| Benchmark | Qwen3.5 397B A17B | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 53.3 | 14.4 |
| AA Intelligence | 34.3 | 7.7 |
| AA-LCR | 72.7 | 23 |
| AA-Omniscience | -30.8 | -68.4 |
| Artificial Analysis Coding Index | 48.2 | 6.7 |
| critpt | 1.7 | 0 |
| ERQA | 67.5 | 47.3 |
| gdpval | 35.8 | 13.8 |
| GPQA Diamond | 89.3 | 64.1 |
| HLE | 29 | 4.6 |
| HMMT 2025 | 92.7 | 53.1 |
| IFBench | 78.8 | 36.6 |
| ifeval | 92.6 | 82.6 |
| include | 85.6 | 64.6 |
| LiveCodeBench v6 | 83.6 | 51.3 |
| mmlu_prox | 84.7 | 65 |
| mmlu_redux | 94.9 | 86 |
| MMLU-Pro | 87.8 | 73.6 |
| MMMU-Pro | 79 | 57 |
| MMStar | 83.8 | 73.2 |
| MV-Bench | 77.6 | 69.3 |
| OmniScience Accuracy | 30.8 | 12.2 |
| OmniScience Non-Hallucination | 17.3 | 8.2 |
| RealWorldQA | 83.9 | 73.2 |
| scicode | 42 | 17.1 |
| supergpqa | 70.4 | 46.8 |
| Terminal-Bench Hard | 40.9 | 1.5 |
| VideoMMMU | 84.7 | 69.4 |
| τ²-Bench Telecom (AA run) | 95.6 | 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.