| Benchmark | Claude Fable 5 | Gemma 4 31B |
|---|---|---|
| AA Agentic Index | 56.6 | 14.4 |
| AA Intelligence | 62.1 | 29.7 |
| AA-LCR | 76.7 | 68.3 |
| AA-Omniscience | 43.3 | -47.9 |
| arena_elo | 1508 | 1451 |
| arena_text_factuality | 1486 | 1449 |
| arena_vision | 1307 | 1252 |
| Artificial Analysis Coding Index | 76.5 | 43.4 |
| charxiv_rq | 89.4 | 67.9 |
| coding_arena_elo | 1626 | 1364 |
| critpt | 28.6 | 1.4 |
| gdpval | 61.9 | 15.5 |
| GPQA Diamond | 92.6 | 85.7 |
| HLE | 64.5 | 26.5 |
| HLE (with tools) | 64.5 | 26.5 |
| IFBench | 63.5 | 75.6 |
| mathvision | 94.8 | 85.6 |
| MCP Atlas | 84.7 | 57.2 |
| MMMU-Pro | 84.2 | 76.9 |
| OmniScience Accuracy | 65.3 | 20 |
| OmniScience Non-Hallucination | 36.4 | 18.1 |
| scicode | 60.2 | 43.4 |
| simpleqa_verified | 68.3 | 9.6 |
| SWE-bench Multilingual | 86.6 | 51.7 |
| SWE-bench Pro | 80 | 35.7 |
| SWE-bench Verified | 95 | 52 |
| TauBench V3 - Banking | 38.1 | 14.8 |
| Terminal-Bench 2.1 | 88 | 43.4 |
| Terminal-Bench Hard | 62.9 | 36.4 |
| WideSearch | 81.2 | 35.2 |
| τ²-Bench Telecom (AA run) | 98.5 | 65.5 |
| τ³-Bench | 26.8 | 67.5 |
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.