| Benchmark | Claude Sonnet 4 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 29.8 | 10 |
| AA-LCR | 69.7 | 20.3 |
| AA-Omniscience | 0.2 | -52.2 |
| aider_polyglot | 70.7 | 7.6 |
| AIME 2024 | 43.4 | 23.3 |
| AIR-Bench 2024 | 88.3 | 59 |
| anthropic_red_team | 99.2 | 99.6 |
| Artificial Analysis Coding Index | 37.6 | 11.9 |
| bbq | 97.9 | 95.4 |
| critpt | 1.1 | 0 |
| Fortress | 24.4 | 56.4 |
| GPQA Diamond | 78 | 49.1 |
| harmbench | 98 | 72.8 |
| HLE | 10.7 | 4.2 |
| IFBench | 55 | 36.9 |
| ifeval | 87.6 | 87.2 |
| livebench | 74.8 | 52.3 |
| LiveCodeBench | 68.5 | 55.5 |
| math_500_em | 94 | 80 |
| mmlu | 91.5 | 86.1 |
| mmlu_redux | 93.6 | 86.8 |
| MMLU-Pro | 84 | 71.6 |
| mmmlu | 86.5 | 74.8 |
| MultiPL-E | 88.6 | 75.1 |
| OmniScience Accuracy | 22.7 | 17.6 |
| OmniScience Non-Hallucination | 70.9 | 15.3 |
| scicode | 40 | 26.7 |
| simple_safety_tests | 100 | 100 |
| simpleqa | 15.9 | 10.3 |
| SWE-bench Verified | 80.2 | 23.8 |
| Terminal-Bench Hard | 31.1 | 4.5 |
| xstest | 97.2 | 97.9 |
| τ²-Bench Telecom (AA run) | 64.6 | 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.