| Benchmark | Claude Fable 5 | DeepSeek-V3.2 |
|---|---|---|
| AA Agentic Index | 56.6 | 39.8 |
| AA Intelligence | 62.1 | 32.8 |
| AA-LCR | 76.7 | 70.7 |
| AA-Omniscience | 43.3 | -22.5 |
| arc_agi_1 | 90.5 | 57 |
| ARC-AGI-2 | 76.8 | 4 |
| arena_text_factuality | 1486 | 1433 |
| Artificial Analysis Coding Index | 76.5 | 44.2 |
| browsecomp | 88 | 51.4 |
| coding_arena_elo | 1626 | 1368 |
| critpt | 28.6 | 2.9 |
| deepsearchqa_f1 | 94.2 | 60.9 |
| gdpval | 61.9 | 18.8 |
| GPQA Diamond | 92.6 | 84 |
| HLE | 64.5 | 40.8 |
| HLE (with tools) | 64.5 | 40.8 |
| IFBench | 63.5 | 61 |
| MCP Atlas | 84.7 | 62.2 |
| OmniScience Accuracy | 65.3 | 33 |
| OmniScience Non-Hallucination | 36.4 | 17.3 |
| PaperBench | 88.8 | 47.1 |
| scicode | 60.2 | 39 |
| simpleqa_verified | 68.3 | 27.5 |
| SWE-bench Multilingual | 86.6 | 70.2 |
| SWE-bench Pro | 80 | 15.6 |
| SWE-bench Verified | 95 | 73.1 |
| TauBench V3 - Banking | 38.1 | 18.8 |
| Terminal-Bench 2.1 | 88 | 46.8 |
| Terminal-Bench Hard | 62.9 | 35.6 |
| toolathlon | 61.7 | 35.2 |
| τ²-Bench Telecom (AA run) | 98.5 | 90.6 |
| τ³-Bench | 26.8 | 69.2 |
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.