| Benchmark | Gemma 4 31B | Kimi K3 |
|---|---|---|
| AA Agentic Index | 14.4 | 54.3 |
| AA Intelligence | 29.7 | 60 |
| AA-LCR | 68.3 | 82.7 |
| AA-Omniscience | -47.9 | 46 |
| aime_2026 | 89.2 | 96.7 |
| arena_elo | 1451 | 1489 |
| arena_text_factuality | 1449 | 1471 |
| Artificial Analysis Coding Index | 43.4 | 76.2 |
| charxiv_rq | 67.9 | 84.8 |
| coding_arena_elo | 1364 | 1674 |
| critpt | 1.4 | 23.4 |
| gdpval | 15.5 | 59.4 |
| GPQA Diamond | 85.7 | 93.5 |
| HLE | 26.5 | 56 |
| hmmt_feb_2026 | 77.2 | 97 |
| itbenchSre | 37.3 | 47.7 |
| mathvision | 85.6 | 97.8 |
| MCP Atlas | 57.2 | 84.2 |
| MMMU-Pro | 76.9 | 83.4 |
| OmniScience Accuracy | 20 | 47.6 |
| OmniScience Non-Hallucination | 18.1 | 49.1 |
| scicode | 43.4 | 58.7 |
| simpleqa_verified | 9.6 | 42.7 |
| TauBench V3 - Banking | 14.8 | 46 |
| Terminal-Bench 2.1 | 43.4 | 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.