| Benchmark | Claude Sonnet 4.5 | Claude Sonnet 5 |
|---|---|---|
| AA Agentic Index | 50.6 | 49.7 |
| AA Intelligence | 37.4 | 55.3 |
| AA-LCR | 68.3 | 77 |
| AA-Omniscience | -0.1 | 16.4 |
| arena_elo | 1456 | 1463 |
| arena_text_factuality | 1448 | 1453 |
| Artificial Analysis Coding Index | 52.1 | 71.5 |
| browsecomp | 24.1 | 84.7 |
| charxiv_rq | 67.2 | 77 |
| coding_arena_elo | 1392 | 1540 |
| critpt | 1.1 | 16.9 |
| gdpval | 40.4 | 54.8 |
| GPQA Diamond | 83.4 | 91.1 |
| HLE | 19.8 | 57.4 |
| MMMU-Pro | 68.7 | 77.3 |
| OmniScience Accuracy | 32.9 | 40 |
| OmniScience Non-Hallucination | 50.9 | 60.6 |
| OSWorld-Verified | 61.4 | 81.2 |
| scicode | 45 | 53.6 |
| simplebench | 54.3 | 60.6 |
| simpleqa_verified | 23.6 | 25 |
| SWE-bench Multilingual | 68 | 78.3 |
| SWE-bench Pro | 43.6 | 63.2 |
| SWE-bench Verified | 82 | 85.2 |
| TauBench V3 - Banking | 24.5 | 37.3 |
| Terminal-Bench 2.0 | 50 | 80.4 |
| Terminal-Bench 2.1 | 55.8 | 80.5 |
| toolathlon | 41 | 54.3 |
| Toolathlon Pass@∞ | 28.7 | 40.7 |
| τ³-Bench | 19 | 28.2 |
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.