| Benchmark | Claude Sonnet 5 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 49.7 | 27.5 |
| AA Intelligence | 55.3 | 37.7 |
| AA-LCR | 77 | 73.3 |
| AA-Omniscience | 16.4 | -20 |
| Artificial Analysis Coding Index | 71.5 | 53.7 |
| charxiv_rq | 77 | 78.4 |
| critpt | 16.9 | 1.1 |
| DeepSWE 1.1 | 54 | 13.3 |
| gdpval | 54.8 | 32 |
| GPQA Diamond | 91.1 | 87.8 |
| HLE | 57.4 | 24 |
| livebench_agentic_coding | 59.4 | 39.3 |
| livebench_coding | 80.7 | 71.8 |
| livebench_data_analysis | 71.7 | 70.4 |
| livebench_instruction_following | 63.9 | 53.2 |
| livebench_language | 75 | 63.3 |
| livebench_math | 92.9 | 79.9 |
| livebench_reasoning | 88.7 | 70.3 |
| MMMU-Pro | 77.3 | 75.8 |
| OmniScience Accuracy | 40 | 19.6 |
| OmniScience Non-Hallucination | 60.6 | 50.7 |
| OSWorld-Verified | 81.2 | 63.9 |
| scicode | 53.6 | 39.8 |
| SWE-bench Multilingual | 78.3 | 71.3 |
| SWE-bench Pro | 63.2 | 53.5 |
| SWE-bench Verified | 85.2 | 77.2 |
| TauBench V3 - Banking | 37.3 | 16.7 |
| Terminal-Bench 2.0 | 80.4 | 59.3 |
| Terminal-Bench 2.1 | 80.5 | 60.7 |
| τ³-Bench | 28.2 | 15.3 |
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.