| Benchmark | Gemini 1.5 Pro | Gemini 3 Pro |
|---|---|---|
| AA Intelligence | 10 | 40.6 |
| ARC-AGI-2 | 0.8 | 54 |
| arena_vision | 1179 | 1290 |
| Artificial Analysis Coding Index | 23.6 | 46.5 |
| Fortress | 53.9 | 41.7 |
| GPQA Diamond | 59.1 | 91.9 |
| HLE | 4.9 | 45.8 |
| LiveCodeBench | 30.5 | 92 |
| MathVista | 70.6 | 89.8 |
| MMLU-Pro | 75.8 | 90.1 |
| MMMU-Pro | 55 | 81 |
| mrcr | 82.6 | 89.7 |
| OCRBench | 800 | 90.3 |
| ocrbench_v2 | 51.6 | 63.4 |
| scicode | 29.5 | 56.1 |
| SEAL VISTA | 37.1 | 51.5 |
| simplebench | 27.1 | 76.4 |
| simpleqa | 23.4 | 72.1 |
| SWE-bench Verified | 34.2 | 78 |
| Video-MME | 78.6 | 88.4 |
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.