| Benchmark | Claude Sonnet 4 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 39.2 | 27.5 |
| AA Intelligence | 29.8 | 37.7 |
| AA-LCR | 69.7 | 73.3 |
| AA-Omniscience | 0.2 | -20 |
| Artificial Analysis Coding Index | 37.6 | 53.7 |
| critpt | 1.1 | 1.1 |
| gdpval | 31.2 | 32 |
| GPQA Diamond | 78 | 87.8 |
| HLE | 10.7 | 24 |
| HMMT 2025 | 15.9 | 90.7 |
| IFBench | 55 | 69.1 |
| LiveCodeBench v6 | 48.5 | 83.9 |
| mmlu_redux | 93.6 | 93.5 |
| MMLU-Pro | 84 | 86.2 |
| MMMU | 74.4 | 82.9 |
| MMMU-Pro | 62.4 | 75.8 |
| OmniScience Accuracy | 22.7 | 19.6 |
| OmniScience Non-Hallucination | 70.9 | 50.7 |
| OSWorld-Verified | 42.2 | 63.9 |
| scicode | 40 | 39.8 |
| simpleqa | 15.9 | 95.7 |
| supergpqa | 55.7 | 66 |
| SWE-bench Multilingual | 56.9 | 71.3 |
| SWE-bench Pro | 42.7 | 53.5 |
| SWE-bench Verified | 80.2 | 77.2 |
| TauBench V3 - Banking | 16.7 | 16.7 |
| Terminal-Bench 2.1 | 36.3 | 60.7 |
| Terminal-Bench Hard | 31.1 | 34.8 |
| τ²-Bench Telecom (AA run) | 64.6 | 94.2 |
| τ³-Bench | 13.8 | 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.