| Benchmark | Kimi K3 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 54.3 | 14.4 |
| AA Intelligence | 60 | 7.7 |
| AA-LCR | 82.7 | 23 |
| AA-Omniscience | 46 | -68.4 |
| Artificial Analysis Coding Index | 76.2 | 6.7 |
| critpt | 23.4 | 0 |
| gdpval | 59.4 | 13.8 |
| GPQA Diamond | 93.5 | 64.1 |
| HLE | 56 | 4.6 |
| mathvision | 97.8 | 60 |
| MMMU-Pro | 83.4 | 57 |
| OmniScience Accuracy | 47.6 | 12.2 |
| OmniScience Non-Hallucination | 49.1 | 8.2 |
| OSWorld-Verified | 84.8 | 31.4 |
| scicode | 58.7 | 17.1 |
| Video-MME | 90 | 59.7 |
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.