| Benchmark | Claude Sonnet 4 | Kimi K3 |
|---|---|---|
| AA Agentic Index | 39.2 | 54.3 |
| AA Intelligence | 29.8 | 60 |
| AA-LCR | 69.7 | 82.7 |
| AA-Omniscience | 0.2 | 46 |
| arc_agi_1 | 63.7 | 65.7 |
| ARC-AGI-2 | 13.6 | 12.4 |
| Artificial Analysis Coding Index | 37.6 | 76.2 |
| browsecomp | 12.2 | 91.2 |
| critpt | 1.1 | 23.4 |
| gdpval | 31.2 | 59.4 |
| GPQA Diamond | 78 | 93.5 |
| HLE | 10.7 | 56 |
| MMMU-Pro | 62.4 | 83.4 |
| OmniScience Accuracy | 22.7 | 47.6 |
| OmniScience Non-Hallucination | 70.9 | 49.1 |
| OSWorld-Verified | 42.2 | 84.8 |
| scicode | 40 | 58.7 |
| simplebench | 45.5 | 60.7 |
| TauBench V3 - Banking | 16.7 | 46 |
| Terminal-Bench 2.1 | 36.3 | 88.3 |
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.