| Benchmark | Qwen3.6 35B A3B | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 21.6 | 14.4 |
| AA Intelligence | 32.1 | 7.7 |
| AA-LCR | 66.7 | 23 |
| AA-Omniscience | -22.2 | -68.4 |
| AI2D | 92.7 | 84.9 |
| Artificial Analysis Coding Index | 41.9 | 6.7 |
| critpt | 0.3 | 0 |
| EmbSpatial-Bench | 0.8 | 80.7 |
| gdpval | 27.8 | 13.8 |
| GPQA Diamond | 86 | 64.1 |
| HallusionBench | 69.8 | 64.1 |
| HLE | 22.2 | 4.6 |
| HMMT 2025 | 89.1 | 53.1 |
| IFBench | 64.4 | 36.6 |
| LiveCodeBench v6 | 80.4 | 51.3 |
| LVBench | 71.4 | 53.5 |
| MathVista | 86.4 | 79.5 |
| mmlu_redux | 93.3 | 86 |
| MMLU-Pro | 85.6 | 73.6 |
| MMMU-Pro | 75.3 | 57 |
| MV-Bench | 74.6 | 69.3 |
| OmniScience Accuracy | 18.8 | 12.2 |
| OmniScience Non-Hallucination | 49.5 | 8.2 |
| RealWorldQA | 85.3 | 73.2 |
| RefSpatial-Bench | 0.6 | 45.3 |
| scicode | 35.8 | 17.1 |
| supergpqa | 64.7 | 46.8 |
| Terminal-Bench Hard | 34.8 | 1.5 |
| VideoMMMU | 83.7 | 69.4 |
| τ²-Bench Telecom (AA run) | 95.3 | 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.