| Benchmark | Claude Fable 5 | Gemini 2.5 Pro |
|---|---|---|
| AA Agentic Index | 56.6 | 7.2 |
| AA Intelligence | 62.1 | 27 |
| AA-LCR | 76.7 | 66 |
| AA-Omniscience | 43.3 | -14.3 |
| arc_agi_1 | 90.5 | 41 |
| ARC-AGI-2 | 76.8 | 4.9 |
| arena_elo | 1508 | 1446 |
| arena_text_factuality | 1486 | 1462 |
| arena_vision | 1307 | 1246 |
| Artificial Analysis Coding Index | 76.5 | 46.7 |
| browsecomp | 88 | 9.9 |
| coding_arena_elo | 1626 | 1224 |
| critpt | 28.6 | 2.6 |
| gdpval | 61.9 | 8.5 |
| Global-MMLU-Lite | 93.3 | 88.6 |
| GPQA Diamond | 92.6 | 84.4 |
| HLE | 64.5 | 22.5 |
| HLE (with tools) | 64.5 | 28.4 |
| IFBench | 63.5 | 49 |
| mathvision | 94.8 | 73.3 |
| MMMU-Pro | 84.2 | 74.9 |
| OmniScience Accuracy | 65.3 | 39 |
| OmniScience Non-Hallucination | 36.4 | 12.6 |
| scicode | 60.2 | 43 |
| simplebench | 81.9 | 62.4 |
| simpleqa_verified | 68.3 | 56 |
| SWE-bench Verified | 95 | 67.2 |
| TauBench V3 - Banking | 38.1 | 9.7 |
| Terminal-Bench 2.1 | 88 | 28.5 |
| Terminal-Bench Hard | 62.9 | 26.5 |
| τ²-Bench Telecom (AA run) | 98.5 | 54.1 |
| τ³-Bench | 26.8 | 9.3 |
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.