| Benchmark | Qwen3.6 27B | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 27.5 | 14.4 |
| AA Intelligence | 37.7 | 7.7 |
| AA-LCR | 73.3 | 23 |
| AA-Omniscience | -20 | -68.4 |
| Artificial Analysis Coding Index | 53.7 | 6.7 |
| critpt | 1.1 | 0 |
| EmbSpatial-Bench | 84.6 | 80.7 |
| ERQA | 62.5 | 47.3 |
| gdpval | 32 | 13.8 |
| GPQA Diamond | 87.8 | 64.1 |
| HLE | 24 | 4.6 |
| HMMT 2025 | 90.7 | 53.1 |
| IFBench | 69.1 | 36.6 |
| LiveCodeBench v6 | 83.9 | 51.3 |
| mathvision | 85.1 | 60 |
| MathVista | 87.4 | 79.5 |
| mmlu_redux | 93.5 | 86 |
| MMLU-Pro | 86.2 | 73.6 |
| MMMU-Pro | 75.8 | 57 |
| MMStar | 81.4 | 73.2 |
| MV-Bench | 75.5 | 69.3 |
| OCRBench | 89.4 | 81.6 |
| OmniScience Accuracy | 19.6 | 12.2 |
| OmniScience Non-Hallucination | 50.7 | 8.2 |
| OSWorld-Verified | 63.9 | 31.4 |
| RealWorldQA | 84.1 | 73.2 |
| RefSpatial-Bench | 70 | 45.3 |
| scicode | 39.8 | 17.1 |
| supergpqa | 66 | 46.8 |
| Terminal-Bench Hard | 34.8 | 1.5 |
| VideoMMMU | 84.4 | 69.4 |
| τ²-Bench Telecom (AA run) | 94.2 | 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.