| Benchmark | DeepSeek-V4-Pro | o3 |
|---|---|---|
| AA Agentic Index | 63.3 | 36.1 |
| AA Intelligence | 53 | 31.1 |
| AA-LCR | 70 | 73.3 |
| AA-Omniscience | -10.6 | -15.3 |
| Artificial Analysis Coding Index | 59.4 | 38.4 |
| browsecomp | 83.4 | 49.7 |
| critpt | 13 | 1.1 |
| gdpval | 49 | 12.8 |
| GPQA Diamond | 90.5 | 83.3 |
| HLE | 48.2 | 20.3 |
| IFBench | 76.5 | 71.4 |
| LiveCodeBench | 93.5 | 84.7 |
| MMLU-Pro | 87.5 | 85 |
| OmniScience Accuracy | 43 | 38.6 |
| OmniScience Non-Hallucination | 12.2 | 12.9 |
| scicode | 50 | 41 |
| simplebench | 50.9 | 53.1 |
| simpleqa | 57.9 | 49.4 |
| simpleqa_verified | 57.9 | 53 |
| strongreject | 98.6 | 97 |
| SWE-bench Verified | 80.6 | 69.1 |
| Terminal-Bench Hard | 46.2 | 37.1 |
| τ²-Bench Telecom (AA run) | 96.2 | 80.7 |
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.