| Benchmark | Claude 4 Opus | o1 |
|---|---|---|
| AA Intelligence | 31.7 | 24 |
| AA-LCR | 40 | 63.3 |
| aider_polyglot | 70.7 | 61.7 |
| AIME 2024 | 76 | 79.2 |
| AIME 2025 | 75.5 | 81.7 |
| AIR-Bench 2024 | 85.7 | 80 |
| anthropic_red_team | 98.9 | 98.3 |
| arena_vision | 1207 | 1193 |
| Artificial Analysis Coding Index | 34 | 39.7 |
| bbq | 99.3 | 97.3 |
| Fortress | 27.6 | 19.4 |
| GPQA Diamond | 79.6 | 75.7 |
| harmbench | 91.7 | 96.3 |
| HLE | 12.3 | 7.7 |
| IFBench | 53.7 | 70.3 |
| livebench | 74.6 | 52.3 |
| math_500_em | 98.2 | 96.4 |
| metr_hcast | 61.7 | 42.6 |
| metr_re_bench | 20 | 0 |
| metr_swaa | 99.8 | 100 |
| mmlu | 92.9 | 91.8 |
| mmmlu | 88.8 | 87.7 |
| MMMU | 73.7 | 77.6 |
| scicode | 40.9 | 35.8 |
| SEAL VISTA | 47 | 45.3 |
| simple_safety_tests | 100 | 99 |
| simplebench | 58.8 | 41.7 |
| simpleqa | 22.8 | 47 |
| SWE-bench Verified | 79.4 | 48.9 |
| TAU-bench (airline) | 59.6 | 50 |
| TAU-bench (retail) | 81.4 | 70.8 |
| Terminal-Bench Hard | 31.1 | 12.9 |
| xstest | 97 | 97 |
| τ²-Bench Telecom (AA run) | 73.4 | 62.6 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.