DeepSeek V4 Pro 0813 ranks first on SWE-bench Verified Extended.
DeepSeek Aug 2026 Open weights Flagship
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Humanity's Last Exam | 60% | #14 of 154 | Lab model card | |
| SimpleQA Verified | 52.9% | #27 of 94 | Epoch AI run | |
| StudyArena Text | 1,243 | #29 of 85 | Official leaderboard | |
| BullshitBench | 32.7% | #46 of 107 | Official leaderboard | |
| AA-LCR Extended | 62.9% | #25 of 49 | Official leaderboard | |
| BullshitBench V2 | 35% | #71 of 129 | Official leaderboard | |
| OpenAI MRCR v2 | 38.9% | #28 of 47 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.6% | #29 of 212 | Epoch AI run | |
| GPQA Diamond | 91.7% | #32 of 229 | Epoch AI run | |
| DTBench | 93.9% | #31 of 180 | Official leaderboard | |
| ProofBench | 49% | #14 of 59 | Official leaderboard | |
| ARC-AGI-2 | 61.3% | #34 of 98 | Official leaderboard | |
| ARC-AGI | 90.5% | #34 of 95 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 64.6% | #39 of 86 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 26.8% | #40 of 68 | Official leaderboard | |
| ProofBench v1.1 | 50% | #31 of 52 | Official leaderboard | |
| FrontierChallenge | 13.4% | #7 of 11 | Official leaderboard | |
| Into the Unknown | 0% | #6 of 9 | Official leaderboard | |
| Prophet Arena | 84.5 | #23 of 24 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified Extended | 83.3% | #1 of 44 | Official leaderboard | |
| NL2Repo-Bench | 61.5% | #4 of 30 | Lab model card | |
| WeirdML | 66.2% | #24 of 112 | Official leaderboard | |
| Vibe Code Bench | 82.3% | #27 of 120 | Official leaderboard | |
| WebDev Arena | 1,581 | #24 of 105 | arena.ai, via Epoch AI | |
| ALE-Bench | 1,730 | #39 of 126 | Official leaderboard | |
| DSBench (DeepSeek internal) | 71.1% | #1 of 3 | Lab model card | |
| Code Migration | 41.5% | #29 of 78 | Official leaderboard | |
| APEX-SWE | 45.9% | #26 of 65 | Official leaderboard | |
| ReactBench | 33.9% | #12 of 28 | Official leaderboard | |
| Code Migration v1.1 | 52% | #15 of 30 | Official leaderboard | |
| IOI v2 (Vals) | 51.6% | #26 of 45 | Official leaderboard | |
| Vibe Code Bench 1-100 | 17.5% | #16 of 26 | Official leaderboard | |
| DeepSWE | 62.7% | #35 of 56 | Lab model card | |
| MirrorCode | 33.3% | #27 of 39 | Epoch AI run | |
| DAXBench v2 | 66.3% | #36 of 47 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Mystery Game Puzzles | 43% | #18 of 84 | Epoch AI run | |
| Terminal-Bench 2.1 | 87.9% | #12 of 56 | Lab model card | |
| Toolathlon-Verified | 74.4% | #6 of 25 | Official leaderboard | |
| Chess Puzzles | 46% | #27 of 93 | Epoch AI run | |
| BrowseComp Extended | 27% | #20 of 42 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 30% | #27 of 44 | Official leaderboard | |
| BabaIsBench | 75% | #21 of 32 | Official leaderboard | |
| AutomationBench | 31.8% | #31 of 47 | Lab model card | |
| WitnessBench | 5.50% | #8 of 12 | Official leaderboard | |
| DuelLab GameBench 2 | 53.6 | #46 of 57 | Official leaderboard | |
| Terminal-Bench 4.0 | 6.67% | #35 of 40 | Official leaderboard | |
| Terminal-Bench 4.1 | 6.67% | #38 of 43 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CorpFin v2 | 65.4% | #29 of 119 | Official leaderboard | |
| FoodTruck Bench | $37,350 | #18 of 59 | Official leaderboard | |
| YC-Bench | $1,459,743 | #20 of 59 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 7.50% | #28 of 80 | Official leaderboard | |
| APEX-1 | 66% | #25 of 71 | Official leaderboard | |
| Legal Research Bench | 40.9% | #35 of 93 | Official leaderboard | |
| TaxEval v2 | 73.1% | #53 of 129 | Official leaderboard | |
| TRLawBench | 9.23 | #15 of 34 | Official leaderboard | |
| Tax Agent Bench | 26.8% | #41 of 84 | Official leaderboard | |
| MedCode | 42.5% | #48 of 96 | Official leaderboard | |
| PolicyBench | 83% | #25 of 50 | Official leaderboard | |
| Vals Index v2 | 52.4% | #33 of 65 | Official leaderboard | |
| LegalBench | 82.4% | #71 of 138 | Official leaderboard | |
| Vals Finance Agent v2 | 50.4% | #44 of 80 | Official leaderboard | |
| MedScribe | 80.2% | #55 of 97 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 10.8% | #22 of 38 | Official leaderboard | |
| Vals Legal Index | 69% | #22 of 38 | Official leaderboard | |
| APEX-Agents | 47.3% | #39 of 61 | Collected by Epoch AI, source not named | |
| APEX-Agents 1.1 | 62.3% | #42 of 65 | Official leaderboard | |
| Excel Modeling Benchmark | 52.8% | #53 of 76 | Official leaderboard | |
| Vals Index v2.1 | 47.6% | #37 of 49 | Official leaderboard | |
| Vals Index v2.2 | 53% | #28 of 37 | Official leaderboard | |
| APEX-Accounting | 53.4% | #39 of 49 | Official leaderboard | |
| Vals Finance Agent v2.1 | 57.1% | #32 of 40 | Official leaderboard | |
| Agents' Last Exam | 25.7% | #21 of 26 | Lab model card | |
| Excel Modeling Benchmark v1.1 | 65.1% | #35 of 42 | Official leaderboard | |
| Vals Finance Index | 52.9% | #34 of 40 | Official leaderboard | |
| E-Bench (Tencent) | 61.3% | #7 of 8 | Official leaderboard | |
| Immigration Bench | 61.7% | #23 of 25 | Official leaderboard | |
| Thrive Holdings helpdesk ticket resolution benchmark | 32.7% | #14 of 15 | Official leaderboard | |
| Bedside Bench | 80.6% | #17 of 17 | Official leaderboard | |
| Holdings-IndividualTaxBench | 64.4% | #13 of 13 | Official leaderboard | |
| E-Bench-Code (Tencent) | 64.3% | #8 of 8 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| VoxelBench | 1,624 | #31 of 56 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 84.1 | #47 of 185 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 72.9% | #12 of 44 | Official leaderboard | |
| Molecular Biology Capabilities Test | 53% | #26 of 90 | Official leaderboard | |
| CyberGym | 83.3% | #10 of 34 | Lab model card | |
| Human Pathogen Capabilities Test | 58% | #29 of 86 | Official leaderboard | |
| BioTIER Permit | 99.7% | #26 of 69 | Official leaderboard | |
| World-Class Bio | 44.7% | #39 of 74 | Official leaderboard | |
| SRE-Bench | 1.91% | #24 of 35 | Official leaderboard | |
| BioTIER Refuse | 17.6% | #61 of 69 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 59.3 | #39 of 151 | Official leaderboard | |
| LMCA | 45.5% | #41 of 147 | Official leaderboard | |
| ACCoRD | 70.2 | #54 of 152 | Official leaderboard |
The Self-Reported Index fits one score to the 7 capability results DeepSeek reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 156.8 to 163.0. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| Toolathlon-Verified | 74.1% Toolathlon-Verified | 74.4% Leaderboard, max | −0.3 pts |
DeepSeek's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
The benchmarks the lab chose to headline, with the scores it gave.