DeepSeek · model
DeepSeek V4 Pro benchmark scores
As of 2026-11-19, DeepSeek V4 Pro (DeepSeek) has measured scores on 50 benchmarks and estimated scores on 112 more.
| Benchmark | Score | Source |
|---|---|---|
| AA-GPQA Diamond | 92.8% | measured |
| AA-HLE | 41.0% | measured |
| AA-SciCode | 51.0% | measured |
| AIME26 | 97.7% | estimated ± 1.2 pp, medium confidence |
| Artificial Analysis Intelligence Index | 53.2% | measured |
| CritPt | 18.0% | measured |
| GDPval-AA | 54.5% | measured |
| GPQA Diamond | 90.1% | measured |
| HLE | 42.7% | measured |
| IFBench | 74.9% | estimated ± 6.6 pp, medium confidence |
| AA-LCR | 80.3% | measured |
| LiveCodeBench v6 | 87.8% | estimated ± 1.7 pp, medium confidence |
| AA-Omniscience Accuracy | 49.1% | measured |
| SciCode | 51.7% | estimated ± 2.9 pp, medium confidence |
| Terminal-Bench 2.1 | 87.9% | measured |
| BrowseComp | 83.4% | measured |
| HLE w/ tools | 60.0% | measured |
| IFEval | 93.1% | estimated ± 2.8 pp, high confidence |
| LongBench v2 | 65.9% | estimated ± 5.5 pp, medium confidence |
| VITA-Bench | 45.8% | estimated ± 6.7 pp, low confidence |
| τ²-bench results | 96.2% | measured |
| AA Coding Index | 68.8% | measured |
| APEX-Agents | 27.7% | estimated ± 4.9 pp, low confidence |
| APEX-Agents-AA | 24.3% | measured |
| BioMysteryBench (human-difficult) | 47.6% | estimated ± 6.8 pp, low confidence |
| DeepSearchQA | 93.3% | estimated ± 4.5 pp, high confidence |
| SWE-bench Verified | 80.6% | measured |
| AutomationBench | 31.8% | measured |
| BFCL v4 | 77.0% | estimated ± 2.7 pp, medium confidence |
| CyberGym | 83.3% | measured |
| JobBench | 37.4% | estimated ± 7.3 pp, low confidence |
| SkillsBench | 55.2% | estimated ± 4.1 pp, medium confidence |
| SWE-bench Pro | 55.4% | measured |
| WideResearch | 83.0% | estimated ± 1.6 pp, medium confidence |
| AA Agentic Index | 49.6% | measured |
| MMLU-Pro | 87.5% | measured |
| AA-IFBench | 76.5% | measured |
| BBH | 100.0% | estimated ± 2.8 pp, medium confidence |
| GPQA | 90.1% | measured |
| MMLU | 90.5% | estimated ± 0.9 pp, medium confidence |
| Vals GPQA Diamond | 92.4% | measured |
| Vals LiveCodeBench | 87.5% | measured |
| Vals MMLU-Pro | 87.0% | measured |
| MMMLU | 90.6% | estimated ± 0.7 pp, low confidence |
| Chess Puzzles | 41.8% | estimated ± 10.8 pp, low confidence |
| AIME 2025 | 98.7% | estimated ± 3.8 pp, low confidence |
| AA MMLU-Pro | 88.7% | estimated ± 0.7 pp, medium confidence |
| Gert Labs | 69.2% | estimated ± 5.5 pp, low confidence |
| AA-AnalystAgent | 43.0% | estimated ± 8.0 pp, low confidence |
| AA EnterpriseOps-Gym | 49.6% | measured |
| AA Harvey LAB v1.0 | 93.4% | estimated ± 1.2 pp, medium confidence |
| AA Tau3 Banking | 39.6% | measured |
| AA Terminal-Bench 2.1 | 80.0% | estimated ± 2.5 pp, medium confidence |
| ApprenticeBench | 22.6% | estimated ± 4.0 pp, high confidence |
| ARC-AGI-1 | 90.0% | measured |
| ARC-AGI-2 | 61.3% | measured |
| CursorBench 3.1 | 46.5% | estimated ± 3.4 pp, low confidence |
| CursorBench 3.2 | 51.1% | estimated ± 3.6 pp, high confidence |
| EBR-bench | 34.7% | estimated ± 11.1 pp, low confidence |
| Terminal-Bench 3.0 | 11.5% | estimated ± 3.2 pp, medium confidence |
| FrontierCode 1.1 Main | 45.8% | estimated ± 3.5 pp, low confidence |
| FrontierMath v2 (Tier 4) | 26.8% | measured |
| FrontierMath v2 (Tiers 1-3) | 64.6% | measured |
| FrontierSWE v2 | 7.6% | estimated ± 8.3 pp, medium confidence |
| Furniture Assembly | 38.2% | estimated ± 12.4 pp, low confidence |
| MaxxBench | 51.9% | estimated ± 5.7 pp, medium confidence |
| MazeBench | 2.7% | estimated ± 4.0 pp, high confidence |
| MirrorCode | 33.3% | measured |
| Mystery Game Puzzles | 29.0% | estimated ± 13.3 pp, low confidence |
| OSWorld-Verified | 79.7% | estimated ± 4.7 pp, high confidence |
| Terminal-Bench 4.1 | 6.7% | measured |
| Terminal-Bench Hard | 54.3% | estimated ± 2.8 pp, medium confidence |
| Vals SWE-bench | 96.4% | measured |
| Terminal-Bench 2.1 (Vals AI) | 54.7% | measured |
| VulcanBench v3 | 85.4% | estimated ± 5.6 pp, low confidence |
| CWE-Bench-AA | 59.2% | estimated ± 5.2 pp, medium confidence |
| AA Cyber Index | 29.0% | estimated ± 9.2 pp, low confidence |
| DeepsecBench-AA | 19.2% | estimated ± 5.2 pp, low confidence |
| HealthBench Professional (length-adjusted, Artificial Analysis) | 62.1% | estimated ± 4.3 pp, medium confidence |
| HealthBench Professional (raw, Artificial Analysis) | 64.2% | estimated ± 0.8 pp, low confidence |
| AA ITBench | 48.7% | estimated ± 4.0 pp, high confidence |
| Bug Hunt Bench | 27.2% | estimated ± 7.2 pp, low confidence |
| CursorBench 4.0 | 30.3% | estimated ± 2.0 pp, medium confidence |
| CWE-bench v1 | 58.7% | estimated ± 6.4 pp, medium confidence |
| DeepSWE | 62.7% | measured |
| DRACO | 85.2% | estimated ± 2.0 pp, medium confidence |
| FrontierCode 2.0 | 19.9% | estimated ± 6.7 pp, low confidence |
| FrontierPhysics | 2.5% | estimated ± 4.5 pp, high confidence |
| HLE w/o tools | 47.0% | estimated ± 2.7 pp, medium confidence |
| OSWorld 2.0 | 19.4% | estimated ± 5.5 pp, low confidence |
| OSWorld 2.1 | 69.7% | estimated ± 6.0 pp, medium confidence |
| OSWorld 2.1 Offline | 70.0% | estimated ± 1.9 pp, medium confidence |
| PostTrainBench v1.1 | 35.1% | estimated ± 3.7 pp, high confidence |
| StealthMark | 42.8% | measured |
| SWE Multilingual | 76.2% | measured |
| Terminal-Bench 4.0 | 9.3% | estimated ± 4.7 pp, medium confidence |
| Toolathlon-Verified | 74.1% | measured |
| VulcanBench Frontier v4 | 81.6% | estimated ± 4.0 pp, low confidence |
| AA AutomationBench | 42.4% | estimated ± 7.9 pp, low confidence |
| GDP.pdf | 17.0% | estimated ± 4.6 pp, high confidence |
| AA Terminal-Bench 4.0 | 6.4% | estimated ± 6.3 pp, medium confidence |
| ARC-AGI-3 | 0.0% | estimated ± 8.3 pp, low confidence |
| ARC-AGI-3 Provider Adapter | 2.7% | estimated ± 7.3 pp, medium confidence |
| CWE-bench | 51.7% | estimated ± 11.0 pp, low confidence |
| MKA | 41.1% | estimated ± 6.4 pp, low confidence |
| Vibe Code Bench | 49.9% | measured |
| FACTS Multimodal | 41.7% | estimated ± 4.7 pp, high confidence |
| Hermes Bench | 70.1% | estimated ± 1.5 pp, medium confidence |
| Hermes Index | 36.1% | estimated ± 7.0 pp, low confidence |
| ExploitBench | 31.9% | estimated ± 12.6 pp, low confidence |
| ExploitGym | 11.0% | estimated ± 9.2 pp, low confidence |
| C-Eval | 92.8% | estimated ± 1.2 pp, high confidence |
| Claw-Eval | 69.9% | estimated ± 6.3 pp, low confidence |
| DeepPlanning | 47.0% | estimated ± 13.1 pp, low confidence |
| HMMT Feb 2026 | 95.2% | measured |
| HMMT Nov 2025 | 95.2% | estimated ± 1.5 pp, low confidence |
| INCLUDE | 85.4% | estimated ± 2.0 pp, medium confidence |
| MCP Atlas | 73.6% | measured |
| MCP-Tasks | 77.5% | estimated ± 7.0 pp, low confidence |
| MMAnswerBench | 90.3% | estimated ± 2.4 pp, low confidence |
| MMLU-ProX | 85.3% | estimated ± 1.3 pp, high confidence |
| MMLU-Redux | 94.5% | estimated ± 0.6 pp, high confidence |
| NL2Repo | 61.5% | measured |
| PostTrainBench v1 | 19.8% | estimated ± 3.1 pp, medium confidence |
| QwenClawBench | 61.5% | estimated ± 2.6 pp, high confidence |
| SuperGPQA | 82.4% | estimated ± 7.2 pp, medium confidence |
| Terminal-Bench 2.0 | 67.9% | measured |
| Toolathlon | 51.8% | measured |
| AA LiveCodeBench | 100.0% | estimated ± 2.6 pp, medium confidence |
| HealthBench Hard | 2.4% | estimated ± 5.1 pp, low confidence |
| LiveCodeBench Pro | 80.0% | estimated ± 6.0 pp, low confidence |
| MedXpertQA (Text) | 54.1% | estimated ± 4.5 pp, medium confidence |
| MMLU-Pro (Arcee) | 86.9% | estimated ± 3.4 pp, low confidence |
| React Native Evals | 79.9% | estimated ± 2.7 pp, medium confidence |
| ResearchClawBench | 18.7% | estimated ± 1.8 pp, high confidence |
| SWE-Rebench | 61.8% | estimated ± 3.6 pp, medium confidence |
| PostTrainBench | 34.8% | estimated ± 3.7 pp, high confidence |
| DAYJOB | 2.3% | estimated ± 3.2 pp, high confidence |
| DAYJOB Healthcare | 1.1% | estimated ± 4.7 pp, high confidence |
| BioMysteryBench (human-solvable) | 88.8% | estimated ± 0.6 pp, medium confidence |
| FrontierCode 1.1 Extended | 61.1% | estimated ± 2.4 pp, medium confidence |
| HealthBench (raw) | 49.6% | estimated ± 6.0 pp, medium confidence |
| HealthBench (length-adjusted) | 55.1% | estimated ± 2.5 pp, low confidence |
| HealthBench Professional | 51.3% | estimated ± 4.2 pp, low confidence |
| HealthBench Professional (raw) | 60.2% | estimated ± 4.5 pp, high confidence |
| HLE-Verified | 52.7% | estimated ± 2.0 pp, medium confidence |
| LABBench2 | 81.5% | estimated ± 1.8 pp, low confidence |
| ScarfBench | 63.8% | estimated ± 3.8 pp, medium confidence |
| STEMBench | 2.6% | estimated ± 0.2 pp, medium confidence |
| SimpleQA | 35.0% | estimated ± 7.8 pp, low confidence |
| LiveCodeBench | 79.6% | estimated ± 14.6 pp, low confidence |
| Agents' Last Exam | 25.7% | measured |
| OpenHarmony Bench v1.0 | 59.0% | measured |
| SEC-Bench Pro | 50.2% | estimated ± 9.3 pp, medium confidence |
| IMOAnswerBench | 89.8% | measured |
| MRCRv2 | 85.8% | estimated ± 11.3 pp, low confidence |
| MMLongBench-Doc | 61.4% | estimated ± 2.6 pp, low confidence |
| BioBasicsBench | 22.0% | estimated ± 2.2 pp, low confidence |
| LifeVal | 74.8% | estimated ± 1.6 pp, medium confidence |
| ExploitBench Internal Port | 0.0% | estimated ± 6.8 pp, low confidence |
| PinchBench | 90.4% | estimated ± 3.6 pp, low confidence |
| SWE-Atlas Codebase QnA | 57.9% | estimated ± 4.8 pp, high confidence |