| Benchmark | Claude Sonnet 4.6 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 48.4 | 10 |
| AA-LCR | 74 | 20.3 |
| AA-Omniscience | 12.2 | -52.2 |
| Artificial Analysis Coding Index | 63 | 11.9 |
| critpt | 3.1 | 0 |
| GPQA Diamond | 89.9 | 49.1 |
| harmbench | 24.2 | 72.8 |
| HLE | 49 | 4.2 |
| IFBench | 56.6 | 36.9 |
| livebench | 75.5 | 52.3 |
| mmmlu | 89.3 | 74.8 |
| OmniScience Accuracy | 40.9 | 17.6 |
| OmniScience Non-Hallucination | 51.6 | 15.3 |
| scicode | 47 | 26.7 |
| SWE-bench Verified | 79.6 | 23.8 |
| Terminal-Bench Hard | 59.1 | 4.5 |
| τ²-Bench Telecom (AA run) | 97.9 | 34.5 |
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.