| Benchmark | Claude Fable 5 | Gemini 3 Flash Preview |
|---|---|---|
| AA Agentic Index | 56.6 | 49.7 |
| AA Intelligence | 62.1 | 38.7 |
| AA-LCR | 76.7 | 73 |
| AA-Omniscience | 43.3 | 10.1 |
| arc_agi_1 | 90.5 | 84.7 |
| ARC-AGI-2 | 76.8 | 33.6 |
| arena_elo | 1508 | 1473 |
| arena_text_factuality | 1486 | 1452 |
| arena_vision | 1307 | 1260 |
| Artificial Analysis Coding Index | 76.5 | 42.6 |
| Blueprint-Bench 2 | 38.6 | 0 |
| charxiv_rq | 89.4 | 80.3 |
| coding_arena_elo | 1626 | 1387 |
| critpt | 28.6 | 8.6 |
| Finance Agent v2 | 56.3 | 42.6 |
| gdpval | 61.9 | 35.2 |
| GDPval-AA (Elo) | 1932 | 1204 |
| GPQA Diamond | 92.6 | 90.4 |
| HLE | 64.5 | 43.5 |
| IFBench | 63.5 | 78 |
| Legal Agent Benchmark | 13.3 | 0 |
| livebench | 78.3 | 72.4 |
| MCP Atlas | 84.7 | 62 |
| MMMU-Pro | 84.2 | 81.2 |
| OmniScience Accuracy | 65.3 | 53.4 |
| OmniScience Non-Hallucination | 36.4 | 7.6 |
| OSWorld-Verified | 85 | 65.1 |
| scicode | 60.2 | 50.6 |
| simplebench | 81.9 | 61.1 |
| simpleqa_verified | 68.3 | 67.4 |
| SWE-bench Multilingual | 86.6 | 72.7 |
| SWE-bench Pro | 80 | 49.6 |
| SWE-bench Verified | 95 | 78 |
| TauBench V3 - Banking | 38.1 | 20.8 |
| Terminal-Bench 2.1 | 88 | 58 |
| Terminal-Bench Hard | 62.9 | 38.6 |
| toolathlon | 61.7 | 49.4 |
| τ²-Bench Telecom (AA run) | 98.5 | 80.4 |
| τ³-Bench | 26.8 | 17.5 |
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.