| Benchmark | Claude Sonnet 4.5 | Gemini 3.6 Flash |
|---|---|---|
| AA Agentic Index | 50.6 | 40.5 |
| AA Intelligence | 37.4 | 52 |
| AA-LCR | 68.3 | 79 |
| AA-Omniscience | -0.1 | 22.1 |
| arc_agi_1 | 63.7 | 34.5 |
| ARC-AGI-2 | 13.6 | 2.6 |
| arena_elo | 1456 | 1481 |
| arena_text_factuality | 1448 | 1470 |
| Artificial Analysis Coding Index | 52.1 | 69.2 |
| charxiv_rq | 67.2 | 85.2 |
| coding_arena_elo | 1392 | 1538 |
| critpt | 1.1 | 10.6 |
| gdpval | 40.4 | 46.1 |
| GPQA Diamond | 83.4 | 92.8 |
| HLE | 19.8 | 40.8 |
| matharena_visual_math_overall | 75.8 | 87.8 |
| MMMU-Pro | 68.7 | 83.2 |
| OmniScience Accuracy | 32.9 | 50 |
| OmniScience Non-Hallucination | 50.9 | 44.4 |
| OSWorld-Verified | 61.4 | 83 |
| scicode | 45 | 52.7 |
| simpleqa_verified | 23.6 | 68.7 |
| SWE-bench Pro | 43.6 | 58.7 |
| TauBench V3 - Banking | 24.5 | 29.9 |
| Terminal-Bench 2.1 | 55.8 | 78 |
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.