| Benchmark | Claude Fable 5 | Qwen2.5 Instruct 72B |
|---|---|---|
| AA Intelligence | 62.1 | 10 |
| AA-LCR | 76.7 | 20.3 |
| AA-Omniscience | 43.3 | -52.2 |
| Artificial Analysis Coding Index | 76.5 | 11.9 |
| critpt | 28.6 | 0 |
| GPQA Diamond | 92.6 | 49.1 |
| HLE | 64.5 | 4.2 |
| IFBench | 63.5 | 36.9 |
| livebench | 78.3 | 52.3 |
| OmniScience Accuracy | 65.3 | 17.6 |
| OmniScience Non-Hallucination | 36.4 | 15.3 |
| scicode | 60.2 | 26.7 |
| SWE-bench Verified | 95 | 23.8 |
| Terminal-Bench Hard | 62.9 | 4.5 |
| τ²-Bench Telecom (AA run) | 98.5 | 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.