| Benchmark | Claude Fable 5 | Qwen3.6 27B |
|---|---|---|
| AA Agentic Index | 56.6 | 27.5 |
| AA Intelligence | 62.1 | 37.7 |
| AA-LCR | 76.7 | 73.3 |
| AA-Omniscience | 43.3 | -20 |
| Agents' Last Exam | 25.7 | 27.3 |
| Artificial Analysis Coding Index | 76.5 | 53.7 |
| charxiv_rq | 89.4 | 78.4 |
| critpt | 28.6 | 1.1 |
| DeepSWE 1.1 | 70 | 13.3 |
| gdpval | 61.9 | 32 |
| GPQA Diamond | 92.6 | 87.8 |
| HLE | 64.5 | 24 |
| IFBench | 63.5 | 69.1 |
| JobBench | 57.4 | 21.8 |
| livebench_agentic_coding | 62.2 | 39.3 |
| livebench_coding | 86 | 71.8 |
| livebench_data_analysis | 80.5 | 70.4 |
| livebench_instruction_following | 75.8 | 53.2 |
| livebench_language | 90.7 | 63.3 |
| livebench_math | 96 | 79.9 |
| livebench_reasoning | 89.7 | 70.3 |
| mathvision | 94.8 | 85.1 |
| MMMU-Pro | 84.2 | 75.8 |
| OmniScience Accuracy | 65.3 | 19.6 |
| OmniScience Non-Hallucination | 36.4 | 50.7 |
| OSWorld-Verified | 85 | 63.9 |
| scicode | 60.2 | 39.8 |
| SWE-bench Multilingual | 86.6 | 71.3 |
| SWE-bench Pro | 80 | 53.5 |
| SWE-bench Verified | 95 | 77.2 |
| TauBench V3 - Banking | 38.1 | 16.7 |
| Terminal-Bench 2.1 | 88 | 60.7 |
| Terminal-Bench Hard | 62.9 | 34.8 |
| τ²-Bench Telecom (AA run) | 98.5 | 94.2 |
| τ³-Bench | 26.8 | 15.3 |
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.