| Benchmark | Claude Fable 5 | GPT-5.1 |
|---|---|---|
| AA Agentic Index | 56.6 | 32.2 |
| AA Intelligence | 62.1 | 37.5 |
| AA-LCR | 76.7 | 76.7 |
| AA-Omniscience | 43.3 | 5.4 |
| arc_agi_1 | 90.5 | 72.8 |
| ARC-AGI-2 | 76.8 | 17.6 |
| arena_elo | 1508 | 1439 |
| arena_text_factuality | 1486 | 1445 |
| arena_vision | 1307 | 1250 |
| Artificial Analysis Coding Index | 76.5 | 49.4 |
| browsecomp | 88 | 50.8 |
| coding_arena_elo | 1626 | 1421 |
| critpt | 28.6 | 4.9 |
| gdpval | 61.9 | 25 |
| GPQA Diamond | 92.6 | 88.1 |
| HLE | 64.5 | 28.5 |
| HLE (with tools) | 64.5 | 42.7 |
| IFBench | 63.5 | 72.9 |
| MCP Atlas | 84.7 | 50.1 |
| MMMU-Pro | 84.2 | 75.5 |
| OmniScience Accuracy | 65.3 | 37.7 |
| OmniScience Non-Hallucination | 36.4 | 48.1 |
| scicode | 60.2 | 43.3 |
| simplebench | 81.9 | 53.2 |
| simpleqa_verified | 68.3 | 48.9 |
| SWE-bench Verified | 95 | 76.3 |
| TauBench V3 - Banking | 38.1 | 15.9 |
| Terminal-Bench 2.1 | 88 | 52.4 |
| Terminal-Bench Hard | 62.9 | 45.5 |
| τ²-Bench Telecom (AA run) | 98.5 | 81.9 |
| τ³-Bench | 26.8 | 14 |
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.