| Benchmark | Qwen2.5 Instruct 72B | Qwen3.6 27B |
|---|---|---|
| AA Intelligence | 10 | 37.7 |
| AA-LCR | 20.3 | 73.3 |
| AA-Omniscience | -52.2 | -20 |
| Artificial Analysis Coding Index | 11.9 | 53.7 |
| C-Eval | 89.2 | 91.4 |
| critpt | 0 | 1.1 |
| GPQA Diamond | 49.1 | 87.8 |
| hellaswag | 84.8 | 90 |
| HLE | 4.2 | 24 |
| humaneval | 86.6 | 78.2 |
| IFBench | 36.9 | 69.1 |
| mmlu_redux | 86.8 | 93.5 |
| MMLU-Pro | 71.6 | 86.2 |
| OmniScience Accuracy | 17.6 | 19.6 |
| OmniScience Non-Hallucination | 15.3 | 50.7 |
| scicode | 26.7 | 39.8 |
| simpleqa | 10.3 | 95.7 |
| SWE-bench Verified | 23.8 | 77.2 |
| Terminal-Bench Hard | 4.5 | 34.8 |
| τ²-Bench Telecom (AA run) | 34.5 | 94.2 |
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.