Benchmark Almanac

DeepSeek V4.1 Flash

DeepSeek V4.1 Flash ranks first on 9 of the 243 benchmarks with five or more models scored.

DeepSeek Released September 9th, 2026 Family DeepSeek-V4.1 The lab's model card Compare with others
246
Benchmarks with a score
9
Ranked first, out of five or more models
19
In the top three
DeepSeek Sep 2026

DeepSeek V4.1 Flash

Open weights Mid-size

156.7 #34 of 256
Epoch Capabilities Index
Self-Reported Index 157.6 #45 of 404
  • MMMU-Pro 56.5% #72 /90
  • NL2Repo-Bench 65.4% #2 /30
  • LongBench v2 45.2% #29 /35

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
Vectara Hallucination Leaderboard (2025 dataset) 4.20% #5 of 104 Official leaderboard
Humanity's Last Exam 63.9% #9 of 154 Lab model card
LMArena (Elo) 1,475 #32 of 335 Official leaderboard
AGIEval 83.4% #4 of 40 Lab model card
C-Eval 92.1% #5 of 46 Lab model card
Wer wird Millionär? LLM Benchmark 175000.0 #14 of 77 Official leaderboard
DesignArena ASCII Art 1,260 #17 of 77 Official leaderboard
Longform Creative Writing (EQ-Bench) 74.0 #29 of 119 Official leaderboard
Extended NYT Connections 89.6% #26 of 103 Official leaderboard
AA-LCR Extended 70.8% #13 of 49 Official leaderboard
DROP 87.9% #11 of 41 Lab model card
Last Translation Benchmark 13.8 #11 of 36 Official leaderboard
BIG-Bench Hard 86.1% #23 of 73 Lab model card
BullshitBench 41.8% #34 of 107 Official leaderboard
BullshitBench V2 57% #41 of 129 Official leaderboard
HellaSwag 87.2% #18 of 53 Lab model card
SimpleQA Verified 42.3% #21 of 56 Official leaderboard
HLE Extended 45.3% #12 of 31 Official leaderboard
Creative Writing v3 70.3 #53 of 120 Official leaderboard
StudyArena Text 1,234 #43 of 85 Official leaderboard
MGSM 80.2% #34 of 57 Lab model card
Lech Mazur Creative Writing -0.55 #30 of 50 Official leaderboard
BBEH micro average 27.2% #9 of 15 Lab model card
MultiLoKo 45.5% #2 of 3 Lab model card
MMLU-Pro 74.1% #112 of 167 Lab model card
SuperGPQA 53.1% #26 of 35 Lab model card
LongBench v2 45.2% #29 of 35 Lab model card
AI Detection Benchmark 65.4% #3 of 3 Official leaderboard

Math and science

BenchmarkScoreRankSource
AIME 2026 100% #1 of 38 Lab model card
OTIS Mock AIME 2024-2025 98.3% #32 of 212 Epoch AI run
GPQA Diamond 89.8% #50 of 229 Epoch AI run
ARC-AGI 94.5% #23 of 95 Official leaderboard
MathArena (Apex, IMO, USAMO) 65.6% #3 of 12 Lab model card
ARC-AGI-2 72.9% #25 of 98 Official leaderboard
PHYSICS-Verified 87% #5 of 19 Official leaderboard
SimpleBench 66.7% #28 of 106 Official leaderboard
xbench-ScienceQA 72.6% #8 of 29 Official leaderboard
GSM8K 93% #26 of 92 Lab model card
DTBench 89.9% #54 of 180 Official leaderboard
FrontierMath Tiers 1-3 v2 67.4% #33 of 86 Official leaderboard
Terminal-Bench-Science 15.7% #7 of 17 Official leaderboard
Prophet Arena 85.7 #11 of 24 Official leaderboard
LatchBio Capabilities v1.0 42.1% #14 of 27 Official leaderboard
Science's Last Exam 26.8 #5 of 9 Official leaderboard
ProofBench v1.1 54% #29 of 52 Official leaderboard
PhysClosureBench 4.50 #4 of 7 Official leaderboard
MATH 61.1% #57 of 99 Lab model card
FrontierMath Tier 4 v2 26.8% #41 of 68 Official leaderboard
scBench (LatchBio SingleCellBench) 54.7% #16 of 26 Official leaderboard
Apodex Discovery 2.55 #5 of 8 Official leaderboard
BioMysteryBench 67.8% #19 of 30 Official leaderboard
Surface Evolver Bench 46.3% #17 of 25 Official leaderboard
FrontierPhysics 3.60% #16 of 22 Official leaderboard
ExplorationBench 59.5% #8 of 10 Official leaderboard
PaperBenchX 7.53% #8 of 10 Official leaderboard
MysteryMechanism 21.2% #25 of 31 Official leaderboard
ArXivLean 5.66% #6 of 7 Official leaderboard

Coding

BenchmarkScoreRankSource
Codeforces rating 3,471 #1 of 47 Lab model card
SWE-bench Verified Extended 82.7% #2 of 44 Official leaderboard
Claw-SWE-Bench 81.7% #1 of 16 Official leaderboard
NL2Repo-Bench 65.4% #2 of 30 Lab model card
BigCodeBench 60.6% #1 of 14 Lab model card
SWE-bench-Live 71% #1 of 12 Official leaderboard
DesignArena Agentic Overall 1,300 #7 of 74 Official leaderboard
DesignArena 3D Design 1,347 #16 of 136 Official leaderboard
Code Arena 1,664 #16 of 128 Official leaderboard
DesignArena Website 1,299 #19 of 151 Official leaderboard
DesignArena Code 1,310 #18 of 143 Official leaderboard
DesignArena Game Dev 1,325 #18 of 143 Official leaderboard
SWE Atlas Codebase QnA Extended 36.7% #5 of 38 Official leaderboard
DesignArena UI Component 1,304 #22 of 139 Official leaderboard
DeepSWE 74.2% #9 of 56 Lab model card
DesignArena Data Visualization 1,286 #19 of 118 Official leaderboard
WebDev Arena 1,614 #18 of 105 arena.ai, via Epoch AI
Vibe Code Bench 84.7% #21 of 120 Official leaderboard
APEX-SWE 51.9% #13 of 65 Official leaderboard
Code Migration 45.6% #17 of 78 Official leaderboard
Image-to-WebDev Arena 1,610 #12 of 55 Official leaderboard
MirrorCode 64.4% #9 of 39 Epoch AI run
DesignArena SVG 1,244 #24 of 103 Official leaderboard
ALE-Bench 1,829 #30 of 126 Official leaderboard
Data Agent Benchmark 92.3% #5 of 18 Official leaderboard
DesignArena Image to HTML 1,226 #17 of 51 Official leaderboard
Code Migration v1.1 58.9% #10 of 30 Official leaderboard
DesignArena Image to Frontend 1,235 #11 of 31 Official leaderboard
ScarfBench 69% #4 of 11 Official leaderboard
SWE Atlas v1.1 62.1% #11 of 28 Official leaderboard
Doc2Harness 40.2% #9 of 19 Official leaderboard
Landing Page Arena 1,482 #7 of 14 Official leaderboard
FDE-Gym 0.21 #6 of 12 Official leaderboard
Real-SWE 1.1 38.4% #8 of 15 Official leaderboard
HumanEval 79.4% #55 of 101 Lab model card
OpenDesign Arena 81.2 #12 of 21 Official leaderboard
MiMo Code Bench 60.2% #5 of 8 Official leaderboard
Brokk Power Ranking 7.80 #12 of 19 Official leaderboard
Senior SWE-Bench 22.7% #15 of 23 Official leaderboard
Vibe Code Bench 1-100 16.4% #17 of 26 Official leaderboard
DAXBench v2 72.8% #31 of 47 Official leaderboard
MiMo Visual Coding 70.6% #6 of 9 Official leaderboard
Legacy-Bench 49% #4 of 6 Official leaderboard
OWE-Bench 12.4% #10 of 14 Official leaderboard
IOI v2 (Vals) 40.3% #35 of 45 Official leaderboard
FrontierCS 2 12.5% #15 of 19 Official leaderboard
MacroscopeBench 72% #13 of 16 Official leaderboard
WeirdML v3 5.95% #23 of 27 Official leaderboard
ProgramBench almost solved 20.3% #7 of 8 Lab model card
FrontierSWE 32.9% #9 of 10 Lab model card
RSI Arena 887 #10 of 10 Official leaderboard
ViBench 2.0 0% #9 of 9 Official leaderboard
RSI Bench 32.1 #8 of 8 Official leaderboard

Agents and tools

BenchmarkScoreRankSource
Terminal-Bench 2.1 90.6% #3 of 56 Lab model card
AdaExploreBench 70.3% #1 of 8 Official leaderboard
AutomationBench 54.8% #6 of 47 Lab model card
BrowseComp Extended 48.2% #7 of 42 Official leaderboard
LLM Chess 1,087 #21 of 123 Official leaderboard
Mystery Game Puzzles 43% #19 of 84 Epoch AI run
Chess Puzzles 48% #23 of 93 Epoch AI run
RuneBench 5.68 #19 of 76 Official leaderboard
CivBench 1,537 #5 of 19 Official leaderboard
Terminal-Bench 2.1 Extended 38.4% #12 of 44 Official leaderboard
Terminal-Bench 3.0 30% #2 of 6 Lab model card
Web Context (Nimble) 73.5% #4 of 11 Official leaderboard
Agent Arena 2.00% #22 of 56 Official leaderboard
DuelLab GameBench 2 68.4 #25 of 57 Official leaderboard
ARC-AGI-3 3.26% #13 of 27 Official leaderboard
GAIA 41.2% #37 of 64 Official leaderboard
Hermes Bench 70.7% #10 of 17 Official leaderboard
Collaborative Intelligence Benchmark 50.8% #11 of 18 Official leaderboard
BU Bench V2.1 70.9 #11 of 16 Official leaderboard
Terminal-Bench 4.0 31.2% #23 of 28 Lab model card
Realtime-Bench 10% #8 of 8 Official leaderboard

Professional work

BenchmarkScoreRankSource
DesignArena Spreadsheet 1,308 #4 of 35 Official leaderboard
FinStepBench LiveSearch 76.7% #1 of 6 Official leaderboard
CorpBench Work 89.2 #3 of 17 Official leaderboard
FoodTruck Bench $50,179 #11 of 59 Official leaderboard
Vals Finance Agent v2.1 65.4% #10 of 40 Official leaderboard
MedScribe 85.5% #26 of 97 Official leaderboard
YC-Bench $1,657,236 #17 of 59 Official leaderboard
Investor's Last Exam 63.0 #3 of 10 Official leaderboard
BlenderWorld 65.7 #4 of 13 Official leaderboard
Vals Index v2 57.9% #21 of 65 Official leaderboard
GDPval Extended 1.70% #14 of 43 Official leaderboard
Legal Research Bench 41.4% #33 of 93 Official leaderboard
Remote Labor Index 9.58% #7 of 19 Official leaderboard
PolicyBench 86.3% #19 of 50 Official leaderboard
Hebbia Financial Services Benchmark 63.1% #4 of 10 Official leaderboard
LegalBench 83.3% #56 of 138 Official leaderboard
APEX-1 65.5% #30 of 71 Official leaderboard
Agents' Last Exam 31.8% #11 of 26 Lab model card
Harvey LAB (Legal Agent Benchmark) 6.67% #34 of 80 Official leaderboard
Vals Finance Agent v2 53.5% #34 of 80 Official leaderboard
RCM-Bench-1 89% #9 of 20 Official leaderboard
Public Benefits Bench 64.3% #23 of 50 Official leaderboard
Tax Agent Bench 26.6% #42 of 84 Official leaderboard
Excel Modeling Benchmark 57.2% #40 of 76 Official leaderboard
Vals Index v2.1 51.3% #26 of 49 Official leaderboard
Vals Finance Index 57.1% #22 of 40 Official leaderboard
Physical-Engineering's Last Exam 0.23 #10 of 18 Official leaderboard
MedCode 41.2% #54 of 96 Official leaderboard
Thrive Holdings helpdesk ticket resolution benchmark 48.1% #9 of 15 Official leaderboard
SheetBench 58.4% #5 of 8 Official leaderboard
OR-License 69.7% #7 of 11 Official leaderboard
EnterpriseBench: CoreCraft 41.5% #33 of 51 Official leaderboard
FinStepBench DeepResearch 50.2% #4 of 6 Official leaderboard
Immigration Bench 65.6% #17 of 25 Official leaderboard
Excel Modeling Benchmark v1.1 67.8% #29 of 42 Official leaderboard
Legora BAR 100.0 #12 of 17 Official leaderboard
EngiWorld 25.9% #5 of 7 Official leaderboard
DeepFinance 27.3% #17 of 23 Official leaderboard
RWE-bench 0% #12 of 15 Official leaderboard
APEX-Agents 39.5% #49 of 61 Collected by Epoch AI, source not named
MedXpertQA MM Extended 48.4% #33 of 41 Official leaderboard
Tuesday Work Index 35.2 #36 of 44 Official leaderboard
FinStepBench CorporateValuation 57.6% #5 of 6 Official leaderboard
Holdings-IndividualTaxBench 69.9% #11 of 13 Official leaderboard
APEX-Agents 1.1 50.5% #56 of 65 Official leaderboard
TaxCalcBench v2 4.00% #28 of 32 Official leaderboard
APEX-Accounting 50.1% #43 of 49 Official leaderboard
Bedside Bench 82.9% #16 of 17 Official leaderboard
Quant-Arena 36.5 #9 of 9 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
ZeroBench 49% #6 of 76 Official leaderboard
DocVQA 95.6% #6 of 36 Lab model card
MineBench 1,804 #17 of 75 Official leaderboard
BabyVision 89.6% #6 of 26 Lab model card
Vision Arena 1,263 #35 of 136 Official leaderboard
VoxelBench 1,881 #17 of 56 Official leaderboard
CineRecon 49.5 #4 of 12 Official leaderboard
VibeVideoBench 3.91 #3 of 9 Official leaderboard
Frontier3DBench 57.5 #4 of 11 Official leaderboard
ExtractBench 87.1% #12 of 27 Official leaderboard
BVB 66.8 #22 of 45 Official leaderboard
4DCodeBench 0.60 #8 of 16 Official leaderboard
Code2Video-Compass 8.00 #9 of 15 Official leaderboard
RoboColiseum 55.1% #6 of 9 Official leaderboard
Furniture Assembly Benchmark 34.2% #24 of 35 Official leaderboard
MMMU-Pro 56.5% #72 of 90 Lab model card
MMMU-Pro Extended 25% #32 of 38 Official leaderboard
CVBench 77.9% #13 of 15 Lab model card
Chartography 11.6% #38 of 43 Official leaderboard
Embodied First Exam 3.07% #8 of 9 Official leaderboard
CharXiv Extended 68.9% #36 of 38 Official leaderboard
InteractiveBench 12.5% #12 of 12 Official leaderboard
DataCenterBench Tier 1 11.7% #9 of 9 Official leaderboard
CodeCap-Bench 44.3 #7 of 7 Official leaderboard
RobotWorld 1.19% #5 of 5 Official leaderboard

Safety and security

BenchmarkScoreRankSource
CyberGym 99.4% #1 of 34 Official leaderboard
SecIT Bench 82.9% #1 of 26 Official leaderboard
DistressBench 79.8% #5 of 42 Official leaderboard
BioSecBench-Surveillance 51.5% #5 of 27 Official leaderboard
CyberBench (Vals) 78.8% #5 of 26 Official leaderboard
CyberBench v1.1 (Vals) 73.7% #9 of 44 Official leaderboard
dfbench 57.3% #5 of 18 Official leaderboard
nocot-bench (NCRI) 81.3 #56 of 185 Official leaderboard
SAFE-Teen Health advice 41% #6 of 17 Official leaderboard
EpiBench 25.5% #10 of 27 Official leaderboard
ExploitGym 23.7% #7 of 18 Official leaderboard
xdev-bench-hard 85.7% #2 of 5 Official leaderboard
BioSecBench-Function 40.8% #11 of 27 Official leaderboard
SAFE-Teen Urgent safety 31% #7 of 17 Official leaderboard
SAFE-Teen Relationship boundaries 51% #8 of 17 Official leaderboard
TxBench-Antibody-Discovery 39.7% #16 of 30 Official leaderboard
MetagenomicsBench 38% #14 of 26 Official leaderboard
TxBench-Oligo-Discovery 34.8% #18 of 30 Official leaderboard
MiMo Cyber Bench 62.7% #3 of 5 Official leaderboard
scBench-Long 14.8% #17 of 27 Official leaderboard
SpatialBench-Long 26.2% #17 of 27 Official leaderboard
BioSecBench-Refusal 11.3% #14 of 22 Official leaderboard
TxBench-Preclinical-Pharmacology 46.7% #20 of 30 Official leaderboard
SEC-bench Pro 62.8% #8 of 11 Lab model card
OncologyBench 14.7% #22 of 30 Official leaderboard
VariantBench 31.1% #20 of 27 Official leaderboard
CWE-bench 55% #9 of 12 Official leaderboard
SRE-Bench 0.76% #29 of 35 Official leaderboard
Biopharma Bench 45.8% #11 of 13 Official leaderboard
FrontierBioML 22.4 #10 of 11 Official leaderboard
VIALS 2 12.6% #13 of 14 Official leaderboard
IPI Bench 35.3% #6 of 6 Official leaderboard

Other

BenchmarkScoreRankSource
LMCA 47% #36 of 147 Official leaderboard
Conceptual Reasoning Index 58.4 #45 of 151 Official leaderboard
SAGE (Vals) 47.9% #29 of 81 Official leaderboard
ACCoRD 68.1 #59 of 152 Official leaderboard
SAGE v1.1 (Vals) 85.4% #17 of 29 Official leaderboard
Political Science LLM Benchmark 66.2% #25 of 42 Official leaderboard
Hermes Index 36.9 #11 of 17 Official leaderboard
JevBench 89.6% #3 of 4 Official leaderboard
Vals Science Index 29.6% #20 of 25 Official leaderboard

Its lab's numbers put it at 157.6, in line with its ECI

The Self-Reported Index fits one score to the 32 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 152.7 to 162.6. How we compute it.

Its lab's numbers next to 6 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
Chartography 78.9% Chartography w/ tools (Pass@1) 11.6% Leaderboard, max +67 pts
CyberGym 88.1% CyberGym (Pass@1) 99.4% Leaderboard, dynamic, TuLongFeng −11 pts
ExploitGym 15.3% ExploitGym (Pass@1) 23.7% Leaderboard, VARAS-OneMind \(智安·宇盾\) −8.4 pts
GPQA Diamond 90.9% GPQA Diamond (Pass@1) 89.8% Epoch AI run, max +1.1 pts
SimpleQA Verified 42.3% SimpleQA-Verified (EM) 42.3% Leaderboard Same
ZeroBench 49% ZeroBench-main w/ tools (Pass@5) 49% Leaderboard, w/ tools; externally reported: DeepSeek-V4.1-Flash model card; Claude Code harness tools; 5 samplings. Same

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-V4.1-Flash model card Sep 10, 2026

Its system card

Every result in the longer documents the lab published with it. 4 of them aren't in the launch post.

DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression Sep 10, 2026

Capabilities

Hallucination

Cyber risk