| Benchmark | DeepSeek-R1 | DeepSeek-V4-Flash |
|---|---|---|
| AA Agentic Index | 3.1 | 61.3 |
| AA Intelligence | 18.6 | 51.8 |
| AA-LCR | 56 | 74.3 |
| AA-Omniscience | -31.3 | -14.3 |
| arc_agi_1 | 21.2 | 84 |
| ARC-AGI-2 | 1.3 | 46 |
| Artificial Analysis Coding Index | 24.6 | 69.1 |
| browsecomp | 8.9 | 73.2 |
| Chinese SimpleQA (C-SimpleQA) | 63.7 | 78.9 |
| Codeforces | 2029 | 3052 |
| critpt | 0.6 | 16.6 |
| gdpval | 1.5 | 52.9 |
| GPQA Diamond | 81 | 90.8 |
| HLE | 17.7 | 38.6 |
| IFBench | 39 | 79.2 |
| LiveCodeBench | 84.4 | 91.6 |
| MMLU-Pro | 85 | 86.4 |
| OmniScience Accuracy | 30.7 | 40.4 |
| OmniScience Non-Hallucination | 10.5 | 15.6 |
| scicode | 35.7 | 49.9 |
| simplebench | 40.8 | 46.3 |
| simpleqa_verified | 27.4 | 45 |
| SWE-bench Multilingual | 30.5 | 73.3 |
| SWE-bench Verified | 57.6 | 79 |
| TauBench V3 - Banking | 6.4 | 39.4 |
| Terminal-Bench 2.1 | 19.1 | 82.7 |
| Terminal-Bench Hard | 6.1 | 38.6 |
| τ²-Bench Telecom (AA run) | 11.4 | 95.6 |
| τ³-Bench | 6.4 | 22.9 |
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.