| Benchmark | GPT-5.1 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 37.5 | 10 |
| AA-LCR | 76.7 | 20.3 |
| AA-Omniscience | 5.4 | -52.2 |
| Artificial Analysis Coding Index | 49.4 | 11.9 |
| critpt | 4.9 | 0 |
| Fortress | 25.7 | 56.4 |
| GPQA Diamond | 88.1 | 49.1 |
| HLE | 28.5 | 4.2 |
| IFBench | 72.9 | 36.9 |
| MMLU-Pro | 87 | 71.6 |
| mmmlu | 91 | 74.8 |
| OmniScience Accuracy | 37.7 | 17.6 |
| OmniScience Non-Hallucination | 48.1 | 15.3 |
| scicode | 43.3 | 26.7 |
| SWE-bench Verified | 76.3 | 23.8 |
| Terminal-Bench Hard | 45.5 | 4.5 |
| τ²-Bench Telecom (AA run) | 81.9 | 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.