| Benchmark | Claude 3.7 Sonnet | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 27.6 | 10 |
| AA-LCR | 62.3 | 20.3 |
| AA-Omniscience | -0.7 | -52.2 |
| aider_polyglot | 64.9 | 7.6 |
| AIR-Bench 2024 | 81.8 | 59 |
| anthropic_red_team | 99.7 | 99.6 |
| Artificial Analysis Coding Index | 36.4 | 11.9 |
| bbq | 92.1 | 95.4 |
| critpt | 0.9 | 0 |
| Fortress | 38 | 56.4 |
| GPQA Diamond | 84.8 | 49.1 |
| harmbench | 84.3 | 72.8 |
| HLE | 9.7 | 4.2 |
| IFBench | 48.3 | 36.9 |
| ifeval | 93.2 | 87.2 |
| math_500_em | 96.2 | 80 |
| mmmlu | 86.1 | 74.8 |
| OmniScience Accuracy | 28.2 | 17.6 |
| OmniScience Non-Hallucination | 60 | 15.3 |
| scicode | 40.3 | 26.7 |
| simple_safety_tests | 100 | 100 |
| SWE-bench Verified | 70.3 | 23.8 |
| Terminal-Bench Hard | 21.2 | 4.5 |
| xstest | 96.4 | 97.9 |
| τ²-Bench Telecom (AA run) | 54.7 | 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.