| Benchmark | Gemini 1.5 Pro | GPT-4o mini |
|---|---|---|
| AA Intelligence | 10 | 7 |
| AI2D | 80.9 | 75.2 |
| aider_polyglot | 16.9 | 3.6 |
| AIR-Bench 2024 | 82.8 | 56.3 |
| anthropic_red_team | 99.9 | 98.3 |
| ARC-AGI-2 | 0.8 | 0 |
| Artificial Analysis Coding Index | 23.6 | 11.4 |
| bbq | 94.5 | 88.2 |
| BLINK | 61 | 51.9 |
| DROP | 74.9 | 79.7 |
| Fortress | 53.9 | 48.1 |
| GPQA Diamond | 59.1 | 42.6 |
| GSM8K | 95.2 | 84.3 |
| harmbench | 79.9 | 84.9 |
| HLE | 4.9 | 4.2 |
| humaneval | 86.6 | 87.2 |
| InterGPS (test) | 58.2 | 39.9 |
| LiveCodeBench | 30.5 | 35.5 |
| MATH | 92 | 80.2 |
| MathVista | 70.6 | 56.7 |
| mgsm | 87.5 | 87 |
| MMBench (dev-en) | 87.9 | 83.8 |
| mmlu | 86.8 | 81.8 |
| MMMU | 68.4 | 59.4 |
| MMMU (val) (Pass@1) | 54.1 | 52.1 |
| MMMU-Pro | 55 | 41.5 |
| narrativeqa | 78.3 | 76.8 |
| naturalquestions_closedbook | 45.5 | 38.5 |
| OpenBookQA | 95.2 | 92 |
| POPE (test) | 89.3 | 83.6 |
| scicode | 29.5 | 22.9 |
| ScienceQA (img-test) | 86 | 84 |
| simple_safety_tests | 97.5 | 97.8 |
| simplebench | 27.1 | 10.7 |
| simpleqa | 23.4 | 9.9 |
| SWE-bench Verified | 34.2 | 8.7 |
| TextVQA (val) | 64.5 | 70.9 |
| Video-MME | 78.6 | 68.9 |
| Video-MME Overall | 62.6 | 61.2 |
| xstest | 98.8 | 96 |
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.