| Benchmark | Claude 3.5 Sonnet | Gemini 3 Pro |
|---|---|---|
| AA Intelligence | 10 | 40.6 |
| AIME 2025 | 7.4 | 100 |
| arena_vision | 1146 | 1290 |
| Artificial Analysis Coding Index | 30.2 | 46.5 |
| Fortress | 13 | 41.7 |
| frontiermath_tier_4 | 0 | 18.8 |
| GPQA Diamond | 67.2 | 91.9 |
| HLE | 3.9 | 45.8 |
| hmmt_feb_2025 | 1.7 | 97.5 |
| LiveCodeBench | 32.8 | 92 |
| LiveCodeBench v6 | 37.2 | 90.7 |
| longbench_v2 | 41 | 68.2 |
| MathVista | 67.7 | 89.8 |
| MMLU-Pro | 78 | 90.1 |
| MMMU-Pro | 54.7 | 81 |
| OCRBench | 790 | 90.3 |
| scicode | 36.6 | 56.1 |
| SEAL VISTA | 38.7 | 51.5 |
| simplebench | 41.4 | 76.4 |
| simpleqa | 28.4 | 72.1 |
| SWE-bench Verified | 50.8 | 78 |
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.