Anthropic
availableShows if the model has enough results for an index.Claude Opus 5.5
Claude Opus 5.5 is a reasoning model from Anthropic. 28 benchmarks count toward its score, in 7 categories.
IndexOverall score. 50 is the middle.82.8 ±3.6
CoverageShare of the index weight with results.95%
SpeedOutput tokens per second.85/s
Input / 1MUS dollars per 1M input tokens.$4
Output / 1MUS dollars per 1M output tokens.$20
ContextMaximum tokens in one request.1M
EloLMArena rating and rank.N/A
50 is the middle of the board. The range shows the doubt in the index.
CapabilitiesScore per category. 50 is the middle.
50 is the middleResults
28 counted| BenchmarkThe test name. | CategoryThe capability that the test measures. | ResultThe score from the publisher. | IndexThis result on the index scale. | RunThe settings of the run. | DateDate of the result. | Published byThe source of the result. |
|---|---|---|---|---|---|---|
| LiveBench Mathematics | Math | 97.1% | 77.6 | max effort | 25 Jun 2026 | LiveBench |
| ArXivMath August 2026 with tools | Math | 96.9% | — | — | — | MathArena and Anthropic |
| BenchCAD Vision2Code voxel IoU with tools | Multimodal | 96.2% | — | — | — | Zhang et al. and Anthropic |
| Global MMLU | Multilingual | 94.3% | — | — | — | Singh et al. |
| Multi-task Indic Language Understanding Benchmark | Multilingual | 93.1% | — | — | — | Verma et al. |
| LiveBench Reasoning | Reasoning | 92.2% | 84.0 | max effort | 25 Jun 2026 | LiveBench |
| Artificial Analysis Harvey LAB-AA | Agentic | 91.2% | 74.3 | — | — | Artificial Analysis |
| Legal Agent Benchmark mean criterion-pass rate — Harvey held-out set | Agentic | 91.2% | — | — | — | Harvey AI |
| ProgramBench: Can Language Models Rebuild Programs From Scratch? | Coding | 91.2% | 84.6 | — | — | John Yang et al. |
| ArXivMath August 2026 without tools | Math | 91.2% | — | — | — | MathArena and Anthropic |
| SWE-bench Pro | Coding | 89.9% | 91.0 | — | — | Xiang Deng et al. |
| BioMysteryBench Human Solvable | Knowledge | 89.3% | — | — | — | Anthropic |
| LiveBench Coding | Coding | 89.3% | 85.8 | max effort | 25 Jun 2026 | LiveBench |
| Chartography with image and code tools | Multimodal | 89.0% | — | — | — | Surge AI and Anthropic |
| Artificial Analysis MMMU-Pro | Multimodal | 87.7% | 74.3 | — | — | Artificial Analysis |
| LiveBench Language | Knowledge | 86.3% | 74.6 | max effort | 25 Jun 2026 | LiveBench |
| Artificial Analysis Long Context Reasoning | Reasoning | 84.7% | 66.8 | — | — | Artificial Analysis |
| Anthropic de novo protein-binder design evaluation | Knowledge | 82.6% | — | — | — | Anthropic |
| Toolathlon Verified Pass@3 | Agentic | 82.4% | — | — | — | Anthropic |
| LiveBench Data Analysis | Reasoning | 80.3% | 67.5 | max effort | 25 Jun 2026 | LiveBench |
| OfficeQA | Multimodal | 78.9% | — | — | — | Databricks and Anthropic |
| Toolathlon-Verified | Agentic | 77.8% | 75.4 | — | — | Moonshot AI |
| HealthBench Professional raw score | Knowledge | 77.1% | — | — | — | Anthropic |
| DeepSWE | Agentic | 74.2% | 77.7 | — | — | Datacurve AI |
| Molecular Biology Protocols Troubleshooting | Knowledge | 73.7% | — | — | — | Anthropic |
| BenchCAD Vision2Code voxel IoU without tools | Multimodal | 73.0% | — | — | — | Zhang et al. and Anthropic |
| Toolathlon Verified Pass cubed | Agentic | 72.2% | — | — | — | Anthropic |
| LatchBio SpatialBench Verified | Knowledge | 72.0% | — | — | — | LatchBio and Anthropic |
| LiveBench Agentic Coding | Agentic | 71.7% | 84.2 | max effort | 25 Jun 2026 | LiveBench |
| Anthropic biomedical-image-analysis evaluation | Multimodal | 71.4% | — | — | — | Anthropic |
| Artificial Analysis AutomationBench | Agentic | 69.5% | 84.5 | — | — | Artificial Analysis |
| Benchling Molecular Biology Protocols Understanding | Knowledge | 69.0% | — | — | — | Benchling and Anthropic |
| HealthBench raw score | Knowledge | 68.1% | — | — | — | Anthropic |
| Humanity's Last Exam with tools | Agentic | 67.7% | 80.1 | — | — | DeepSeek-AI |
| OfficeQA Pro | Multimodal | 67.7% | 80.0 | — | — | OfficeQA Pro authors |
| GDPval-AA normalized | Agentic | 67.3% | 87.3 | — | — | Artificial Analysis |
| Artificial Analysis SciCode | Coding | 66.9% | 85.7 | — | — | Artificial Analysis |
| Artificial Analysis Omniscience Accuracy | Knowledge | 66.2% | 95.0 | — | — | Artificial Analysis |
| LiveBench Instruction Following | Instruction | 65.7% | 63.8 | max effort | 25 Jun 2026 | LiveBench |
| HealthBench Professional | Knowledge | 65.6% | — | — | — | Rebecca Soskin Hicks et al. |
| Chartography without tools | Multimodal | 64.4% | — | — | — | Surge AI and Anthropic |
| Humanity's Last Exam without tools | Knowledge | 64.4% | 83.3 | — | — | OpenAI |
| FrontierCode 1.1 Extended | Coding | 63.6% | — | — | — | Cognition |
| Anthropic medicinal-chemistry evaluation | Knowledge | 63.5% | — | — | — | Anthropic |
| FrontierSWE v2 | Coding | 62.3% | 88.9 | — | — | Proximal |
| Artificial Analysis Humanity's Last Exam | Knowledge | 61.4% | 91.1 | — | — | Artificial Analysis |
| LatchBio SingleCellBench | Knowledge | 61.2% | — | — | — | LatchBio and Anthropic |
| HealthBench length-adjusted score | Knowledge | 60.6% | — | — | — | Anthropic |
| Anthropic Protein Design evaluation | Knowledge | 60.2% | — | — | — | Anthropic |
| Terminal-Bench-Science 0.1 | Agentic | 58.7% | — | — | — | Terminal-Bench-Science Team |
| cursorBench40 | Coding | 57.8% | — | — | — | Benchmark authors |
| Artificial Analysis Intelligence Index | Knowledge | 57.6% | 94.5 | — | — | Artificial Analysis |
| Anthropic protein-design library-ranking task | Knowledge | 56.0% | — | — | — | Anthropic |
| FrontierCode 1.1 Main | Coding | 54.4% | 82.0 | — | — | Cognition |
| BioMysteryBench Human Difficult | Knowledge | 50.0% | — | — | — | Anthropic |
| OSWorld 2.0 | Agentic | 48.7% | 79.1 | — | — | Mengqi Yuan et al. |
| AutomationBench | Agentic | 40.0% | 82.4 | — | — | Moonshot AI |
| Axiom Bio morphology-to-molecule matching | Knowledge | 34.0% | — | — | — | Axiom Bio and Anthropic |
| Critical Physics Tasks | Reasoning | 31.7% | 95.0 | — | — | Artificial Analysis |
| Toolathlon Verified average assistant turns | Agentic | 26.9% | — | — | — | Anthropic |
| Artificial Analysis GDP.pdf | Agentic | 26.2% | 77.7 | — | — | Artificial Analysis |
| Legal Agent Benchmark all-pass rate — Harvey held-out set | Agentic | 8.3% | — | — | — | Harvey AI |
28 benchmarks count, from 29 of 62 results. A grey row does not count. Too few models took that benchmark.