Benchmark Almanac

DeepSeek-V4-Pro

DeepSeek-V4-Pro ranks first on 5 of the 271 benchmarks with five or more models scored.

DeepSeek Released April 24th, 2026 Family DeepSeek-V4 The lab's model card Compare with others
283
Benchmarks with a score
5
Ranked first, out of five or more models
19
In the top three
DeepSeek Apr 2026

DeepSeek-V4-Pro

Open weights Flagship

150.2 #69 of 256
Epoch Capabilities Index
Self-Reported Index 152.7 #77 of 404
  • SWE-bench Verified 77.6% #6 /32
  • DeepSWE 62.8% #14 /28
  • Agents' Last Exam 12.4% #23 /31

Where it ranks

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.

Knowledge and language

BenchmarkScoreRankSource
AgentIF (2026-10 setting) 67.4% #1 of 24 Official leaderboard
MMLU-Pro 87.5% #7 of 167 Lab model card
C-Eval 93.1% #3 of 46 Lab model card
EQ-Bench 3 1,555 #7 of 78 Official leaderboard
MMLU 90.1% #13 of 131 Lab model card
MMMLU 90.3% #8 of 71 Lab model card
CMMLU 90.8% #2 of 17 Lab model card
LMArena (Elo) 1,464 #47 of 335 Official leaderboard
AGIEval 83.1% #6 of 40 Lab model card
Chinese SimpleQA 84.4% #2 of 13 Lab model card
BIG-Bench Hard 87.5% #14 of 73 Lab model card
CulturalRiddles 63.8% #4 of 20 Official leaderboard
Extended NYT Connections 91.3% #21 of 103 Official leaderboard
Longform Creative Writing (EQ-Bench) 75.6 #26 of 119 Official leaderboard
DROP 88.7% #9 of 41 Lab model card
HUMAINE 31 #11 of 50 Official leaderboard
TriviaQA 85.6% #10 of 40 Lab model card
HELM Arabic Enterprise 72.8 #6 of 23 Official leaderboard
HellaSwag 88% #14 of 53 Lab model card
Creative Writing v3 75.0 #32 of 120 Official leaderboard
MRCR v2 83.5% #8 of 30 Lab model card
Wer wird Millionär? LLM Benchmark 73489.0 #23 of 77 Official leaderboard
Humanity's Last Exam 48.2% #48 of 154 Lab model card
Hemingway-bench 1,046 #14 of 42 Official leaderboard
MultiLoKo 51.1% #1 of 3 Lab model card
WinoGrande 81.5% #16 of 47 Lab model card
SEA-HELM 72.6 #25 of 71 Official leaderboard
EvoArena PersonaMem-Evo 51.6% #3 of 8 Official leaderboard
Lech Mazur Creative Writing 0.44 #20 of 50 Official leaderboard
CompassArena 1,020 #4 of 10 Official leaderboard
SimpleQA Verified 47% #39 of 94 Epoch AI run
Vectara Hallucination Leaderboard (2025 dataset) 8.60% #44 of 104 Official leaderboard
Last Translation Benchmark 10.8 #16 of 36 Official leaderboard
MGSM 84.4% #28 of 57 Lab model card
CL-bench Life 13.5% #8 of 16 Official leaderboard
CorpusQA 1M 62% #1 of 2 Lab model card
Structured Output Benchmark 85.3% #10 of 19 Official leaderboard
Antidote 1,018 #19 of 36 Official leaderboard
MMLU-Redux 90.8% #25 of 45 Lab model card
CLUEWSC 85.2% #5 of 8 Lab model card
EvoArena SWE-Chain-Evo 27.7% #5 of 8 Official leaderboard
DesignArena ASCII Art 1,143 #49 of 77 Official leaderboard
Continual Learning Exam 27.3% #7 of 11 Official leaderboard
LongBench v2 51.5% #23 of 35 Lab model card
OpenAI MRCR v2 33.2% #31 of 47 Official leaderboard
Judgemark v4 47.1 #27 of 40 Official leaderboard
EQ-Bench 4 1,156 #22 of 32 Official leaderboard
SuperGPQA 53.9% #25 of 35 Lab model card
BullshitBench V2 14% #107 of 129 Official leaderboard
BullshitBench 14.6% #90 of 107 Official leaderboard
Phare 58.9% #53 of 62 Official leaderboard
EvoArena Terminal-Bench-Evo 40.4% #7 of 8 Official leaderboard
AA-LCR Extended 37.6% #43 of 49 Official leaderboard
HLE-Diamond 16.6% #20 of 20 Official leaderboard
DeepSeek Chinese functional writing evaluation 62.6% #1 of 1 Lab model card
DeepSeek complex instruction following and multi-turn writing evaluation 45.9% #1 of 1 Lab model card
DeepSeek search Q&A evaluation 28.1% #1 of 1 Lab model card
FACTS Parametric 62.6% #1 of 1 Lab model card

Math and science

BenchmarkScoreRankSource
ResearchClawBench 38.6 #1 of 23 Official leaderboard
Echo General AI Prediction Leaderboard 1,017 #1 of 17 Official leaderboard
NatureBench 25.6% #2 of 18 Official leaderboard
IMO-AnswerBench 89.8% #5 of 37 Lab model card
MathArena Apex Shortlist 87.8% #6 of 38 Official leaderboard
GPQA Diamond 90.9% #37 of 229 Epoch AI run
OTIS Mock AIME 2024-2025 96.7% #41 of 212 Epoch AI run
MathArena (Apex, IMO, USAMO) 28.1% #10 of 47 Official leaderboard
DTBench 90.7% #46 of 180 Official leaderboard
LiveMathematicianBench 30.4% #5 of 17 Official leaderboard
GSM8K 92.6% #28 of 92 Lab model card
AIME 2026 96.7% #10 of 31 Official leaderboard
CMATH 90.9% #2 of 6 Lab model card
HMMT 2026 93.9% #12 of 31 Official leaderboard
USAMO 2026 60.7% #4 of 9 Official leaderboard
Putnam 2025 120.0 #1 of 2 Lab model card
Verifying the Verifiers (Project APE CRED benchmark) 83.5% #12 of 23 Official leaderboard
MATH 64.5% #53 of 99 Lab model card
SimpleBench 50.9% #59 of 106 Official leaderboard
HorizonMath 2.70% #5 of 8 Official leaderboard
FrontierMath Tiers 1-3 v2 45.3% #56 of 86 Official leaderboard
ProofBench 10% #43 of 59 Official leaderboard
xbench-ScienceQA 69.4% #22 of 29 Official leaderboard
Surface Evolver Bench 40% #20 of 25 Official leaderboard
ForecastBench 0.15 #67 of 82 Official leaderboard
Leipzig Benchmark 10% #16 of 19 Official leaderboard
ProofBench v1.1 16% #45 of 52 Official leaderboard
FrontierPhysics 0.60% #20 of 22 Official leaderboard
ASI-Bench 25.0 #10 of 11 Official leaderboard
FrontierMath Tier 4 v2 2.44% #64 of 68 Official leaderboard
Terminal-Bench-Science 3.81% #16 of 17 Official leaderboard
ExplorationBench 12.9% #10 of 10 Official leaderboard

Coding

BenchmarkScoreRankSource
LiveCodeBench v6 93.5% #1 of 90 Lab model card
LiveCodeBench 93.5% #1 of 54 Lab model card
Codeforces rating 3,206 #3 of 47 Lab model card
SWE-bench Verified 77.6% #6 of 32 Epoch AI run
BigCodeBench 59.2% #3 of 14 Lab model card
IOI (Vals) 35.8% #15 of 55 Official leaderboard
Code Arena 1,581 #39 of 128 Official leaderboard
DesignArena 3D Design 1,235 #45 of 136 Official leaderboard
ContextBench 41.1% #3 of 9 Official leaderboard
QwenWebDev 1,570 #2 of 6 Official leaderboard
SWE-bench Multilingual 76.2% #28 of 73 Lab model card
DesignArena Game Dev 1,222 #55 of 143 Official leaderboard
EvoCode-Bench v2 14.1% #5 of 13 Official leaderboard
SWE-bench Science 39.5% #4 of 10 Official leaderboard
Monthly-SWEBench 57% #5 of 12 Official leaderboard
DesignArena Code 1,231 #60 of 143 Official leaderboard
WeirdML 48.6% #49 of 112 Official leaderboard
FeatureBench 59.6% #7 of 16 Official leaderboard
FeatureBench Lite 59.6% #7 of 16 Official leaderboard
DesignArena Website 1,228 #68 of 151 Official leaderboard
ALE-Bench 1,489 #58 of 126 Official leaderboard
EvoCode-Bench 30.6% #6 of 13 Official leaderboard
WebDev Arena 1,464 #51 of 105 arena.ai, via Epoch AI
DesignArena UI Component 1,212 #69 of 139 Official leaderboard
DeepSWE 62.8% #14 of 28 Official leaderboard
Claw-SWE-Bench 71.7% #8 of 16 Official leaderboard
Tencent WorkBuddy Bench 75.6% #7 of 14 Official leaderboard
Vibe Code Bench 49.9% #64 of 120 Official leaderboard
RoadmapBench 18.3% #6 of 11 Official leaderboard
DesignArena Data Visualization 1,201 #67 of 118 Official leaderboard
Code Migration 26.2% #46 of 78 Official leaderboard
HumanEval 76.8% #60 of 101 Lab model card
AutoLab 0.38 #6 of 10 Official leaderboard
SWE-Bench Pro 55.4% #50 of 81 Lab model card
DesignArena SVG 1,129 #67 of 103 Official leaderboard
SWE Atlas Codebase QnA Extended 5.30% #26 of 38 Official leaderboard
QwenSVG 1,506 #5 of 7 Official leaderboard
SWE Atlas 27.1% #16 of 22 Official leaderboard
AndroidBench 59.5% #24 of 32 Official leaderboard
SWE-Marathon 4.00% #9 of 12 Official leaderboard
APEX-SWE 33.8% #49 of 65 Official leaderboard
SWE-rebench 40.2% #10 of 13 Official leaderboard
FrogsGame post-training task 0.40% #14 of 17 Official leaderboard
FrontierSWE v1 24.6% #14 of 17 Official leaderboard
HarnessDev 48.8% #5 of 6 Official leaderboard
TestPrism 13.3% #5 of 6 Official leaderboard
Doc2Harness 31.9% #16 of 19 Official leaderboard
Snorkel Agentic Coding 2.0 14.8% #14 of 16 Official leaderboard
SWE-bench CLI 2.30% #14 of 16 Official leaderboard
OpenHands Index 40.7% #29 of 33 Official leaderboard
RSI-Exam 32.3% #16 of 18 Official leaderboard
SWE-bench Verified Extended 22% #40 of 44 Official leaderboard
SWE Atlas Test Writing 27.1% #20 of 22 Official leaderboard
DesignArena Agentic Overall 919 #71 of 74 Official leaderboard
SWE-Together 48% #22 of 22 Official leaderboard
SWE-Marathon v1.1 10.6% #20 of 20 Official leaderboard
GameCraft-Bench 2.15% #15 of 15 Official leaderboard
MLS-Bench 24.4% #15 of 15 Official leaderboard
EvoPolicyGym 0.36 #4 of 4 Official leaderboard
DeepSeek R&D Coding Benchmark 67% #1 of 1 Lab model card

Agents and tools

BenchmarkScoreRankSource
Toolathlon 52.8% #5 of 50 Official leaderboard
PACT 1,570 #3 of 26 Official leaderboard
Terminal-Bench 2.0 67.9% #11 of 63 Lab model card
VitaBench 2.0 47.2% #3 of 17 Official leaderboard
ClawProBench 64.4% #10 of 49 Official leaderboard
GAIA 74.8% #14 of 64 Official leaderboard
LisanBench 3256.3 #35 of 133 Official leaderboard
BrowseComp 83.4% #24 of 81 Lab model card
Clocktower Radio 1,661 #8 of 23 Official leaderboard
Long-Horizon-Terminal-Bench 2.0 52.9% #6 of 15 Official leaderboard
Claw-Eval Pass^3 59.8% #10 of 24 Official leaderboard
Claw-Eval 77.2% #11 of 24 Official leaderboard
SkillsBench 50.1% #12 of 22 Official leaderboard
TERMS-Bench 0.62 #6 of 11 Official leaderboard
DeepTerminalBench 32% #5 of 9 Official leaderboard
ClawBench V2 33.9% #4 of 7 Official leaderboard
TUA-Bench 46.2% #7 of 12 Official leaderboard
MCP Atlas 73.6% #27 of 46 Lab model card
RuneBench 4.30 #49 of 76 Official leaderboard
WildClawBench 43.7% #20 of 31 Official leaderboard
clembench 59.5 #4 of 6 Official leaderboard
Long-Horizon-Terminal-Bench (LHTB) 30.7% #17 of 25 Official leaderboard
Toolathlon-Verified 55.9% #17 of 25 Official leaderboard
LiveClawBench 65.6% #12 of 17 Official leaderboard
CLASHarness 19.8% #10 of 14 Official leaderboard
PinchBench 2.0 61.1% #41 of 57 Official leaderboard
Mystery Game Puzzles 17% #61 of 84 Epoch AI run
Harbor-Index 4.90% #7 of 9 Official leaderboard
TraceTrial-Bench 72% #7 of 9 Official leaderboard
Chess Puzzles 19% #74 of 93 Epoch AI run
Bench on the Clocktower 48% #8 of 10 Official leaderboard
Terminal-Bench 2.1 Extended 21.2% #36 of 44 Official leaderboard
BrowseComp Extended 8.30% #36 of 42 Official leaderboard
Collaborative Intelligence Benchmark 39.4% #16 of 18 Official leaderboard
Learn2Play Bench 54.1 #10 of 11 Official leaderboard
BabaIsBench 29% #30 of 32 Official leaderboard
Pencil Puzzle Bench 0% #52 of 53 Official leaderboard
CEO-Bench $0 #19 of 19 Official leaderboard
MCPMark Verified 71.7% #8 of 8 Official leaderboard
EdgeBench 31.0 #5 of 5 Official leaderboard
STATE-Bench Main 0.7.0 47.2% #4 of 4 Official leaderboard

Professional work

BenchmarkScoreRankSource
Vals Finance Agent 60.4% #4 of 49 Official leaderboard
LEXam-hard 38.9% #2 of 15 Official leaderboard
Vals Index v1 62.2% #7 of 34 Official leaderboard
SnorkelUnderwrite 2.0 28.5% #4 of 16 Official leaderboard
EnterpriseBench: CoreCraft 53.3% #17 of 51 Official leaderboard
EnterpriseClawBench 58.8% #4 of 9 Official leaderboard
FoodTruck Bench $27,142 #27 of 59 Official leaderboard
CommerceAgentBench 50.5% #6 of 13 Official leaderboard
CorpFin v2 61.4% #55 of 119 Official leaderboard
FORTE 70.5% #10 of 21 Official leaderboard
ProfBench 48.2% #35 of 72 Official leaderboard
GDPval-AA 1,554 #8 of 16 Lab model card
TaxEval v2 72.1% #65 of 129 Official leaderboard
CaseLaw v2 59.4% #27 of 53 Official leaderboard
Harvey LAB (Legal Agent Benchmark) 3.75% #41 of 80 Official leaderboard
Public Benefits Bench 62.9% #26 of 50 Official leaderboard
YC-Bench $1,066,426 #31 of 59 Official leaderboard
Tax Agent Bench 25.2% #45 of 84 Official leaderboard
$OneMillion-Bench 26% #7 of 13 Official leaderboard
PhysicianBench 18.7% #7 of 13 Official leaderboard
Clipping Bench $462 #11 of 19 Official leaderboard
TRLawBench 8.97 #20 of 34 Official leaderboard
Investor's Last Exam 59.3 #6 of 10 Official leaderboard
SimuVerity 28.4 #3 of 5 Official leaderboard
LegalBench 80.3% #84 of 138 Official leaderboard
SnorkelLegal 30.4% #10 of 16 Official leaderboard
Vals Index v2 42.9% #41 of 65 Official leaderboard
MedCode 40.5% #61 of 96 Official leaderboard
E-Commerce Bench 1.90 #13 of 19 Official leaderboard
CHI-Bench 14.2% #16 of 23 Official leaderboard
DAYJOB: Healthcare 1.20% #26 of 37 Official leaderboard
Vending-Bench 2 $3,285 #48 of 67 Official leaderboard
Excel Modeling Benchmark 51.6% #55 of 76 Official leaderboard
Vals Finance Agent v2 44.1% #58 of 80 Official leaderboard
Legal Research Bench 23.1% #68 of 93 Official leaderboard
Agents' Last Exam 12.4% #23 of 31 Official leaderboard
DAYJOB: Finance 3.50% #29 of 39 Official leaderboard
APEX-1 59.7% #54 of 71 Official leaderboard
MedScribe 75.1% #74 of 97 Official leaderboard
Tuesday Work Index 39.6 #34 of 44 Official leaderboard
PolicyBench 79.3% #39 of 50 Official leaderboard
prinzbench 23.0 #21 of 26 Official leaderboard
SnorkelFinance 2.0 9.50% #13 of 16 Official leaderboard
CAD Arena 0.19 #10 of 12 Official leaderboard
GDPval Extended 0.20% #36 of 43 Official leaderboard
Vals Index v2.1 38.6% #42 of 49 Official leaderboard
DeepFinance 20.3% #20 of 23 Official leaderboard
SnorkelManufacturing 8.10% #14 of 16 Official leaderboard
SnorkelRevOps 9.50% #14 of 16 Official leaderboard
Ivo-micro1 ContractBench 48.4% #7 of 8 Official leaderboard
FrontierFinance 40.5% #15 of 17 Official leaderboard
GDP.xlsx 6.90% #18 of 20 Official leaderboard
ATLAS-Finance 1.67% #10 of 11 Official leaderboard
ATLAS-Finance v1.1.0 1.67% #10 of 11 Official leaderboard
WorkplaceAgents 8.60% #13 of 14 Official leaderboard
Digital Work Index 11.1 #15 of 16 Official leaderboard
Legora BAR 93.0 #16 of 17 Official leaderboard
MerchantBench 16.7 #8 of 8 Official leaderboard
GDPevo 48.8% #4 of 4 Official leaderboard
DeepSeek Chinese white-collar tasks evaluation 53% #1 of 1 Lab model card

Vision, audio and robotics

BenchmarkScoreRankSource
MineBench 1,442 #43 of 75 Official leaderboard
AgentWorldBench 53.0 #9 of 13 Official leaderboard
VoxelBench 1,289 #51 of 56 Official leaderboard
ArtifactArena 765 #20 of 21 Official leaderboard
Code2Video Benchmark 1,387 #19 of 19 Official leaderboard

Safety and security

BenchmarkScoreRankSource
CyberGym 95.4% #4 of 34 Official leaderboard
ABC Bench Liquid-Handling Robot 99.3% #6 of 34 Official leaderboard
nocot-bench (NCRI) 84.1 #48 of 185 Official leaderboard
ABC Bench Fragment Design 91.1% #11 of 27 Official leaderboard
PCT (Polarized Contrastive Pairs) 44% #22 of 46 Official leaderboard
Human Pathogen Capabilities Test v2 (HPCT v2) 52.7% #38 of 71 Official leaderboard
CyberBench (Vals) 68.6% #15 of 26 Official leaderboard
KORA 30.5% #26 of 42 Official leaderboard
SecCodeBench 58.3% #4 of 6 Official leaderboard
World-Class Bio 37.3% #53 of 74 Official leaderboard
DrugDiscoveryBench 31.7% #15 of 20 Official leaderboard
CheatBench 75.1% #8 of 10 Official leaderboard
Last Secure Code Benchmark 18% #5 of 6 Official leaderboard
Human Pathogen Capabilities Test 38.8% #72 of 86 Official leaderboard
DTap Prompt Injection 41.7% #7 of 8 Official leaderboard
Dictatorship Eval 11.7% #18 of 20 Official leaderboard
GeneBench Pro 2.42% #11 of 12 Official leaderboard
Molecular Biology Capabilities Test 33.6% #83 of 90 Official leaderboard
BioSimBench 0.11 #12 of 12 Official leaderboard
DTap Direct Misuse 59.6% #8 of 8 Official leaderboard

Other

BenchmarkScoreRankSource
ACCoRD 70.9 #48 of 152 Official leaderboard
Conceptual Reasoning Index 55.8 #50 of 151 Official leaderboard
LMCA 41.2% #54 of 147 Official leaderboard
Political Science LLM Benchmark 66.3% #24 of 42 Official leaderboard
CompassRank 77.5 #8 of 14 Official leaderboard
CAIS Text Capabilities Index 26.4 #41 of 68 Official leaderboard
CAIS Risk Index 65.5 #60 of 67 Official leaderboard

Its lab's numbers put it at 152.7, 2.5 points above its ECI

The Self-Reported Index fits one score to the 38 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 149.7 to 155.7. How we compute it.

Its lab's numbers next to 8 leaderboards

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
OpenAI MRCR v2 83.5% MRCR 1M (MMR), OpenAI MRCR, DS-V4-Pro Max (Table 6). Figure 9 prints 8-needle MMR by context length 33.2% Leaderboard, 8-needle, 128K to 256K bin, enabled reasoning +50 pts
SimpleQA Verified 57.9% SimpleQA-Verified (Pass@1) 47% Epoch AI run, max +11 pts
MathArena (Apex, IMO, USAMO) 38.3% Apex (Pass@1) 28.1% Leaderboard, max +10 pts
SWE-bench Verified 80.6% SWE Verified (Resolved) 77.6% Epoch AI run, max +3.0 pts
MathArena Apex Shortlist 90.2% Apex Shortlist (Pass@1) 87.8% Leaderboard, max +2.4 pts
HMMT 2026 95.2% HMMT 2026 Feb (Pass@1) 93.9% Leaderboard, max +1.3 pts
Toolathlon 51.8% Toolathlon (Pass@1) 52.8% Leaderboard −1.0 pts
GPQA Diamond 90.1% GPQA Diamond (Pass@1) 90.9% Epoch AI run, high −0.8 pts

DeepSeek's models over time

DeepSeek's models in color, against every other model on the Epoch Capabilities Index, by release date.

50 100 150 2023 2024 2025 2026 2027 DeepSeek V4.1 Flash 1017 DeepSeek-V4-Flash DeepSeek-V3.1 DeepSeek-V3 DeepSeek-V2 (MoE-236B, May 2024) DeepSeek Coder 33B DeepSeek Coder 6.7B 50 100 150 2023 2024 2025 2026 2027 DeepSeek V4.1 Flash 1017 DeepSeek-V4-Flash

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

DeepSeek-V4-Pro model card Apr 24, 2026

Its system card

Every result in the longer documents the lab published with it. 9 of them aren't in the launch post.

DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence (arXiv 2606.19348) Apr 26, 2026

Capabilities

Hallucination