| Benchmark | Claude Sonnet 4.6 | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 61.6 | 14.4 |
| AA Intelligence | 48.4 | 7.7 |
| AA-LCR | 74 | 23 |
| AA-Omniscience | 12.2 | -68.4 |
| Artificial Analysis Coding Index | 63 | 6.7 |
| critpt | 3.1 | 0 |
| gdpval | 54.8 | 13.8 |
| GPQA Diamond | 89.9 | 64.1 |
| HLE | 49 | 4.6 |
| IFBench | 56.6 | 36.6 |
| MMMU-Pro | 75.6 | 57 |
| OmniScience Accuracy | 40.9 | 12.2 |
| OmniScience Non-Hallucination | 51.6 | 8.2 |
| OSWorld-Verified | 78.5 | 31.4 |
| scicode | 47 | 17.1 |
| Terminal-Bench Hard | 59.1 | 1.5 |
| τ²-Bench Telecom (AA run) | 97.9 | 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.