| Benchmark | GPT-5.6 Sol | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 57.8 | 14.4 |
| AA Intelligence | 61 | 7.7 |
| AA-LCR | 77.7 | 23 |
| AA-Omniscience | 22 | -68.4 |
| Artificial Analysis Coding Index | 78.3 | 6.7 |
| critpt | 32.3 | 0 |
| gdpval | 61.1 | 13.8 |
| GPQA Diamond | 94.6 | 64.1 |
| HLE | 49.5 | 4.6 |
| IFBench | 72.7 | 36.6 |
| mathvision | 95.8 | 60 |
| MMMU-Pro | 84.6 | 57 |
| OmniScience Accuracy | 59.4 | 12.2 |
| OmniScience Non-Hallucination | 10.6 | 8.2 |
| OSWorld-Verified | 83 | 31.4 |
| scicode | 56.9 | 17.1 |
| Terminal-Bench Hard | 65.9 | 1.5 |
| Video-MME | 89.5 | 59.7 |
| τ²-Bench Telecom (AA run) | 85.1 | 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.