| Benchmark | Claude Opus 4.5 | Claude Sonnet 5 |
|---|---|---|
| AA Agentic Index | 59.6 | 49.7 |
| AA Intelligence | 41.9 | 55.3 |
| AA-LCR | 76 | 77 |
| AA-Omniscience | 14 | 16.4 |
| arena_elo | 1473 | 1463 |
| arena_text_factuality | 1458 | 1453 |
| Artificial Analysis Coding Index | 47.8 | 71.5 |
| browsecomp | 37 | 84.7 |
| charxiv_rq | 68.5 | 77 |
| coding_arena_elo | 1495 | 1540 |
| critpt | 4.6 | 16.9 |
| gdpval | 47.3 | 54.8 |
| GPQA Diamond | 87 | 91.1 |
| HLE | 30.8 | 57.4 |
| livebench_agentic_coding | 39.7 | 59.4 |
| livebench_coding | 79.7 | 80.7 |
| livebench_data_analysis | 74.4 | 71.7 |
| livebench_instruction_following | 62.5 | 63.9 |
| livebench_language | 81.3 | 75 |
| livebench_math | 90.4 | 92.9 |
| livebench_reasoning | 80.1 | 88.7 |
| MMMU-Pro | 74 | 77.3 |
| OmniScience Accuracy | 46.6 | 40 |
| OmniScience Non-Hallucination | 39 | 60.6 |
| OSWorld-Verified | 66.3 | 81.2 |
| scicode | 50 | 53.6 |
| simplebench | 62 | 60.6 |
| simpleqa_verified | 41.8 | 25 |
| SWE-bench Multilingual | 77.5 | 78.3 |
| SWE-bench Pro | 57.1 | 63.2 |
| SWE-bench Verified | 81.5 | 85.2 |
| Terminal-Bench 2.0 | 59.3 | 80.4 |
| toolathlon | 43.5 | 54.3 |
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.