| Benchmark | Claude Fable 5 | DeepSeek-R1 |
|---|---|---|
| AA Agentic Index | 56.6 | 3.1 |
| AA Intelligence | 62.1 | 18.6 |
| AA-LCR | 76.7 | 56 |
| AA-Omniscience | 43.3 | -31.3 |
| arc_agi_1 | 90.5 | 21.2 |
| ARC-AGI-2 | 76.8 | 1.3 |
| Artificial Analysis Coding Index | 76.5 | 24.6 |
| browsecomp | 88 | 8.9 |
| critpt | 28.6 | 0.6 |
| gdpval | 61.9 | 1.5 |
| GPQA Diamond | 92.6 | 81 |
| HLE | 64.5 | 17.7 |
| IFBench | 63.5 | 39 |
| OmniScience Accuracy | 65.3 | 30.7 |
| OmniScience Non-Hallucination | 36.4 | 10.5 |
| scicode | 60.2 | 35.7 |
| simplebench | 81.9 | 40.8 |
| simpleqa_verified | 68.3 | 27.4 |
| SWE-bench Multilingual | 86.6 | 30.5 |
| SWE-bench Verified | 95 | 57.6 |
| TauBench V3 - Banking | 38.1 | 6.4 |
| Terminal-Bench 2.1 | 88 | 19.1 |
| Terminal-Bench Hard | 62.9 | 6.1 |
| τ²-Bench Telecom (AA run) | 98.5 | 11.4 |
| τ³-Bench | 26.8 | 6.4 |
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.