| Benchmark | Claude Sonnet 5 | Gemini 3 Pro |
|---|---|---|
| AA Agentic Index | 49.7 | 52 |
| AA Intelligence | 55.3 | 40.6 |
| AA-LCR | 77 | 73 |
| AA-Omniscience | 16.4 | 15.3 |
| arena_elo | 1463 | 1485 |
| arena_text_factuality | 1453 | 1481 |
| Artificial Analysis Coding Index | 71.5 | 46.5 |
| browsecomp | 84.7 | 59.2 |
| charxiv_rq | 77 | 81.4 |
| coding_arena_elo | 1540 | 1438 |
| critpt | 16.9 | 9.1 |
| gdpval | 54.8 | 34.2 |
| GPQA Diamond | 91.1 | 91.9 |
| HLE | 57.4 | 45.8 |
| LVBench | 68.5 | 73.5 |
| MMMU-Pro | 77.3 | 81 |
| mrcr_v2_8needle_128k_average | 81.5 | 77 |
| OmniScience Accuracy | 40 | 55.8 |
| OmniScience Non-Hallucination | 60.6 | 10 |
| scicode | 53.6 | 56.1 |
| simplebench | 60.6 | 76.4 |
| simpleqa_verified | 25 | 72.9 |
| SWE-bench Multilingual | 78.3 | 68.7 |
| SWE-bench Pro | 63.2 | 43.3 |
| SWE-bench Verified | 85.2 | 78 |
| Terminal-Bench 2.0 | 80.4 | 54.2 |
| toolathlon | 54.3 | 36.4 |
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.