The LLM benchmark aggregator.

DeepSeek V4 Pro

DeepSeek · 1.7T · open weights

Strongest in Agentic coding (#10 of 157, on 2 of its 4 boards), weakest in Toxicity avoidance (#213 of 272). Above par in 25 of 33 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5.1 and Claude Fable 5 and ahead of Claude Sonnet 4.6 and GPT-5.

§ 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.

Coding58.1−11#17/1652/5

Agentic coding60.5−8.2#10/1572/4

Code generation49.6−20.4#48/771/2

Repository Q&A50.5−16.2—/0✱✱

Agents59.7−8.4#32/2682/5

Knowledge work62.6−11#27/1782/2

Workflow automation44.7−17.3—/0✱✱

Reasoning55.5−11.4#36/1784/4

Science61.4−2.1#17/1221/1

Reasoning53.7−15.4#47/1403/3

Mathematics55−10.7#50/1402/2

Expert reasoning46.5−14.4—/0✱✱

Knowledge56−7.5#38/1381/2

Academic knowledge57.2−6.7#35/1231/1

Professional54.3−7.3#39/1681/1

Legal55.9−11.2#33/1511/1

Finance55.5−7.4#41/1511/1

Medical51.2−15.6#72/1401/1

Cybersecurity47.5−12.2—/9✱0/1

Human preference63.2−4.3#45/3421/1

Core abilities55−12.4#47/2043/3

Data analysis59.7−6.4#14/571/1

Language55.3−16.2#22/571/1

Instruction following49.6−24.1#30/571/1

General intelligence55−15.1#61/2043/3

Safety50.5−11#168/3372/3

Factual grounding53.5−17.1#42/1011/1

Jailbreak resistance54.4−10.5#124/2721/1

Harm refusal51.3−10.3#135/3001/2

Secure code49.9−16.6#156/2741/1

Fairness45.2−24.8#191/3001/2

Toxicity avoidance48.3−10.7#213/2721/1

§ 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 Pro, left for the other.

Mistral Large 3won 20 · lost 0

Gemini 3.5 Flash Litewon 22 · lost 0

Grok 4.3won 27 · lost 1

Command Awon 22 · lost 1

DeepSeek V4 Flashwon 30 · lost 2

DeepSeek V3won 24 · lost 2

GPT-5.4 Miniwon 21 · lost 2

Grok 3won 19 · lost 2

Mistral Smallwon 18 · lost 2

GPT-5.4 Nanowon 21 · lost 3

Grok 4.1 Fastwon 19 · lost 3

Qwen3.6 Pluswon 19 · lost 3

Kimi K2won 20 · lost 4

GLM-4.6won 20 · lost 4

Llama 4 Scout Instructwon 19 · lost 4

GPT OSS 20Bwon 17 · lost 4

Claude Haiku 4.5won 20 · lost 5

DeepSeek V3.2won 20 · lost 5

Kimi K2.6won 23 · lost 6

GPT-4.1 Nanowon 19 · lost 5

GPT-4.1 Miniwon 19 · lost 5

Qwen3.8 27Bwon 18 · lost 5

Gemma 4 31Bwon 18 · lost 5

Mimo V2.5 Prowon 18 · lost 5

Gemini 2.0 Flashwon 17 · lost 5

Llama 4 Maverick Instructwon 19 · lost 6

GPT OSS 120Bwon 19 · lost 6

Gemini 2.5 Flashwon 19 · lost 6

DeepSeek R1won 19 · lost 6

GPT-6 Lunawon 19 · lost 6

Kimi K2.5won 19 · lost 6

Claude 3.5 Haikuwon 15 · lost 5

GPT-4o Miniwon 18 · lost 6

GPT-5 Nanowon 16 · lost 6

Claude 3.7 Sonnetwon 15 · lost 6

MiniMax M3won 22 · lost 9

GPT-4owon 15 · lost 7

Qwen3.7 Maxwon 19 · lost 9

GLM-5.2won 21 · lost 10

Claude Sonnet 4won 17 · lost 9

Grok 4won 15 · lost 8

Kimi K2 Thinkingwon 13 · lost 7

Claude 3.5 Sonnetwon 13 · lost 7

O1won 13 · lost 7

Inklingwon 20 · lost 11

GLM-5.1won 14 · lost 8

Qwen3 235B A22B Instructwon 15 · lost 9

MiniMax M2.7won 13 · lost 8

Gemini 2.5 Prowon 16 · lost 10

O3 Miniwon 14 · lost 9

O4 Miniwon 15 · lost 10

Claude Sonnet 4.5won 15 · lost 10

GPT-5.6 Lunawon 18 · lost 12

Claude Opus 4won 14 · lost 10

GPT-5.2won 16 · lost 12

GLM-5.3 Flashwon 16 · lost 12

GPT-5.1won 13 · lost 10

GPT-4.1won 13 · lost 11

Claude Opus 4.5won 12 · lost 11

GPT-5 Miniwon 13 · lost 12

GPT-5won 13 · lost 12

Claude Sonnet 4.6won 14 · lost 13

O3won 12 · lost 12

Gemini 3.1 Prowon 12 · lost 12

Gemini 3.6 Flashwon 14 · lost 14

Gemini 3.5 Flashwon 13 · lost 15

Muse Spark 1.1won 12 · lost 16

Grok 4.7won 10 · lost 15

Gemini 3.8 Flashwon 11 · lost 18

Claude Opus 4.6won 9 · lost 16

Grok 4.5won 10 · lost 18

GPT-6 Solwon 9 · lost 17

GLM-5.3won 9 · lost 17

Gemini 3 Prowon 7 · lost 14

Claude Sonnet 5won 10 · lost 20

DeepSeek V4.1 Flashwon 8 · lost 17

Claude Opus 4.7won 9 · lost 20

GPT-5.4won 8 · lost 21

Gemini 3.7 Flashwon 6 · lost 16

GPT-5.6 Terrawon 8 · lost 22

Qwen3.8 Maxwon 7 · lost 21

Grok 4.6won 7 · lost 22

GPT-5.5won 7 · lost 22

Muse Spark 1.2won 5 · lost 16

Mimo V2.6 Prowon 4 · lost 16

Kimi K3won 5 · lost 21

GPT-6 Astrawon 4 · lost 18

Claude Opus 4.8won 5 · lost 23

GPT-5.6 Solwon 5 · lost 28

Claude Opus 5won 3 · lost 29

Claude Opus 5.5won 2 · lost 23

Claude Fable 5won 0 · lost 22

Claude Fable 5.1won 0 · lost 23

§ 3 · Sources

Where the numbers come from

12 publications, 51 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-09-26

LMArena Text 1464

Artificial Analysis ↗1 measureread 2026-09-26

Artificial Analysis Intelligence Index 36

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 1441

AA-Briefcase ↗1 measureread 2026-09-26

AA-Briefcase 1258

ARC-AGI-2 ↗1 measureread 2026-09-26

ARC-AGI-2 61.3%

LiveBench ↗8 measuresread 2026-09-26

LiveBench 77.4LiveBench · Reasoning 85.8LiveBench · Coding 77.2LiveBench · Agentic Coding 54.9LiveBench · Mathematics 95.1LiveBench · Data Analysis 79.2LiveBench · Language 82.1LiveBench · Instruction Following 67.7

Kagi LLM Benchmark ↗1 measureread 2026-09-26

Kagi LLM Benchmark 53.5%

SimpleBench ↗1 measureread 2026-09-26

SimpleBench 50.9%

Vals.ai ↗14 measuresread 2026-09-26

Vals · Legal Research Bench 40.87%Vals · LegalBench 82.36%Vals · Harvey Legal Agent Benchmark 7.5%Vals · Finance Agent 50.39%Vals · CorpFin 65.42%Vals · TaxEval 73.06%Vals · MedCode 42.47%Vals · MedScribe 80.17%Vals · SWE-bench Verified 96.4% (Mini-SWE-agent)Vals · Vibe Code Bench 82.3% (OpenHands)Vals · Code Migration 41.54%Vals · GPQA Diamond 92.42%Vals · MMLU Pro 86.97%Vals · ProofBench 50%

Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-09-26

Enkrypt · Jailbreak risk 9.1%Enkrypt · Harmful content risk 22.2%Enkrypt · CBRN risk 17.3%Enkrypt · Toxicity risk 7.5%Enkrypt · Bias risk 85.3%Enkrypt · Insecure code risk 33.8%

Vectara Hallucination Leaderboard ↗1 measureread 2026-09-26

Vectara · Factual consistency 91.4%

✱✱ DeepSeek — V4.1 Flash model card ↗the model’s own publisher15 measuresread 2026-09-26

GPQA Diamond (Pass@1) 92.4%Codeforces (Rating) 3348MathArena Apex (Pass@1) 65.3%Terminal-Bench 2.1 (Pass@1) 87.9%Terminal-Bench 3.0 (Pass@1) 11.8%Terminal-Bench 4.0 (Pass@1) 12.4%DeepSWE v1.1 (Resolved) 62.7%ProgramBench (Almost@1) 15.5%NL2Repo-Bench (Score) 61.5%CyberGym (Pass@1) 83.3%SEC-Bench Pro (Pass@1) 56.4%ExploitGym (Pass@1) 5.4%HLE w/ tools (Pass@1) 60%AutomationBench (Pass@1) 43.2%Agent's Last Exam (Pass@1) 25.7%

Badge

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