| Benchmark | Claude 3.7 Sonnet | Qwen3 VL 4B (Reasoning) |
|---|---|---|
| AA Agentic Index | 37 | 14.4 |
| AA Intelligence | 27.6 | 7.7 |
| AA-LCR | 62.3 | 23 |
| AA-Omniscience | -0.7 | -68.4 |
| AIME 2025 | 54.8 | 74.5 |
| Artificial Analysis Coding Index | 36.4 | 6.7 |
| critpt | 0.9 | 0 |
| gdpval | 27.4 | 13.8 |
| GPQA Diamond | 84.8 | 64.1 |
| HLE | 9.7 | 4.6 |
| IFBench | 48.3 | 36.6 |
| ifeval | 93.2 | 82.6 |
| MMMU-Pro | 60.1 | 57 |
| OmniScience Accuracy | 28.2 | 12.2 |
| OmniScience Non-Hallucination | 60 | 8.2 |
| scicode | 40.3 | 17.1 |
| Terminal-Bench Hard | 21.2 | 1.5 |
| τ²-Bench Telecom (AA run) | 54.7 | 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.