DeepSeek · model
DeepSeek V4.1 Flash benchmark scores
As of 2026-10-07, DeepSeek V4.1 Flash (DeepSeek) has measured scores on 30 benchmarks and estimated scores on 79 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 88.7% | estimated ± 3.4 pp, high confidence |
| AA-HLE | 39.2% | measured |
| AA-SciCode | 51.9% | measured |
| Artificial Analysis Intelligence Index | 39.5% | measured |
| CritPt | 14.3% | measured |
| GDPval-AA | 55.0% | measured |
| GPQA Diamond | 90.9% | measured |
| HLE | 36.8% | measured |
| AA-LCR | 84.0% | measured |
| LiveCodeBench v6 | 93.1% | estimated ± 3.3 pp, medium confidence |
| AA-Omniscience Accuracy | 46.4% | measured |
| SciCode | 51.5% | estimated ± 3.2 pp, high confidence |
| Terminal-Bench 2.1 | 90.6% | measured |
| BrowseComp | 85.1% | estimated ± 1.8 pp, medium confidence |
| HLE w/ tools | 63.9% | measured |
| LongBench v2 | 61.5% | estimated ± 5.9 pp, low confidence |
| τ²-bench results | 89.0% | estimated ± 4.8 pp, medium confidence |
| AA Coding Index | 77.5% | estimated ± 2.5 pp, high confidence |
| AA-MMMU-Pro | 77.0% | measured |
| APEX-Agents | 44.3% | estimated ± 10.0 pp, low confidence |
| APEX-Agents-AA | 39.2% | estimated ± 3.2 pp, medium confidence |
| BioMysteryBench (human-difficult) | 44.1% | estimated ± 5.9 pp, low confidence |
| DeepSearchQA | 90.4% | estimated ± 8.4 pp, medium confidence |
| SWE-bench Verified | 84.1% | estimated ± 2.4 pp, medium confidence |
| AutomationBench | 54.8% | measured |
| BFCL v4 | 77.8% | estimated ± 2.6 pp, medium confidence |
| CyberGym | 88.1% | measured |
| JobBench | 58.2% | estimated ± 8.6 pp, medium confidence |
| SWE-bench Pro | 62.8% | estimated ± 3.6 pp, medium confidence |
| WideResearch | 86.2% | estimated ± 3.4 pp, low confidence |
| JevBench 1.4 | 4.8% | measured |
| AA Agentic Index | 51.0% | estimated ± 1.7 pp, high confidence |
| MMLU-Pro | 86.5% | estimated ± 3.0 pp, high confidence |
| GPQA | 90.9% | measured |
| Gert Labs | 70.9% | estimated ± 6.0 pp, low confidence |
| AA-AnalystAgent | 55.6% | estimated ± 2.5 pp, medium confidence |
| AA EnterpriseOps-Gym | 37.5% | estimated ± 4.4 pp, high confidence |
| AA Harvey LAB | 92.0% | estimated ± 1.7 pp, medium confidence |
| ApprenticeBench | 31.0% | estimated ± 4.0 pp, low confidence |
| ARC-AGI-1 | 90.4% | estimated ± 2.7 pp, high confidence |
| ARC-AGI-2 | 59.8% | estimated ± 10.1 pp, low confidence |
| CursorBench 3.1 | 48.7% | estimated ± 3.4 pp, medium confidence |
| CursorBench 3.2 | 21.8% | estimated ± 3.6 pp, low confidence |
| Terminal-Bench 3.0 | 18.5% | estimated ± 6.9 pp, medium confidence |
| FrontierCode 1.1 Main | 53.5% | estimated ± 3.7 pp, low confidence |
| FrontierSWE v2 | 17.5% | estimated ± 14.6 pp, low confidence |
| OfficeQA Pro | 50.3% | estimated ± 7.2 pp, medium confidence |
| OSWorld-Verified | 80.7% | estimated ± 4.7 pp, high confidence |
| Vals GPQA Diamond | 90.2% | estimated ± 2.0 pp, high confidence |
| Vals LiveCodeBench | 77.4% | estimated ± 6.1 pp, low confidence |
| Vals MMLU-Pro | 87.4% | estimated ± 1.2 pp, high confidence |
| Vals SWE-bench | 97.3% | estimated ± 5.4 pp, low confidence |
| Terminal-Bench 2.1 (Vals AI) | 74.5% | measured |
| AA AutomationBench | 68.9% | measured |
| GDP.pdf | 12.8% | measured |
| AA ITBench | 46.9% | measured |
| AA Tau3 Banking | 46.8% | estimated ± 3.0 pp, low confidence |
| AA Terminal-Bench 2.1 | 86.5% | estimated ± 2.6 pp, high confidence |
| AA Terminal-Bench 4.0 | 26.8% | measured |
| CursorBench 4.0 | 32.4% | estimated ± 3.5 pp, medium confidence |
| CWE-bench v1 | 55.0% | measured |
| DeepSWE | 74.2% | measured |
| HLE w/o tools | 35.1% | estimated ± 2.9 pp, high confidence |
| OSWorld 2.0 | 61.0% | estimated ± 13.3 pp, low confidence |
| PostTrainBench v1.1 | 42.8% | estimated ± 4.8 pp, high confidence |
| ProgramBench | 20.3% | measured |
| Terminal-Bench 4.0 | 31.2% | measured |
| Terminal-Bench-Science 0.1 | 4.7% | estimated ± 4.1 pp, low confidence |
| Toolathlon-Verified | 74.4% | estimated ± 2.2 pp, medium confidence |
| CharXiv | 84.4% | estimated ± 7.5 pp, low confidence |
| CharXiv w/o tools | 80.8% | estimated ± 2.4 pp, high confidence |
| ExploitBench | 52.5% | estimated ± 12.2 pp, low confidence |
| ExploitGym | 15.3% | measured |
| C-Eval | 93.2% | estimated ± 0.8 pp, low confidence |
| MathVision | 89.4% | estimated ± 4.9 pp, high confidence |
| MCP Atlas | 87.4% | estimated ± 2.7 pp, low confidence |
| MMLU-Redux | 97.1% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro | 77.9% | estimated ± 4.4 pp, medium confidence |
| NL2Repo | 65.4% | measured |
| QwenClawBench | 100.0% | estimated ± 6.0 pp, low confidence |
| ScreenSpot Pro | 75.8% | estimated ± 10.7 pp, low confidence |
| SuperGPQA | 84.2% | estimated ± 7.5 pp, low confidence |
| Terminal-Bench 2.0 | 76.4% | estimated ± 5.9 pp, medium confidence |
| Toolathlon | 55.9% | estimated ± 6.3 pp, medium confidence |
| ERQA | 66.6% | estimated ± 3.9 pp, high confidence |
| HealthBench Hard | 9.5% | estimated ± 5.1 pp, low confidence |
| MedXpertQA (MM) | 75.0% | estimated ± 6.5 pp, low confidence |
| MedXpertQA (Text) | 56.5% | estimated ± 4.5 pp, medium confidence |
| MMLU-Pro (Arcee) | 87.2% | estimated ± 3.4 pp, low confidence |
| React Native Evals | 77.5% | estimated ± 4.7 pp, medium confidence |
| ResearchClawBench | 19.8% | estimated ± 2.2 pp, low confidence |
| ARC-AGI-3 | 0.0% | estimated ± 2.1 pp, medium confidence |
| BioMysteryBench (human-solvable) | 87.1% | estimated ± 0.6 pp, medium confidence |
| DRACO | 84.1% | estimated ± 3.0 pp, medium confidence |
| HealthBench (raw) | 48.3% | estimated ± 6.4 pp, low confidence |
| HealthBench (length-adjusted) | 53.9% | estimated ± 3.7 pp, medium confidence |
| HealthBench Professional | 51.9% | estimated ± 4.8 pp, low confidence |
| HealthBench Professional (raw) | 58.7% | estimated ± 5.0 pp, medium confidence |
| HLE-Verified | 41.3% | estimated ± 6.0 pp, low confidence |
| LABBench2 | 79.5% | estimated ± 2.4 pp, low confidence |
| MMMU | 84.7% | estimated ± 0.9 pp, high confidence |
| Agents' Last Exam | 31.8% | measured |
| OpenHarmony Bench | 60.3% | measured |
| SEC-Bench Pro | 62.8% | measured |
| MMMLU | 90.1% | estimated ± 1.1 pp, medium confidence |
| MMLU | 94.4% | estimated ± 1.1 pp, low confidence |
| MMMU-Pro w/ Python | 79.9% | estimated ± 1.6 pp, high confidence |
| RealWorldQA | 86.0% | estimated ± 1.2 pp, medium confidence |
| OmniDocBench 1.5 | 91.9% | estimated ± 6.4 pp, low confidence |