| Benchmark | Claude Sonnet 4 | Gemini 1.5 Pro |
|---|---|---|
| AA Intelligence | 29.8 | 10 |
| aider_polyglot | 70.7 | 16.9 |
| AIR-Bench 2024 | 88.3 | 82.8 |
| anthropic_red_team | 99.2 | 99.9 |
| ARC-AGI-2 | 13.6 | 0.8 |
| arena_vision | 1207 | 1179 |
| Artificial Analysis Coding Index | 37.6 | 23.6 |
| bbq | 97.9 | 94.5 |
| Fortress | 24.4 | 53.9 |
| GPQA Diamond | 78 | 59.1 |
| harmbench | 98 | 79.9 |
| HLE | 10.7 | 4.9 |
| ifeval | 87.6 | 89.4 |
| LiveCodeBench | 68.5 | 30.5 |
| mmlu | 91.5 | 86.8 |
| MMLU-Pro | 84 | 75.8 |
| MMMU | 74.4 | 68.4 |
| MMMU-Pro | 62.4 | 55 |
| scicode | 40 | 29.5 |
| SEAL VISTA | 45.5 | 37.1 |
| simple_safety_tests | 100 | 97.5 |
| simplebench | 45.5 | 27.1 |
| simpleqa | 15.9 | 23.4 |
| SWE-bench Verified | 80.2 | 34.2 |
| xstest | 97.2 | 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.