| Benchmark | Gemini 1.5 Pro | Gemini 2.5 Pro |
|---|---|---|
| AA Intelligence | 10 | 27 |
| aider_polyglot | 16.9 | 82.2 |
| AIR-Bench 2024 | 82.8 | 73.6 |
| anthropic_red_team | 99.9 | 99.5 |
| ARC-AGI-2 | 0.8 | 4.9 |
| arena_vision | 1179 | 1246 |
| Artificial Analysis Coding Index | 23.6 | 46.7 |
| bbq | 94.5 | 96.4 |
| Fortress | 53.9 | 54.9 |
| GPQA Diamond | 59.1 | 84.4 |
| harmbench | 79.9 | 65.4 |
| HLE | 4.9 | 22.5 |
| LiveCodeBench | 30.5 | 82.7 |
| MathVista | 70.6 | 83.9 |
| MMLU-Pro | 75.8 | 86 |
| MMMU | 68.4 | 83.9 |
| MMMU-Pro | 55 | 74.9 |
| mrcr | 82.6 | 93 |
| OCRBench | 800 | 85.9 |
| ocrbench_v2 | 51.6 | 59.3 |
| scicode | 29.5 | 43 |
| SEAL VISTA | 37.1 | 50.8 |
| simple_safety_tests | 97.5 | 97 |
| simplebench | 27.1 | 62.4 |
| simpleqa | 23.4 | 50.8 |
| SWE-bench Verified | 34.2 | 67.2 |
| Video-MME | 78.6 | 84.8 |
| xstest | 98.8 | 98.7 |
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.