| Benchmark | Claude Opus 4.8 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 57.3 | 10 |
| AA-LCR | 73 | 20.3 |
| AA-Omniscience | 28.8 | -52.2 |
| Artificial Analysis Coding Index | 74.3 | 11.9 |
| critpt | 20.9 | 0 |
| Fortress | 18.2 | 56.4 |
| GPQA Diamond | 93.6 | 49.1 |
| HLE | 57.9 | 4.2 |
| IFBench | 62.2 | 36.9 |
| livebench | 77.2 | 52.3 |
| longbench_v2 | 69.1 | 39.4 |
| OmniScience Accuracy | 48.8 | 17.6 |
| OmniScience Non-Hallucination | 60.7 | 15.3 |
| scicode | 53.5 | 26.7 |
| SWE-bench Verified | 88.6 | 23.8 |
| Terminal-Bench Hard | 58.3 | 4.5 |
| τ²-Bench Telecom (AA run) | 94.4 | 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.