| Benchmark | Claude Sonnet 4 | Gemini 3.5 Flash |
|---|---|---|
| AA Agentic Index | 39.2 | 70.4 |
| AA Intelligence | 29.8 | 52 |
| AA-LCR | 69.7 | 81 |
| AA-Omniscience | 0.2 | 21.2 |
| arc_agi_1 | 63.7 | 92.5 |
| ARC-AGI-2 | 13.6 | 72.1 |
| Artificial Analysis Coding Index | 37.6 | 70.1 |
| critpt | 1.1 | 13.1 |
| frontiermath_tier_4 | 0 | 14.6 |
| gdpval | 31.2 | 57.8 |
| GPQA Diamond | 78 | 92.2 |
| HLE | 10.7 | 42.7 |
| IFBench | 55 | 76.3 |
| livebench | 74.8 | 75 |
| MMMU-Pro | 62.4 | 84.3 |
| OmniScience Accuracy | 22.7 | 51.4 |
| OmniScience Non-Hallucination | 70.9 | 38.2 |
| OSWorld-Verified | 42.2 | 78.4 |
| scicode | 40 | 53.1 |
| simplebench | 45.5 | 76.7 |
| SWE-bench Pro | 42.7 | 55.1 |
| TauBench V3 - Banking | 16.7 | 32.2 |
| Terminal-Bench 2.1 | 36.3 | 78.7 |
| Terminal-Bench Hard | 31.1 | 46.2 |
| τ²-Bench Telecom (AA run) | 64.6 | 95.6 |
| τ³-Bench | 13.8 | 25.4 |
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.