| Benchmark | Qwen2.5 Instruct 72B | Qwen3.7 Max |
|---|---|---|
| AA Intelligence | 10 | 46.7 |
| AA-LCR | 20.3 | 74.7 |
| AA-Omniscience | -52.2 | 13.5 |
| Artificial Analysis Coding Index | 11.9 | 66 |
| critpt | 0 | 13.4 |
| GPQA Diamond | 49.1 | 92.4 |
| HLE | 4.2 | 41.4 |
| IFBench | 36.9 | 80.5 |
| ifeval | 87.2 | 94.3 |
| livebench | 52.3 | 74.3 |
| longbench_v2 | 39.4 | 65.3 |
| mmlu_redux | 86.8 | 95 |
| MMLU-Pro | 71.6 | 89.6 |
| mmmlu | 74.8 | 90.3 |
| OmniScience Accuracy | 17.6 | 31.1 |
| OmniScience Non-Hallucination | 15.3 | 74.4 |
| scicode | 26.7 | 53.5 |
| SWE-bench Verified | 23.8 | 80.4 |
| Terminal-Bench Hard | 4.5 | 50.8 |
| τ²-Bench Telecom (AA run) | 34.5 | 94.7 |
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.