| Benchmark | Claude Sonnet 4.6 | Qwen3.6 35B A3B |
|---|---|---|
| AA Agentic Index | 61.6 | 21.6 |
| AA Intelligence | 48.4 | 32.1 |
| AA-LCR | 74 | 66.7 |
| AA-Omniscience | 12.2 | -22.2 |
| Artificial Analysis Coding Index | 63 | 41.9 |
| browsecomp | 76.2 | 48.7 |
| charxiv_rq | 72.4 | 78 |
| critpt | 3.1 | 0.3 |
| gdpval | 54.8 | 27.8 |
| GDPval-AA v2 | 1381 | 1015 |
| GPQA (unspecified) | 89.9 | 86 |
| GPQA Diamond | 89.9 | 86 |
| HLE | 49 | 22.2 |
| IFBench | 56.6 | 64.4 |
| MCP Atlas | 69.5 | 62.8 |
| MMMU-Pro | 75.6 | 75.3 |
| OmniScience Accuracy | 40.9 | 18.8 |
| OmniScience Non-Hallucination | 51.6 | 49.5 |
| scicode | 47 | 35.8 |
| SWE-bench Pro | 58.1 | 49.5 |
| SWE-bench Verified | 79.6 | 73.4 |
| TauBench V3 - Banking | 34.4 | 9.3 |
| Terminal-Bench 2.0 | 59.1 | 51.5 |
| Terminal-Bench 2.1 | 71.2 | 44.9 |
| Terminal-Bench Hard | 59.1 | 34.8 |
| toolathlon | 49.4 | 26.9 |
| τ²-Bench Telecom (AA run) | 97.9 | 95.3 |
| τ³-Bench | 30.5 | 67.2 |
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.