| Benchmark | Claude Sonnet 5 | Gemini 2.5 Pro |
|---|---|---|
| AA Agentic Index | 49.7 | 7.2 |
| AA Intelligence | 55.3 | 27 |
| AA-LCR | 77 | 66 |
| AA-Omniscience | 16.4 | -14.3 |
| arena_elo | 1463 | 1446 |
| arena_text_factuality | 1453 | 1462 |
| Artificial Analysis Coding Index | 71.5 | 46.7 |
| browsecomp | 84.7 | 9.9 |
| coding_arena_elo | 1540 | 1224 |
| critpt | 16.9 | 2.6 |
| gdpval | 54.8 | 8.5 |
| GPQA Diamond | 91.1 | 84.4 |
| HLE | 57.4 | 22.5 |
| MMMU-Pro | 77.3 | 74.9 |
| OmniScience Accuracy | 40 | 39 |
| OmniScience Non-Hallucination | 60.6 | 12.6 |
| scicode | 53.6 | 43 |
| simplebench | 60.6 | 62.4 |
| simpleqa_verified | 25 | 56 |
| SWE-bench Verified | 85.2 | 67.2 |
| TauBench V3 - Banking | 37.3 | 9.7 |
| Terminal-Bench 2.1 | 80.5 | 28.5 |
| τ³-Bench | 28.2 | 9.3 |
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.