| Benchmark | Claude Sonnet 4.5 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 37.4 | 10 |
| AA-LCR | 68.3 | 20.3 |
| AA-Omniscience | -0.1 | -52.2 |
| AIR-Bench 2024 | 89.8 | 59 |
| Artificial Analysis Coding Index | 52.1 | 11.9 |
| critpt | 1.1 | 0 |
| Fortress | 17 | 56.4 |
| GPQA Diamond | 83.4 | 49.1 |
| HLE | 19.8 | 4.2 |
| IFBench | 57.3 | 36.9 |
| LiveCodeBench | 71 | 55.5 |
| longbench_v2 | 61.8 | 39.4 |
| mmlu_redux | 95.6 | 86.8 |
| MMLU-Pro | 88.2 | 71.6 |
| mmmlu | 89.1 | 74.8 |
| OmniScience Accuracy | 32.9 | 17.6 |
| OmniScience Non-Hallucination | 50.9 | 15.3 |
| scicode | 45 | 26.7 |
| SWE-bench Verified | 82 | 23.8 |
| Terminal-Bench Hard | 35.6 | 4.5 |
| τ²-Bench Telecom (AA run) | 78.1 | 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.