| Benchmark | Claude Fable 5 | Claude Sonnet 4 |
|---|---|---|
| AA Agentic Index | 56.6 | 39.2 |
| AA Intelligence | 62.1 | 29.8 |
| AA-LCR | 76.7 | 69.7 |
| AA-Omniscience | 43.3 | 0.2 |
| arc_agi_1 | 90.5 | 63.7 |
| ARC-AGI-2 | 76.8 | 13.6 |
| arena_vision | 1307 | 1207 |
| Artificial Analysis Coding Index | 76.5 | 37.6 |
| browsecomp | 88 | 12.2 |
| critpt | 28.6 | 1.1 |
| gdpval | 61.9 | 31.2 |
| GPQA Diamond | 92.6 | 78 |
| HLE | 64.5 | 10.7 |
| HLE (with tools) | 64.5 | 20.3 |
| IFBench | 63.5 | 55 |
| livebench | 78.3 | 74.8 |
| MMMU-Pro | 84.2 | 62.4 |
| OmniScience Accuracy | 65.3 | 22.7 |
| OmniScience Non-Hallucination | 36.4 | 70.9 |
| OSWorld-Verified | 85 | 42.2 |
| scicode | 60.2 | 40 |
| simplebench | 81.9 | 45.5 |
| SWE-bench Multilingual | 86.6 | 56.9 |
| SWE-bench Pro | 80 | 42.7 |
| SWE-bench Verified | 95 | 80.2 |
| TauBench V3 - Banking | 38.1 | 16.7 |
| Terminal-Bench 2.1 | 88 | 36.3 |
| Terminal-Bench Hard | 62.9 | 31.1 |
| τ²-Bench Telecom (AA run) | 98.5 | 64.6 |
| τ³-Bench | 26.8 | 13.8 |
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.