The LLM benchmark aggregator.

DeepSeek V4.1 Flash

DeepSeek · 763B · open weights

Strongest in Factual grounding (#6 of 104), weakest in Mathematics (#73 of 152). Above par in 32 of 35 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5.1 and Claude Fable 5 and ahead of GPT-5.6 Terra and Grok 4.7.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Coding61.4−4.8#7/1793/6

Agentic coding61.9−5#7/1733/5

Code generation62.4−2.9—/20✱0/1

Repository Q&A53.7−13—/0✱✱

Human preference63.9−4.9#32/3541/1

Agents56.9−9.5#38/2983/6

Knowledge work66.8−3.9#11/2082/2

Long-horizon work50.8−23.2#13/311/1

Web research50.1−13.8#23/351/2

Computer use53.3−14.8#30/681/2

Workflow automation66leads—/0✱✱

Reasoning55.6−9.8#38/1955/5

Reasoning58.5−10.5#28/1473/3

Mathematics52.6−11.2#73/1522/2

Science46.4−17.1—/122✱0/1

Expert reasoning64.3−2.1—/0✱✱

Core abilities54.8−10.5#47/1993/4

Data analysis57.8−7.4#10/311/1

Language54.1−12.5#26/621/1

Instruction following53.6−20.2#29/621/1

General intelligence54.2−15.2#55/1782/3

Knowledge53.9−8.6#47/1971/3

Document parsing54.2−9.1#37/831/1

Multimodal understanding50.8−21.3—/19✱0/1

Professional53.5−8.9#53/1942/2

Document design60.5−8.2#6/311/1

Cybersecurity57.9−6#10/471/1

Biology research47.3−23.4#19/311/1

IT operations39.6−34.2#29/331/1

Legal56.1−9.8#32/1731/1

Finance55.9−6.5#37/1621/1

Medical52.5−14#70/1491/1

Safety54.6−6.8#59/3291/3

Factual grounding64.6−6.1#6/1041/1

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for DeepSeek V4.1 Flash, left for the other.

DeepSeek V4 Flashwon 21 · lost 0

Step 5won 21 · lost 0

Gemma 4 31Bwon 22 · lost 0

Mistral Large 4won 23 · lost 1

Inklingwon 23 · lost 2

Qwen3.8 27Bwon 22 · lost 3

DeepSeek V4 Prowon 19 · lost 3

GPT-5.4 nanowon 18 · lost 3

GPT-6 Lunawon 24 · lost 4

GLM-5.3 Flashwon 21 · lost 5

Qwen3.8 Maxwon 20 · lost 5

MiniMax M3won 16 · lost 5

GPT-5.6 Lunawon 17 · lost 6

Claude Sonnet 5won 15 · lost 6

Grok 4.5won 14 · lost 7

Claude Haiku 5.5won 16 · lost 11

Grok 4.6won 15 · lost 11

Gemini 3.6 Flashwon 12 · lost 9

Gemini 3.8 Flashwon 16 · lost 12

MiMo V2.6 Prowon 13 · lost 10

Grok 4.7won 14 · lost 12

GPT-5.6 Terrawon 14 · lost 13

Kimi K3won 12 · lost 13

GPT-6 Solwon 11 · lost 12

Gemini 3.1 Prowon 10 · lost 11

GPT-5.4won 9 · lost 12

Claude Opus 4.7won 9 · lost 12

Muse Spark 1.3won 8 · lost 15

Muse Wonderwon 7 · lost 16

Claude Sonnet 5.5won 8 · lost 19

GPT-6 Astrawon 8 · lost 21

GPT-6.1 Solwon 7 · lost 20

GPT-5.5won 5 · lost 17

GPT-5.6 Solwon 5 · lost 20

Gemini 4 Argonwon 4 · lost 21

Claude Opus 5.5won 4 · lost 25

Claude Opus 5won 3 · lost 19

Claude Fable 5won 1 · lost 20

Claude Fable 5.1won 1 · lost 25

§ 3 · Sources

Where the numbers come from

12 publications, 59 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-11-10

LMArena Text 1475

Artificial Analysis ↗1 measureread 2026-11-10

Artificial Analysis Intelligence Index 33

GDPval-AA ↗1 measureread 2026-11-10

GDPval-AA 1746

AA-Briefcase ↗1 measureread 2026-11-10

AA-Briefcase 1570

ARC-AGI-2 ↗1 measureread 2026-11-10

ARC-AGI-2 78.1%

LiveBench ↗8 measuresread 2026-11-10

LiveBench 70.3LiveBench · Agentic 38.8LiveBench · Reasoning 89.1LiveBench · Coding 80.4LiveBench · Browser Use 67.2LiveBench · Data Analysis 79.5LiveBench · Research 78.6LiveBench · Design 79.8

LiveBench 1.0 ↗8 measuresread 2026-11-10

LiveBench 1.0 81.1LiveBench 1.0 · Reasoning 86.7LiveBench 1.0 · Coding 80LiveBench 1.0 · Agentic Coding 77.3LiveBench 1.0 · Mathematics 93.3LiveBench 1.0 · Data Analysis 79.3LiveBench 1.0 · Language 81.2LiveBench 1.0 · Instruction Following 70

SimpleBench ↗1 measureread 2026-11-10

SimpleBench 66.7%

Vals.ai ↗12 measuresread 2026-11-10

Vals · Legal Research Bench 45.19%Vals · LegalBench 83.28%Vals · Harvey Legal Agent Benchmark 14.17%Vals · Finance Agent 56.96%Vals · MedCode 41.17%Vals · MedScribe 85.5%Vals · BioMysteryBench 67.78%Vals · CyberBench 73.69%Vals · SRE Bench 0.76%Vals · Vibe Code Bench 87.73% (OpenHands)Vals · Code Migration 44.21%Vals · ProofBench 54%

ParseBench ↗6 measuresread 2026-11-10

ParseBench · Overall 64.38%ParseBench · Tables 82.84%ParseBench · Charts 51.11%ParseBench · Content faithfulness 86.92%ParseBench · Semantic formatting 73.48%ParseBench · Visual grounding 27.53%

Vectara Hallucination Leaderboard ↗1 measureread 2026-11-10

Vectara · Factual consistency 95.8%

✱✱ DeepSeek — V4.1 Flash model card ↗the model’s own publisher18 measuresread 2026-11-10

GPQA Diamond (Pass@1) 90.9%Codeforces (Rating) 3471MathArena Apex (Pass@1) 65.6%Terminal-Bench 2.1 (Pass@1) 90.6%Terminal-Bench 3.0 (Pass@1) 30%Terminal-Bench 4.0 (Pass@1) 31.2%DeepSWE v1.1 (Resolved) 74.2%ProgramBench (Almost@1) 20.3%NL2Repo-Bench (Score) 64%CyberGym (Pass@1) 88.1%SEC-Bench Pro (Pass@1) 62.8%ExploitGym (Pass@1) 15.3%HLE w/ tools (Pass@1) 63.9%AutomationBench (Pass@1) 54.8%Agent's Last Exam (Pass@1) 31.8%Chartography w/ tools (Pass@1) 78.9%BabyVision w/ tools (Pass@1) 89.6%ZeroBench-main w/ tools (Pass@5) 49%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=deepseek-v4-1-flash)](https://publicai.io/model-index/m/deepseek-v4-1-flash)