| Benchmark | Gemini 1.5 Pro | o1 |
|---|---|---|
| AA Intelligence | 10 | 24 |
| aider_polyglot | 16.9 | 61.7 |
| AIR-Bench 2024 | 82.8 | 80 |
| anthropic_red_team | 99.9 | 98.3 |
| arena_vision | 1179 | 1193 |
| Artificial Analysis Coding Index | 23.6 | 39.7 |
| bbq | 94.5 | 97.3 |
| Fortress | 53.9 | 19.4 |
| GPQA Diamond | 59.1 | 75.7 |
| GSM8K | 95.2 | 97.1 |
| harmbench | 79.9 | 96.3 |
| HLE | 4.9 | 7.7 |
| humaneval | 86.6 | 88.1 |
| MATH | 92 | 96.4 |
| MathVista | 70.6 | 71.8 |
| mgsm | 87.5 | 90.8 |
| mmlu | 86.8 | 91.8 |
| MMMU | 68.4 | 77.6 |
| scicode | 29.5 | 35.8 |
| SEAL VISTA | 37.1 | 45.3 |
| simple_safety_tests | 97.5 | 99 |
| simplebench | 27.1 | 41.7 |
| simpleqa | 23.4 | 47 |
| SWE-bench Verified | 34.2 | 48.9 |
| xstest | 98.8 | 97 |
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.