| Benchmark | Gemini 1.5 Pro | Kimi K2.5 |
|---|---|---|
| AA Intelligence | 10 | 36 |
| ARC-AGI-2 | 0.8 | 11.8 |
| arena_vision | 1179 | 1247 |
| Artificial Analysis Coding Index | 23.6 | 46.8 |
| Fortress | 53.9 | 41.1 |
| GPQA Diamond | 59.1 | 87.9 |
| HLE | 4.9 | 50.2 |
| MathVista | 70.6 | 90.1 |
| MMLU-Pro | 75.8 | 87.1 |
| MMMU-Pro | 55 | 78.5 |
| OCRBench | 800 | 92.3 |
| scicode | 29.5 | 49 |
| SEAL VISTA | 37.1 | 41.9 |
| simplebench | 27.1 | 46.8 |
| SWE-bench Verified | 34.2 | 76.8 |
| Video-MME | 78.6 | 87.4 |
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.