The LLM benchmark aggregator.

DeepSeek V3

DeepSeek · 685B · open weights

Strongest in Factual grounding (#24 of 101), weakest in Agents (#259 of 268). Above par in 6 of 27 scopes. Among the models it meets almost everywhere, it finishes behind Gemini 3.8 Flash and Claude Opus 5 and ahead of Gemini 2.5 Flash Lite and GPT OSS 120B.

§ 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.

Professional49.3−12.3#90/1681/1

Legal49.6−17.5#81/1511/1

Medical49.2−17.6#86/1401/1

Finance48.8−14.1#93/1511/1

Knowledge48.2−15.3#96/1381/2

Academic knowledge47.8−16.1#88/1231/1

Core abilities49.9−17.5#97/2041/3

General intelligence49.8−20.3#98/2041/3

Coding45.9−23.2#111/1652/5

Agentic coding52.7−16#52/1571/4

Code generation35−35#71/771/2

Human preference56.6−10.9#127/3421/1

Reasoning42.2−24.7#145/1782/4

Mathematics46.3−19.4#92/1401/2

Science39.9−23.6#99/1221/1

Reasoning39.7−29.4#114/1401/3

Safety47.3−14.2#247/3373/3

Factual grounding59.8−10.8#24/1011/1

Safe-prompt compliance56.6−3.8#24/821/1

Fairness50.7−19.3#104/3002/2

Secure code48.6−17.9#171/2741/1

Toxicity avoidance44.2−14.8#233/2721/1

Harm refusal45.6−16#238/3002/2

Jailbreak resistance28.8−36.1#256/2721/1

Agents37.5−30.6#259/2682/5

Knowledge work33.7−39.9#166/1782/2

§ 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 V3, left for the other.

Command R Pluswon 15 · lost 3

Command Awon 18 · lost 5

Mistral Medium 3.5won 13 · lost 4

Gemini 2.0 Flashwon 17 · lost 6

Gemini 1.5 Prowon 14 · lost 5

Llama 4 Scout Instructwon 17 · lost 7

Gemini 1.5 Flashwon 12 · lost 5

Mistral Smallwon 14 · lost 6

GPT-4o Miniwon 17 · lost 8

GPT-4.1 Nanowon 17 · lost 8

Llama 4 Maverick Instructwon 17 · lost 9

Mistral Largewon 11 · lost 7

Claude 3.5 Haikuwon 12 · lost 8

GPT OSS 120Bwon 14 · lost 12

Gemini 2.5 Flash Litewon 10 · lost 9

Ling 3.0 Flashwon 9 · lost 9

GPT OSS 20Bwon 11 · lost 11

Claude Haiku 4.5won 13 · lost 13

GPT-4owon 11 · lost 12

GPT-5.4 Nanowon 10 · lost 11

Gemini 3.5 Flash Litewon 9 · lost 10

Grok 4.1 Fastwon 10 · lost 12

Mistral Large 3won 9 · lost 11

Grok 4.3won 11 · lost 14

Qwen3.6 Pluswon 8 · lost 11

GPT-5.4 Miniwon 8 · lost 12

GPT-5 Nanowon 9 · lost 14

Claude 3.5 Sonnetwon 8 · lost 13

O3 Miniwon 9 · lost 15

GLM-4.6won 9 · lost 15

Gemini 3.1 Flash Litewon 7 · lost 12

GPT-4.1 Miniwon 9 · lost 16

Kimi K2won 9 · lost 16

DeepSeek V3.2won 9 · lost 16

MiniMax M3won 8 · lost 17

Kimi K2.6won 8 · lost 18

Grok 4won 7 · lost 17

MiniMax M2.7won 6 · lost 15

Qwen3.7 Maxwon 7 · lost 18

Claude 3.7 Sonnetwon 6 · lost 16

Gemma 4 31Bwon 6 · lost 16

Gemini 3.5 Flashwon 6 · lost 19

Gemini 3.6 Flashwon 6 · lost 19

Grok 3won 5 · lost 16

DeepSeek R1won 6 · lost 20

Grok 4 Fastwon 4 · lost 14

Mimo V2.5 Prowon 5 · lost 18

Claude Sonnet 4.6won 5 · lost 19

GPT-4.1won 5 · lost 20

Inklingwon 5 · lost 20

Kimi K2.5won 5 · lost 20

GPT-5 Miniwon 5 · lost 21

Claude Sonnet 4won 5 · lost 22

GLM-5.1won 4 · lost 18

Claude Opus 4.6won 4 · lost 18

GLM-5won 3 · lost 15

GLM-5.2won 4 · lost 21

GLM-4.5won 3 · lost 16

O4 Miniwon 4 · lost 22

Claude Sonnet 4.5won 4 · lost 22

Gemini 2.5 Prowon 4 · lost 23

Claude Opus 4won 3 · lost 22

DeepSeek V4 Flashwon 3 · lost 22

GPT-5.6 Lunawon 3 · lost 22

GLM-5.3 Flashwon 3 · lost 22

Qwen3.8 27Bwon 2 · lost 17

Kimi K2 Thinkingwon 2 · lost 18

Claude Opus 4.5won 2 · lost 18

Gemini 3.1 Prowon 2 · lost 19

GPT-6 Lunawon 2 · lost 20

Gemini 3 Prowon 2 · lost 20

GPT-6 Solwon 2 · lost 21

GPT-5.1won 2 · lost 22

GPT-5.2won 2 · lost 23

Gemini 2.5 Flashwon 2 · lost 24

GPT-5won 2 · lost 24

DeepSeek V4 Prowon 2 · lost 24

GPT-5.4won 2 · lost 24

GPT-6 Astrawon 1 · lost 17

Mimo V2.6 Prowon 1 · lost 18

O1won 1 · lost 20

Qwen3 235B A22B Instructwon 1 · lost 24

Claude Sonnet 5won 1 · lost 24

Grok 4.5won 1 · lost 24

Qwen3.8 Maxwon 1 · lost 24

Muse Spark 1.1won 1 · lost 24

GPT-5.5won 1 · lost 25

GPT-5.6 Solwon 1 · lost 25

Claude Opus 4.7won 1 · lost 25

DeepSeek V4.1 Flashwon 0 · lost 18

Muse Spark 1.2won 0 · lost 18

GLM-5.3won 0 · lost 19

Gemini 3.7 Flashwon 0 · lost 19

Kimi K3won 0 · lost 19

Claude Fable 5.1won 0 · lost 19

Claude Fable 5won 0 · lost 19

Grok 4.7won 0 · lost 22

Claude Opus 5.5won 0 · lost 22

O3won 0 · lost 25

Grok 4.6won 0 · lost 25

GPT-5.6 Terrawon 0 · lost 25

Claude Opus 4.8won 0 · lost 25

Claude Opus 5won 0 · lost 25

Gemini 3.8 Flashwon 0 · lost 25

§ 3 · Sources

Where the numbers come from

11 publications, 26 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-09-26

LMArena Text 1396

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 56

AA-Briefcase ↗1 measureread 2026-09-26

AA-Briefcase 71

LiveCodeBench ↗1 measureread 2026-09-26

LiveCodeBench 27.2%

Aider polyglot ↗1 measureread 2026-09-26

Aider polyglot 55.1%

Kagi LLM Benchmark ↗1 measureread 2026-09-26

Kagi LLM Benchmark 52.3%

SimpleBench ↗1 measureread 2026-09-26

SimpleBench 27.2%

Vals.ai ↗7 measuresread 2026-09-26

Vals · LegalBench 77.73%Vals · CorpFin 54.74%Vals · TaxEval 71.1%Vals · MedQA 82%Vals · GPQA Diamond 61.62%Vals · MMLU Pro 79.47%Vals · AIME 52.2%

HELM Safety ↗5 measuresread 2026-09-26

HELM Safety · HarmBench 49.7%HELM Safety · SimpleSafetyTests 95.3%HELM Safety · Anthropic Red Team 97.1%HELM Safety · BBQ 96.7%HELM Safety · XSTest 97.1%

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

Enkrypt · Jailbreak risk 30.7%Enkrypt · Harmful content risk 47.8%Enkrypt · CBRN risk 12.7%Enkrypt · Toxicity risk 10.4%Enkrypt · Bias risk 86.8%Enkrypt · Insecure code risk 36.4%

Vectara Hallucination Leaderboard ↗1 measureread 2026-09-26

Vectara · Factual consistency 93.9%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=deepseek-v3)](https://publicai.io/model-index/m/deepseek-v3)