| Benchmark | Claude Sonnet 4 | Claude Sonnet 5 |
|---|---|---|
| AA Agentic Index | 39.2 | 49.7 |
| AA Intelligence | 29.8 | 55.3 |
| AA-LCR | 69.7 | 77 |
| AA-Omniscience | 0.2 | 16.4 |
| Artificial Analysis Coding Index | 37.6 | 71.5 |
| browsecomp | 12.2 | 84.7 |
| critpt | 1.1 | 16.9 |
| gdpval | 31.2 | 54.8 |
| GPQA Diamond | 78 | 91.1 |
| HLE | 10.7 | 57.4 |
| MMMU-Pro | 62.4 | 77.3 |
| OmniScience Accuracy | 22.7 | 40 |
| OmniScience Non-Hallucination | 70.9 | 60.6 |
| OSWorld-Verified | 42.2 | 81.2 |
| scicode | 40 | 53.6 |
| simplebench | 45.5 | 60.6 |
| SWE-bench Multilingual | 56.9 | 78.3 |
| SWE-bench Pro | 42.7 | 63.2 |
| SWE-bench Verified | 80.2 | 85.2 |
| TauBench V3 - Banking | 16.7 | 37.3 |
| Terminal-Bench 2.1 | 36.3 | 80.5 |
| τ³-Bench | 13.8 | 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.