| Benchmark | DeepSeek-V3 | DeepSeek-V4-Pro |
|---|---|---|
| AA Agentic Index | 1.6 | 63.3 |
| AA Intelligence | 15.4 | 53 |
| AA-LCR | 41.3 | 70 |
| AA-Omniscience | -37.6 | -10.6 |
| Artificial Analysis Coding Index | 23 | 59.4 |
| Chinese SimpleQA (C-SimpleQA) | 68 | 77.7 |
| Codeforces | 1134 | 3206 |
| critpt | 0 | 13 |
| gdpval | 0 | 49 |
| GPQA Diamond | 68.4 | 90.5 |
| HLE | 5.2 | 48.2 |
| IFBench | 41 | 76.5 |
| livebench | 72.4 | 73.6 |
| LiveCodeBench | 49.6 | 93.5 |
| MMLU-Pro | 81.2 | 87.5 |
| OmniScience Accuracy | 25.4 | 43 |
| OmniScience Non-Hallucination | 23.3 | 12.2 |
| scicode | 35.8 | 50 |
| simplebench | 27.2 | 50.9 |
| simpleqa | 27.7 | 57.9 |
| SWE-bench Verified | 42 | 80.6 |
| TauBench V3 - Banking | 4.7 | 30.1 |
| Terminal-Bench 2.1 | 16.9 | 72.1 |
| Terminal-Bench Hard | 15.2 | 46.2 |
| vectara_answer_rate | 97.5 | 97.2 |
| vectara_avg_summary_length | 81.7 | 153.8 |
| vectara_factual_consistency | 93.9 | 91.4 |
| vectara_hallucination_rate ↓ | 6.1 | 8.6 |
| τ²-Bench Telecom (AA run) | 47.1 | 96.2 |
| τ³-Bench | 4.7 | 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.