| Benchmark | Claude Sonnet 4.6 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 61.6 | 27.5 |
| AA Intelligence | 48.4 | 37.7 |
| AA-LCR | 74 | 73.3 |
| AA-Omniscience | 12.2 | -20 |
| Artificial Analysis Coding Index | 63 | 53.7 |
| charxiv_rq | 72.4 | 78.4 |
| critpt | 3.1 | 1.1 |
| DeepSWE 1.1 | 30 | 13.3 |
| gdpval | 54.8 | 32 |
| GPQA Diamond | 89.9 | 87.8 |
| HLE | 49 | 24 |
| IFBench | 56.6 | 69.1 |
| livebench_agentic_coding | 42.6 | 39.3 |
| livebench_coding | 79.3 | 71.8 |
| livebench_data_analysis | 78 | 70.4 |
| livebench_instruction_following | 63.2 | 53.2 |
| livebench_language | 76.1 | 63.3 |
| livebench_math | 87 | 79.9 |
| livebench_reasoning | 84.8 | 70.3 |
| MMMU-Pro | 75.6 | 75.8 |
| OmniScience Accuracy | 40.9 | 19.6 |
| OmniScience Non-Hallucination | 51.6 | 50.7 |
| OSWorld-Verified | 78.5 | 63.9 |
| scicode | 47 | 39.8 |
| SWE-bench Pro | 58.1 | 53.5 |
| SWE-bench Verified | 79.6 | 77.2 |
| TauBench V3 - Banking | 34.4 | 16.7 |
| Terminal-Bench 2.0 | 59.1 | 59.3 |
| Terminal-Bench 2.1 | 71.2 | 60.7 |
| Terminal-Bench Hard | 59.1 | 34.8 |
| τ²-Bench Telecom (AA run) | 97.9 | 94.2 |
| τ³-Bench | 30.5 | 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.