| Benchmark | Kimi K2.6 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 58.7 | 14.4 |
| AA Intelligence | 45.1 | 7.7 |
| AA-LCR | 76.7 | 23 |
| AA-Omniscience | 6.4 | -68.4 |
| Artificial Analysis Coding Index | 61.8 | 6.7 |
| critpt | 8 | 0 |
| gdpval | 41.4 | 13.8 |
| GPQA Diamond | 91.1 | 64.1 |
| HLE | 37.5 | 4.6 |
| IFBench | 76 | 36.6 |
| LiveCodeBench | 89.6 | 51.3 |
| LiveCodeBench v6 | 89.6 | 51.3 |
| mathvision | 93.2 | 60 |
| MMLU-Pro | 87.1 | 73.6 |
| MMMU-Pro | 80.1 | 57 |
| OmniScience Accuracy | 32.8 | 12.2 |
| OmniScience Non-Hallucination | 60.7 | 8.2 |
| OSWorld-Verified | 73.1 | 31.4 |
| scicode | 53.5 | 17.1 |
| Terminal-Bench Hard | 43.9 | 1.5 |
| τ²-Bench Telecom (AA run) | 95.9 | 15.5 |
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.