| Benchmark | Claude Fable 5 | Qwen3.5 27B |
|---|---|---|
| AA Agentic Index | 56.6 | 54.6 |
| AA Intelligence | 62.1 | 34.6 |
| AA-LCR | 76.7 | 72.3 |
| AA-Omniscience | 43.3 | -44 |
| arena_vision | 1307 | 1221 |
| Artificial Analysis Coding Index | 76.5 | 34.9 |
| browsecomp | 88 | 61 |
| charxiv_rq | 89.4 | 79.5 |
| coding_arena_elo | 1626 | 1357 |
| critpt | 28.6 | 0.9 |
| gdpval | 61.9 | 33 |
| GPQA Diamond | 92.6 | 85.8 |
| HLE | 64.5 | 48.5 |
| IFBench | 63.5 | 76.5 |
| mathvision | 94.8 | 86 |
| MCP Atlas | 84.7 | 68.4 |
| MMMU-Pro | 84.2 | 75 |
| OmniScience Accuracy | 65.3 | 20.7 |
| OmniScience Non-Hallucination | 36.4 | 24.9 |
| OSWorld-Verified | 85 | 56.2 |
| scicode | 60.2 | 39.5 |
| SWE-bench Multilingual | 86.6 | 69.3 |
| SWE-bench Pro | 80 | 51.2 |
| SWE-bench Verified | 95 | 75 |
| Terminal-Bench Hard | 62.9 | 32.6 |
| WideSearch | 81.2 | 66.4 |
| τ²-Bench Telecom (AA run) | 98.5 | 93.9 |
| τ³-Bench | 26.8 | 68.4 |
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.