| Benchmark | Gemini 1.5 Pro | GPT-5 |
|---|---|---|
| AA Intelligence | 10 | 35.3 |
| aider_polyglot | 16.9 | 88 |
| AIR-Bench 2024 | 82.8 | 87.7 |
| ARC-AGI-2 | 0.8 | 7.5 |
| arena_vision | 1179 | 1212 |
| Artificial Analysis Coding Index | 23.6 | 38.9 |
| Fortress | 53.9 | 17 |
| GPQA Diamond | 59.1 | 87.3 |
| GSM8K | 95.2 | -1.8 |
| harmbench | 79.9 | 70 |
| HLE | 4.9 | 28.5 |
| humaneval | 86.6 | 93.4 |
| LiveCodeBench | 30.5 | 85 |
| MMLU-Pro | 75.8 | 87.5 |
| MMMU | 68.4 | 84.2 |
| MMMU-Pro | 55 | 78.4 |
| ocrbench_v2 | 51.6 | 55.5 |
| scicode | 29.5 | 43 |
| SEAL VISTA | 37.1 | 49.7 |
| simplebench | 27.1 | 56.7 |
| SWE-bench Verified | 34.2 | 74.9 |
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.