| Benchmark | Gemini 2.5 Pro | Kimi K3 |
|---|---|---|
| AA Agentic Index | 7.2 | 54.3 |
| AA Intelligence | 27 | 60 |
| AA-LCR | 66 | 82.7 |
| AA-Omniscience | -14.3 | 46 |
| arc_agi_1 | 41 | 65.7 |
| ARC-AGI-2 | 4.9 | 12.4 |
| arena_elo | 1446 | 1489 |
| arena_text_factuality | 1462 | 1471 |
| Artificial Analysis Coding Index | 46.7 | 76.2 |
| browsecomp | 9.9 | 91.2 |
| coding_arena_elo | 1224 | 1674 |
| critpt | 2.6 | 23.4 |
| gdpval | 8.5 | 59.4 |
| GPQA Diamond | 84.4 | 93.5 |
| HLE | 22.5 | 56 |
| matharena_visual_math_overall | 77.2 | 91.3 |
| mathvision | 73.3 | 97.8 |
| MMMU-Pro | 74.9 | 83.4 |
| OmniScience Accuracy | 39 | 47.6 |
| OmniScience Non-Hallucination | 12.6 | 49.1 |
| scicode | 43 | 58.7 |
| simplebench | 62.4 | 60.7 |
| simpleqa_verified | 56 | 42.7 |
| TauBench V3 - Banking | 9.7 | 46 |
| Terminal-Bench 2.1 | 28.5 | 88.3 |
| Video-MME | 84.8 | 90 |
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.