| Benchmark | Claude Fable 5 | Claude Opus 4.5 |
|---|---|---|
| AA Agentic Index | 56.6 | 59.6 |
| AA Intelligence | 62.1 | 41.9 |
| AA-LCR | 76.7 | 76 |
| AA-Omniscience | 43.3 | 14 |
| arc_agi_1 | 90.5 | 82 |
| ARC-AGI-2 | 76.8 | 37.6 |
| arena_elo | 1508 | 1473 |
| arena_text_factuality | 1486 | 1458 |
| Artificial Analysis Coding Index | 76.5 | 47.8 |
| browsecomp | 88 | 37 |
| charxiv_rq | 89.4 | 68.5 |
| coding_arena_elo | 1626 | 1495 |
| critpt | 28.6 | 4.6 |
| deepsearchqa_f1 | 94.2 | 76.1 |
| gdpval | 61.9 | 47.3 |
| GPQA Diamond | 92.6 | 87 |
| HLE | 64.5 | 30.8 |
| HLE (with tools) | 64.5 | 43.4 |
| IFBench | 63.5 | 58 |
| livebench | 78.3 | 76 |
| livebench_agentic_coding | 62.2 | 39.7 |
| livebench_coding | 86 | 79.7 |
| livebench_data_analysis | 80.5 | 74.4 |
| livebench_instruction_following | 75.8 | 62.5 |
| livebench_language | 90.7 | 81.3 |
| livebench_math | 96 | 90.4 |
| livebench_reasoning | 89.7 | 80.1 |
| mathvision | 94.8 | 77.1 |
| MCP Atlas | 84.7 | 69.8 |
| MMMU-Pro | 84.2 | 74 |
| OmniScience Accuracy | 65.3 | 46.6 |
| OmniScience Non-Hallucination | 36.4 | 39 |
| OSWorld-Verified | 85 | 66.3 |
| PaperBench | 88.8 | 72.9 |
| scicode | 60.2 | 50 |
| simplebench | 81.9 | 62 |
| simpleqa_verified | 68.3 | 41.8 |
| SWE-bench Multilingual | 86.6 | 77.5 |
| SWE-bench Pro | 80 | 57.1 |
| SWE-bench Verified | 95 | 81.5 |
| Terminal-Bench Hard | 62.9 | 47 |
| toolathlon | 61.7 | 43.5 |
| τ²-Bench Telecom (AA run) | 98.5 | 89.5 |
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.