| Benchmark | DeepSeek-V4-Pro | Llama 3.1 Instruct 405B |
|---|---|---|
| AA Agentic Index | 63.3 | 6.3 |
| AA Intelligence | 53 | 8.5 |
| AA-LCR | 70 | 24.7 |
| AA-Omniscience | -10.6 | -17.1 |
| Artificial Analysis Coding Index | 59.4 | 14.5 |
| Chinese SimpleQA (C-SimpleQA) | 77.7 | 54.7 |
| Codeforces | 3206 | 25.3 |
| critpt | 13 | 0 |
| gdpval | 49 | 0 |
| GPQA Diamond | 90.5 | 51.5 |
| HLE | 48.2 | 4.2 |
| IFBench | 76.5 | 39 |
| LiveCodeBench | 93.5 | 30.1 |
| MMLU-Pro | 87.5 | 73.4 |
| OmniScience Accuracy | 43 | 23.2 |
| OmniScience Non-Hallucination | 12.2 | 49 |
| scicode | 50 | 29.9 |
| simplebench | 50.9 | 23 |
| simpleqa | 57.9 | 23.2 |
| SWE-bench Pro | 55.4 | 11.2 |
| SWE-bench Verified | 80.6 | 24.5 |
| Terminal-Bench Hard | 46.2 | 6.8 |
| τ²-Bench Telecom (AA run) | 96.2 | 19 |
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.