| Benchmark | DeepSeek-R1 | Kimi K3 |
|---|---|---|
| AA Agentic Index | 3.1 | 54.3 |
| AA Intelligence | 18.6 | 60 |
| AA-LCR | 56 | 82.7 |
| AA-Omniscience | -31.3 | 46 |
| arc_agi_1 | 21.2 | 65.7 |
| ARC-AGI-2 | 1.3 | 12.4 |
| Artificial Analysis Coding Index | 24.6 | 76.2 |
| browsecomp | 8.9 | 91.2 |
| critpt | 0.6 | 23.4 |
| gdpval | 1.5 | 59.4 |
| GPQA Diamond | 81 | 93.5 |
| HLE | 17.7 | 56 |
| OmniScience Accuracy | 30.7 | 47.6 |
| OmniScience Non-Hallucination | 10.5 | 49.1 |
| scicode | 35.7 | 58.7 |
| simplebench | 40.8 | 60.7 |
| simpleqa_verified | 27.4 | 42.7 |
| TauBench V3 - Banking | 6.4 | 46 |
| Terminal-Bench 2.1 | 19.1 | 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.