| Benchmark | Claude Sonnet 5 | DeepSeek-V4-Flash |
|---|---|---|
| AA Agentic Index | 49.7 | 61.3 |
| AA Intelligence | 55.3 | 51.8 |
| AA-LCR | 77 | 74.3 |
| AA-Omniscience | 16.4 | -14.3 |
| arena_elo | 1463 | 1439 |
| arena_text_factuality | 1453 | 1434 |
| Artificial Analysis Coding Index | 71.5 | 69.1 |
| browsecomp | 84.7 | 73.2 |
| coding_arena_elo | 1540 | 1431 |
| critpt | 16.9 | 16.6 |
| gdpval | 54.8 | 52.9 |
| GDPval-AA v2 | 1618 | 1189 |
| GPQA Diamond | 91.1 | 90.8 |
| HLE | 57.4 | 38.6 |
| livebench_agentic_coding | 59.4 | 46.8 |
| livebench_coding | 80.7 | 75 |
| livebench_data_analysis | 71.7 | 79.3 |
| livebench_instruction_following | 63.9 | 65.5 |
| livebench_language | 75 | 79.2 |
| livebench_math | 92.9 | 86.8 |
| livebench_reasoning | 88.7 | 86.6 |
| OmniScience Accuracy | 40 | 40.4 |
| OmniScience Non-Hallucination | 60.6 | 15.6 |
| scicode | 53.6 | 49.9 |
| simplebench | 60.6 | 46.3 |
| simpleqa_verified | 25 | 45 |
| SWE-bench Multilingual | 78.3 | 73.3 |
| SWE-bench Pro | 63.2 | 52.6 |
| SWE-bench Verified | 85.2 | 79 |
| TauBench V3 - Banking | 37.3 | 39.4 |
| Terminal-Bench 2.0 | 80.4 | 56.9 |
| Terminal-Bench 2.1 | 80.5 | 82.7 |
| toolathlon | 54.3 | 70.3 |
| τ³-Bench | 28.2 | 22.9 |
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.