| Benchmark | DeepSeek-V4-Pro | o1 |
|---|---|---|
| AA Agentic Index | 63.3 | 31.1 |
| AA Intelligence | 53 | 24 |
| AA-LCR | 70 | 63.3 |
| AA-Omniscience | -10.6 | -10.6 |
| Artificial Analysis Coding Index | 59.4 | 39.7 |
| Codeforces | 3206 | 2061 |
| critpt | 13 | 0.3 |
| gdpval | 49 | 11.5 |
| GPQA Diamond | 90.5 | 75.7 |
| HLE | 48.2 | 7.7 |
| IFBench | 76.5 | 70.3 |
| livebench | 73.6 | 52.3 |
| OmniScience Accuracy | 43 | 34.7 |
| OmniScience Non-Hallucination | 12.2 | 30.7 |
| scicode | 50 | 35.8 |
| simplebench | 50.9 | 41.7 |
| simpleqa | 57.9 | 47 |
| strongreject | 98.6 | 97 |
| SWE-bench Verified | 80.6 | 48.9 |
| Terminal-Bench Hard | 46.2 | 12.9 |
| τ²-Bench Telecom (AA run) | 96.2 | 62.6 |
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.