| Benchmark | DeepSeek-R1 | DeepSeek-V4-Pro |
|---|---|---|
| AA Agentic Index | 3.1 | 63.3 |
| AA Intelligence | 18.6 | 53 |
| AA-LCR | 56 | 70 |
| AA-Omniscience | -31.3 | -10.6 |
| Artificial Analysis Coding Index | 24.6 | 59.4 |
| browsecomp | 8.9 | 83.4 |
| Chinese SimpleQA (C-SimpleQA) | 63.7 | 77.7 |
| Codeforces | 2029 | 3206 |
| critpt | 0.6 | 13 |
| gdpval | 1.5 | 49 |
| GPQA Diamond | 81 | 90.5 |
| HLE | 17.7 | 48.2 |
| IFBench | 39 | 76.5 |
| LiveCodeBench | 84.4 | 93.5 |
| MMLU-Pro | 85 | 87.5 |
| OmniScience Accuracy | 30.7 | 43 |
| OmniScience Non-Hallucination | 10.5 | 12.2 |
| scicode | 35.7 | 50 |
| simplebench | 40.8 | 50.9 |
| simpleqa | 92.3 | 57.9 |
| simpleqa_verified | 27.4 | 57.9 |
| SWE-bench Multilingual | 30.5 | 76.2 |
| SWE-bench Verified | 57.6 | 80.6 |
| TauBench V3 - Banking | 6.4 | 30.1 |
| Terminal-Bench 2.1 | 19.1 | 72.1 |
| Terminal-Bench Hard | 6.1 | 46.2 |
| vectara_answer_rate | 97 | 97.2 |
| vectara_avg_summary_length | 93.5 | 153.8 |
| vectara_factual_consistency | 88.7 | 91.4 |
| vectara_hallucination_rate ↓ | 11.3 | 8.6 |
| τ²-Bench Telecom (AA run) | 11.4 | 96.2 |
| τ³-Bench | 6.4 | 25.8 |
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.