard | |
| VWE-Bench | 24.3% | #6 of 6 | Official leaderboard | |
| RedlineBench | 44.4% | #4 of 4 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vision Arena | 1,286 | #15 of 136 | Official leaderboard | |
| ChartQA Pro | 72.3% | #1 of 8 | Lab model card | |
| ZeroBench | 34% | #15 of 76 | Official leaderboard | |
| ChartMuseum | 75.8% | #1 of 5 | Lab model card | |
| BabyVision | 81.2% | #6 of 27 | Official leaderboard | |
| AgentWorldBench | 56.6 | #3 of 13 | Official leaderboard | |
| MineBench | 1,768 | #21 of 75 | Official leaderboard | |
| GeoBench | 81% | #14 of 48 | Official leaderboard | |
| Drone-Bench | 82.3% | #6 of 19 | Official leaderboard | |
| LongExtractionBench | 70.7% | #1 of 3 | Official leaderboard | |
| DesignArena Code to Video | 1,187 | #13 of 29 | Official leaderboard | |
| CharXiv | 80.5% | #34 of 75 | Lab model card | |
| Math Kangaroo 2025 (MathArena) | 81.6% | #10 of 22 | Official leaderboard | |
| ExtractBench | 86.8% | #13 of 27 | Official leaderboard | |
| Furniture Assembly Benchmark | 42.5% | #17 of 35 | Official leaderboard | |
| QwenVisualOffice | 34.5% | #3 of 6 | Official leaderboard | |
| CC-OCR-Bench-V2 | 65.4% | #8 of 15 | Official leaderboard | |
| ParseBench | 62.3% | #24 of 44 | Official leaderboard | |
| GlobeBench | 20% | #11 of 19 | Official leaderboard | |
| InteractiveBench | 28.3% | #7 of 12 | Official leaderboard | |
| 1H-VideoQA | 71.3% | #12 of 20 | Official leaderboard | |
| VoxelBench | 1,586 | #35 of 56 | Official leaderboard | |
| PerceptionBench (Moonshot) | 47.2% | #10 of 16 | Official leaderboard | |
| MMMU-Pro Extended | 31.4% | #24 of 38 | Official leaderboard | |
| BVB | 62.5 | #33 of 45 | Official leaderboard | |
| Chartography | 15.9% | #32 of 43 | Official leaderboard | |
| CineRecon | 40.3 | #9 of 12 | Official leaderboard | |
| CharXiv Extended | 76.7% | #29 of 38 | Official leaderboard | |
| Blueprint-Bench 2 | 14.5% | #31 of 38 | Official leaderboard | |
| ArtifactArena | 929 | #19 of 21 | Official leaderboard | |
| Humanity's Sixth Sense | 23.5% | #31 of 34 | Official leaderboard | |
| EnigmaEval | 23.5% | #5 of 5 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| World-Class Bio | 66.7% | #2 of 74 | Official leaderboard | |
| BioTIER Refuse | 95.9% | #5 of 69 | Official leaderboard | |
| ABC Bench Liquid-Handling Robot | 100% | #3 of 34 | Official leaderboard | |
| LAB-Bench FigQA | 80.4% | #1 of 11 | Lab model card | |
| nocot-bench (NCRI) | 102.0 | #20 of 185 | Official leaderboard | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 66.3% | #8 of 71 | Official leaderboard | |
| Malicious use of Claude Code (Anthropic) | 95.1% | #2 of 17 | Lab model card | |
| KORA | 65.4% | #7 of 42 | Official leaderboard | |
| Molecular Biology Capabilities Test | 56.4% | #17 of 90 | Official leaderboard | |
| MetagenomicsBench | 50% | #5 of 26 | Official leaderboard | |
| Human Pathogen Capabilities Test | 63.1% | #17 of 86 | Official leaderboard | |
| SusVibes | 32.8% | #2 of 10 | Official leaderboard | |
| Over-eager behavior on GUI computer use tasks (Anthropic) | 9.40% | #1 of 5 | Lab model card | |
| Gray Swan Agent Red Teaming (ART) benchmark | 0.10% | #3 of 14 | Lab model card | |
| Dictatorship Eval | 86.4% | #5 of 20 | Official leaderboard | |
| SpatialBench-Long | 33.3% | #7 of 27 | Official leaderboard | |
| EpiBench | 25.8% | #8 of 27 | Official leaderboard | |
| VariantBench | 42.4% | #8 of 27 | Official leaderboard | |
| SHADE-Arena | 12% | #5 of 16 | Lab model card | |
| MASK | 95.4% | #7 of 19 | Lab model card | |
| BioSecBench-Function | 43.1% | #10 of 27 | Official leaderboard | |
| FORTRESS | 18.2% | #28 of 73 | Official leaderboard | |
| VCT-v2 | 45.9% | #10 of 25 | Official leaderboard | |
| DNA Synthesis Screening Evasion | 70% | #2 of 5 | Lab model card | |
| UK AISI unprompted sabotage evaluation | 0% | #2 of 5 | Lab model card | |
| scBench-Long | 21% | #11 of 27 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 57.7% | #13 of 30 | Official leaderboard | |
| TxBench-Oligo-Discovery | 37.5% | #14 of 30 | Official leaderboard | |
| GeneBench Pro | 16% | #6 of 12 | Official leaderboard | |
| BioSecBench-Surveillance | 46.9% | #14 of 27 | Official leaderboard | |
| DistressBench | 71.4% | #22 of 42 | Official leaderboard | |
| OncologyBench | 16% | #16 of 30 | Official leaderboard | |
| DrugDiscoveryBench | 46.8% | #11 of 20 | Official leaderboard | |
| CyberGym | 78.8% | #19 of 34 | Lab model card | |
| Single-turn harmful request evaluations (Anthropic) | 98% | #11 of 19 | Lab model card | |
| ExploitGym | 13.8% | #11 of 18 | Official leaderboard | |
| HackBench | 52.5 | #5 of 8 | Official leaderboard | |
| TxBench-Antibody-Discovery | 35.3% | #19 of 30 | Official leaderboard | |
| Anthropic political even-handedness eval | 96.7% | #9 of 14 | Lab model card | |
| Adherence to its constitution (Anthropic) | 0.99 | #2 of 3 | Lab model card | |
| Protocols (Anthropic life sciences) | 59.6% | #2 of 3 | Lab model card | |
| Uncritically reporting flawed results (Anthropic) | 9.40 | #2 of 3 | Lab model card | |
| UK AISI sabotage continuation evaluation | 1.00% | #3 of 4 | Lab model card | |
| CyberBench (Vals) | 50.9% | #20 of 26 | Official leaderboard | |
| Virology Capabilities Test (VCT) | 47% | #24 of 31 | Lab model card | |
| Prompt injection robustness against adaptive attackers, coding (Anthropic) | 7.03% | #11 of 14 | Lab model card | |
| SecIT Bench | 67.1% | #21 of 26 | Official leaderboard | |
| BBQ (Bias Benchmark for QA) | 72.1% | #21 of 25 | Lab model card | |
| Single-turn benign request evaluations (Anthropic) | 0.36% | #16 of 19 | Lab model card | |
| PCT (Polarized Contrastive Pairs) | 53.8% | #39 of 46 | Official leaderboard | |
| Malicious computer use (Anthropic) | 81.7% | #12 of 14 | Lab model card | |
| Prompt injection robustness against adaptive attackers, computer use (Anthropic) | 7.14% | #12 of 14 | Lab model card | |
| BioTIER Permit | 84.4% | #61 of 69 | Official leaderboard | |
| DeepsecBench | 6.86 | #26 of 29 | Official leaderboard | |
| ABC Bench Fragment Design | 19.8% | #27 of 27 | Official leaderboard | |
| Agentic BAIM-LLM Evaluation (ABLE) | 0% | #18 of 18 | Official leaderboard | |
| Prompt injection robustness against adaptive attackers, browser use (Anthropic) | 31.5% | #13 of 13 | Lab model card | |
| Binary Exploitation Benchmark (Anthropic, OSS-Fuzz based) | 38.5% | #6 of 6 | Lab model card | |
| Firefox 147 exploit evaluation (Anthropic) | 8.80% | #5 of 5 | Lab model card | |
| Minimal-LinuxBench (Anthropic) | 4.70% | #3 of 3 | Lab model card | |
| Prompt injection live bug bounty (Anthropic with Gray Swan) | 0.40% | #3 of 3 | Lab model card | |
| Lying about identity (Anthropic) | 97% | #2 of 2 | Lab model card | |
| Overconfidence: verifying command-line syntax (Anthropic) | 4.51 | #2 of 2 | Lab model card | |
| AAV capsid packaging prediction | 0.80 | #1 of 1 | Lab model card | |
| Code summary honesty (Anthropic) | 3.70% | #1 of 1 | Lab model card | |
| Lazy investigation on misleading codebases (Anthropic) | 0% | #1 of 1 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SAGE (Vals) | 54.8% | #3 of 81 | Official leaderboard | |
| Conceptual Reasoning Index | 68.3 | #13 of 151 | Official leaderboard | |
| StudentBench | 16.5 | #1 of 11 | Official leaderboard | |
| LMCA | 57.5% | #14 of 147 | Official leaderboard | |
| ACCoRD | 77.4 | #21 of 152 | Official leaderboard | |
| Vals Multimodal Index | 70.9% | #5 of 33 | Official leaderboard | |
| CAIS Risk Index | 38.1 | #13 of 67 | Official leaderboard | |
| CAIS Text Capabilities Index | 41.2 | #19 of 68 | Official leaderboard | |
| CAIS Vision Capabilities Index | 53.0 | #25 of 53 | Official leaderboard | |
| CompassRank | 75.1 | #10 of 14 | Official leaderboard | |
| Anthropic ECI (AECI) | 155.5 | #5 of 6 | Lab model card |
Its lab's numbers put it at 159.1, in line with its ECI
The Self-Reported Index fits one score to the 28 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 157.0 to 161.2. How we compute it.
- GraphWalks 99.3
- GPQA Diamond 93.6
- DeepSearchQA 93.1
- Global MMLU 90.4
- MILU 90.3
- SWE-bench Verified 88.6
- BrowseComp 88.5
- ScreenSpot-Pro 87.9
- INCLUDE 87.6
- SWE-bench Multilingual 84.4
- OSWorld-Verified 83.4
- MCP Atlas 82.2
- CharXiv 80.5
- DRACO 80.4
- ProgramBench tests passed 79
- OfficeQA 77.6
- Terminal-Bench 2.1 74.6
- ChartQA Pro 72.3
- ArXivMath 71.8
- SWE-Bench Pro 69.2
- OfficeQA Pro 66.2
- Toolathlon 59.9
- Humanity's Last Exam 57.9
- HealthBench Professional 55.8
- Vals Finance Agent v2 53.9
- SWE-bench Multimodal 38.4
- AutomationBench 15.5
- Harvey LAB (Legal Agent Benchmark) 9.62
Its lab's numbers next to 11 leaderboards
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 |
|---|---|---|---|---|
| Vending-Bench 2 | $2,992 Vending-Bench 2 final balance at max effort ($5,787.43 at high effort), run by Andon Labs (8.13.5) | $5,787 Leaderboard, high | −$2,795 | The lab reports max reasoning effort, while the board selects high reasoning effort. |
| OfficeQA Pro | 66.2% OfficeQA Pro, 133 questions, internal agentic harness, exact match (8.13.1) | 48.9% Leaderboard, Claude Agent SDK | +17 pts | Anthropic uses an internal agentic harness, while Databricks evaluates Opus 4.8 through Claude Agent SDK. |
| ProgramBench tests passed | 79% ProgramBench hidden test pass rate on 166 of 200 tasks, one episode of up to 1M tokens, near-final snapshot. Text gives 79 to 88% across 1 to 5 episodes (8.5, Figure 8.5.A) | 70.9% Leaderboard, xhigh, mini-SWE-agent | +8.1 pts | |
| Terminal-Bench 2.1 | 74.6% Agentic terminal coding Terminal-Bench 2.1 | 78.9% Leaderboard, high, Claude Code | −4.3 pts | |
| GPQA Diamond | 93.6% GPQA Diamond | 91% Epoch AI run, max | +2.6 pts | |
| SpatialBench (LatchBio) | 53.3% LatchBio Bioinformatics, SpatialBench, no extended thinking (8.16.3) | 55.4% Leaderboard, claude-code | −2.1 pts | |
| AutomationBench | 15.5% Automation Bench | 16.9% Leaderboard, xhigh | −1.4 pts | |
| scBench (LatchBio SingleCellBench) | 58.2% LatchBio Bioinformatics, SingleCellBench, no extended thinking (8.16.3) | 57.1% Leaderboard, published v1, Pi | +1.1 pts | |
| Harvey LAB (Legal Agent Benchmark) | 9.62% Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, internal reimplementation, max effort, five trials, mean criterion pass rate 89.01% (8.13.3) | 9.58% Leaderboard | Same | |
| Vals Finance Agent v2 | 53.9% Agentic financial analysis Finance Agent v2 | 53.9% Leaderboard | Same | |
| MCP Atlas | 82.2% MCP-Atlas | 82.2% Leaderboard, max | Same |
Anthropic's models over time
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)
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it. 81 of them aren't in the launch post.
System Card: Claude Opus 4.8 May 28, 2026
Capabilities
- SWE-bench Verified 88.6
- SWE-Bench Pro 69.2
- SWE-bench Multilingual 84.4
- SWE-bench Multimodal 38.4
- BrowseComp 84.3
- Terminal-Bench 2.1 74.6
- Humanity's Last Exam 49.8
- ChartQA Pro 69.4
- OSWorld-Verified 83.4
- GPQA Diamond 93.6
- ScreenSpot-Pro 82.3
- Vals Finance Agent v2 53.9
- GDPval-AA 1890
- MCP Atlas 82.2
- AutomationBench 15.5
- GraphWalks
- FrontierSWE v1
- ProgramBench tests passed 79
- USAMO 2026 96.7
- ArXivMath 71.8
- DeepSearchQA 93.1
- DRACO 80.4
- ChartMuseum 75.8
- LAB-Bench FigQA 80.4
- CharXiv 80.5
- OfficeQA 77.6
- OfficeQA Pro 66.2
- Harvey LAB (Legal Agent Benchmark) 9.62
- Vending-Bench 2 2992.3
- Toolathlon 59.9
- Global MMLU 90.4
- MILU 90.3
- INCLUDE 87.6
- BioPipelineBench (Anthropic) 87.7
- BioMysteryBench
- SpatialBench (LatchBio) 53.3
- scBench (LatchBio SingleCellBench) 58.2
- Structural Biology (Anthropic) 79
- ProteinGym 39.6
- Organic Chemistry (Anthropic) 86.2
- Protocols (Anthropic life sciences) 59.6
- LABBench2
Health
Hallucination
Safety
- Single-turn harmful request evaluations (Anthropic) 98
- Single-turn benign request evaluations (Anthropic) 0.36
- Multi-turn testing (Anthropic)
- Child safety evaluations (Anthropic)
- Suicide and self-harm evaluations (Anthropic)
- Disordered eating evaluations (Anthropic)
- Election integrity evaluations (Anthropic)
- Malicious use of Claude Code (Anthropic) 95.1
- Malicious computer use (Anthropic) 81.7
Robustness
- Gray Swan Agent Red Teaming (ART) benchmark 0.1
- Prompt injection live bug bounty (Anthropic with Gray Swan) 0.4
- Prompt injection robustness against adaptive attackers, coding (Anthropic) 7.03
- Prompt injection robustness against adaptive attackers, computer use (Anthropic) 7.14
- Prompt injection robustness against adaptive attackers, browser use (Anthropic) 31.5
Alignment
- Anthropic automated behavioral audit
- Petri
- Evaluation awareness discriminability (Anthropic)
- UK AISI unprompted sabotage evaluation 0
- UK AISI sabotage continuation evaluation 1
- Over-eager behavior on GUI computer use tasks (Anthropic) 9.4
- Adherence to its constitution (Anthropic) 0.99
- MASK 95.4
- Lying about identity (Anthropic) 97
- Refusal to assist with AI safety R&D (Anthropic)
- Self-preference bias (Anthropic)
- Uncritically reporting flawed results (Anthropic) 9.4
- Code summary honesty (Anthropic) 3.7
- Lazy investigation on misleading codebases (Anthropic) 0
- Overconfidence: verifying command-line syntax (Anthropic) 4.51
Monitorability
Bio and chem risk
Cyber risk
AI R&D risk
- Anthropic ECI (AECI) 155.5