‹ PublicAI Index
The LLM benchmark aggregator.
DeepSeek V3
DeepSeek · 685B · open weights
Strongest in Code generation (#19 of 20), weakest in Agents (#295 of 298). Above par in 6 of 27 scopes. Among the models it meets almost everywhere, it finishes behind O3 and Claude Opus 4.8 and ahead of Gemma 4 31B and Gemini 2.5 Flash-Lite.
§ 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.
Core abilities49.9−15.4#99/1991/4
General intelligence49.8−19.6#89/1781/3
Professional49.3−13.1#107/1941/2
Medical49.2−17.3#93/1491/1
Legal49.6−16.3#94/1731/1
Finance48.8−13.6#103/1621/1
Coding46−20.2#114/1792/6
Code generation32−33.3#19/201/1
Agentic coding52.7−14.2#62/1731/5
Knowledge49.1−13.4#125/1971/3
Academic knowledge47.8−16.1#88/1231/1
Human preference56.2−12.6#137/3541/1
Human preference56.2−12.6#137/3541/1
Reasoning42.1−23.3#161/1952/5
Science39.9−23.6#99/1221/1
Mathematics46.3−17.5#102/1521/2
Reasoning39.3−29.7#123/1471/3
Safety47.2−14.2#243/3293/3
Safe-prompt compliance56.6−3.8#23/821/1
Factual grounding59.8−10.9#25/1041/1
Fairness50.7−19.3#103/2922/2
Secure code48.4−17.8#171/2701/1
Harm refusal45.6−15.9#229/2922/2
Toxicity avoidance43.8−15.2#231/2681/1
Jailbreak resistance28.7−36.3#253/2681/1
Agents38.8−27.6#295/2982/6
Knowledge work32.1−38.6#197/2082/2
§ 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.
§ 3 · Sources
Where the numbers come from
11 publications, 26 figures. Every one links to the page it was read from.
LMArena Text 1395
GDPval-AA 77
AA-Briefcase 69
LiveCodeBench 27.2%
Aider polyglot 55.1%
Kagi LLM Benchmark 52.3%
SimpleBench 27.2%
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 · 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 · 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 · Factual consistency 93.9%
Badge
[](https://publicai.io/model-index/m/deepseek-v3)