| Benchmark | Claude 3.7 Sonnet | Gemini 1.5 Pro |
|---|---|---|
| AA Intelligence | 27.6 | 10 |
| aider_polyglot | 64.9 | 16.9 |
| AIR-Bench 2024 | 81.8 | 82.8 |
| anthropic_red_team | 99.7 | 99.9 |
| ARC-AGI-2 | 0 | 0.8 |
| arena_vision | 1195 | 1179 |
| Artificial Analysis Coding Index | 36.4 | 23.6 |
| bbq | 92.1 | 94.5 |
| Fortress | 38 | 53.9 |
| GPQA Diamond | 84.8 | 59.1 |
| harmbench | 84.3 | 79.9 |
| HLE | 9.7 | 4.9 |
| ifeval | 93.2 | 89.4 |
| MMMU | 75 | 68.4 |
| MMMU-Pro | 60.1 | 55 |
| scicode | 40.3 | 29.5 |
| SEAL VISTA | 48.2 | 37.1 |
| simple_safety_tests | 100 | 97.5 |
| simplebench | 46.4 | 27.1 |
| SWE-bench Verified | 70.3 | 34.2 |
| xstest | 96.4 | 98.8 |
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.