| Benchmark | Claude Sonnet 5 | DeepSeek-V3.2 |
|---|---|---|
| AA Agentic Index | 49.7 | 39.8 |
| AA Intelligence | 55.3 | 32.8 |
| AA-LCR | 77 | 70.7 |
| AA-Omniscience | 16.4 | -22.5 |
| arena_text_factuality | 1453 | 1433 |
| Artificial Analysis Coding Index | 71.5 | 44.2 |
| browsecomp | 84.7 | 51.4 |
| coding_arena_elo | 1540 | 1368 |
| critpt | 16.9 | 2.9 |
| gdpval | 54.8 | 18.8 |
| GPQA Diamond | 91.1 | 84 |
| HLE | 57.4 | 40.8 |
| OmniScience Accuracy | 40 | 33 |
| OmniScience Non-Hallucination | 60.6 | 17.3 |
| scicode | 53.6 | 39 |
| simpleqa_verified | 25 | 27.5 |
| SWE-bench Multilingual | 78.3 | 70.2 |
| SWE-bench Pro | 63.2 | 15.6 |
| SWE-bench Verified | 85.2 | 73.1 |
| TauBench V3 - Banking | 37.3 | 18.8 |
| Terminal-Bench 2.0 | 80.4 | 46.4 |
| Terminal-Bench 2.1 | 80.5 | 46.8 |
| toolathlon | 54.3 | 35.2 |
| τ³-Bench | 28.2 | 69.2 |
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.