| Benchmark | Claude 3.5 Sonnet | Claude 3.7 Sonnet |
|---|---|---|
| AA Intelligence | 10 | 27.6 |
| aider_polyglot | 45.3 | 64.9 |
| AIME 2025 | 7.4 | 54.8 |
| AIR-Bench 2024 | 85.9 | 81.8 |
| anthropic_red_team | 99.8 | 99.7 |
| arena_vision | 1146 | 1195 |
| Artificial Analysis Coding Index | 30.2 | 36.4 |
| bbq | 94.9 | 92.1 |
| Fortress | 13 | 38 |
| GPQA Diamond | 67.2 | 84.8 |
| harmbench | 98.1 | 84.3 |
| HLE | 3.9 | 9.7 |
| hmmt_feb_2025 | 1.7 | 31.7 |
| ifeval | 90.1 | 93.2 |
| math_500_em | 78.3 | 96.2 |
| MMMU | 72 | 75 |
| MMMU-Pro | 54.7 | 60.1 |
| scicode | 36.6 | 40.3 |
| SEAL VISTA | 38.7 | 48.2 |
| simple_safety_tests | 100 | 100 |
| simplebench | 41.4 | 46.4 |
| SWE-bench Verified | 50.8 | 70.3 |
| TAU-bench (airline) | 46 | 58.4 |
| TAU-bench (retail) | 69.2 | 81.2 |
| xstest | 95.6 | 96.4 |
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.