‹ PublicAI Index
The LLM benchmark aggregator.
Mistral Large 4
Mistral
Strongest in Biology research (#19 of 27), weakest in Jailbreak resistance (#235 of 271). Above par in 12 of 29 scopes. Among the models it meets almost everywhere, it finishes behind Kimi K3 and DeepSeek V4.1 Flash and ahead of Gemini 3.6 Flash and GPT-4.1 Mini.
§ 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.
Coding50.8−15.6#77/1763/6
Agentic coding50.8−16.2#74/1693/5
Professional50.5−12.6#80/1872/2
Biology research44.9−26.2#19/271/1
Legal55.6−10.8#35/1621/1
Document design39.1−28#39/471/1
Finance54.6−7.9#55/1611/1
Medical49.8−16.7#89/1491/1
Human preference59.8−8.8#91/3461/1
Human preference59.8−8.8#91/3461/1
Reasoning47−17.8#117/1893/5
Reasoning49.5−17.2#70/1431/3
Mathematics45.2−18.8#106/1492/2
Core abilities44−21.2#177/2253/4
Data analysis55−10.9#21/471/1
Instruction following44.4−29.4#42/621/1
Language26−40.6#61/621/1
General intelligence47.3−21.5#128/2142/3
Safety49.4−12#199/3311/3
Fairness55.4−14.6#64/2961/2
Secure code58.7−7.6#77/2731/1
Toxicity avoidance54.8−4.2#111/2711/1
Harm refusal46.5−15#214/2961/2
Jailbreak resistance34.7−30.3#235/2711/1
Agents44.2−21.7#224/2893/6
Knowledge work62.3−9.5#32/1982/2
Long-horizon work33.9−40.1#44/471/1
Web research30−33.7#48/481/2
Computer use35.9−32#70/801/2
§ 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 Mistral Large 4, left for the other.
§ 3 · Sources
Where the numbers come from
8 publications, 35 figures. Every one links to the page it was read from.
LMArena Text 1432
Artificial Analysis Intelligence Index 38
GDPval-AA 1422
AA-Briefcase 1392
LiveBench 47.7LiveBench · Agentic 11.8LiveBench · Reasoning 83.9LiveBench · Coding 60LiveBench · Browser Use 8LiveBench · Data Analysis 76.5LiveBench · Research 28.1LiveBench · Design 65.6
LiveBench 1.0 71.8LiveBench 1.0 · Reasoning 83.9LiveBench 1.0 · Coding 77.2LiveBench 1.0 · Agentic Coding 57.2LiveBench 1.0 · Mathematics 93.6LiveBench 1.0 · Data Analysis 76.5LiveBench 1.0 · Language 49.6LiveBench 1.0 · Instruction Following 64.8
Vals · Legal Research Bench 31.73%Vals · Harvey Legal Agent Benchmark 15.83%Vals · Finance Agent 54.68%Vals · MedCode 40.69%Vals · MedScribe 80.43%Vals · BioMysteryBench 67.04%Vals · Vibe Code Bench 78.4% (OpenHands)Vals · Code Migration 30.56%Vals · ProofBench 10%
Enkrypt · Jailbreak risk 25.5%Enkrypt · Harmful content risk 1.1%Enkrypt · CBRN risk 49.3%Enkrypt · Toxicity risk 2.9%Enkrypt · Bias risk 69.3%Enkrypt · Insecure code risk 15.6%
Badge
[](https://publicai.io/model-index/m/mistral-large-4)