| Benchmark | Kimi K3 | Qwen3.5 35B A3B |
|---|---|---|
| AA Agentic Index | 54.3 | 44.1 |
| AA Intelligence | 60 | 29.9 |
| AA-LCR | 82.7 | 68.3 |
| AA-Omniscience | 46 | -48.1 |
| aime_2026 | 96.7 | 93.3 |
| Artificial Analysis Coding Index | 76.2 | 37 |
| baby_vision | 85.7 | 38.4 |
| browsecomp | 91.2 | 61 |
| charxiv_rq | 84.8 | 77.5 |
| coding_arena_elo | 1674 | 1250 |
| critpt | 23.4 | 0.9 |
| gdpval | 59.4 | 20.3 |
| GPQA Diamond | 93.5 | 84.5 |
| HLE | 56 | 47.4 |
| hmmt_feb_2026 | 97 | 81.8 |
| itbenchSre | 47.7 | 21.5 |
| mathvision | 97.8 | 83.9 |
| MCP Atlas | 84.2 | 62.4 |
| MMMU-Pro | 83.4 | 75.1 |
| MMVU | 82.1 | 72.3 |
| OmniScience Accuracy | 47.6 | 20.1 |
| OmniScience Non-Hallucination | 49.1 | 14.6 |
| OSWorld-Verified | 84.8 | 54.5 |
| scicode | 58.7 | 37.7 |
| TauBench V3 - Banking | 46 | 4.9 |
| Terminal-Bench 2.1 | 88.3 | 40.8 |
| ZeroBench | 0.4 | 0.1 |
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.