Mistral AILanguage modelPreviewOpen weightsNew

Mistral Large 4

Europe's trillion-parameter answer: 160+ languages, weights promised to developers.

Compare head-to-head Race it in 3D

Overall Intelligence

#28

of 56
12.1 pts behind #1

races led

weeks at #1

76.6

score

Released: Oct 6, 2026; 2 wk ago
Context: 256K; 64K max output
Price: $2 / $6; per 1M in/out · $3.00 blended
Speed: 108 tok/s; 1.2s to first token
Modalities: text · image; outputs text
Weights: Open; 1T MoE (49B active)

The 30-second version

Europe's trillion-parameter answer, in preview: 160+ languages, with weights promised to developers.

Key takeaways

1Exceptional multilingual coverage.
2EU-hosted option for data-sovereignty needs.
3Preview: capabilities and licence terms may change.
4Behind the US frontier on reasoning and coding.

What you need to know

Purpose

A sovereign European frontier model for multilingual enterprises and governments.

What Mistral optimised for

Multilinguality, efficiency and European data sovereignty.

Unique strengths

Over 160 languages.
European hosting and governance.
Planned open weights.

Insight

How it differs from rivals. The leading choice where EU hosting and broad language coverage matter more than peak benchmark scores.

Where it ranks

Current snapshot · illustrative scores · click any race to watch it

Large 4Best on each axis

Overall

28Overall+12.1 pts

Coding

33Coding+15.6 pts 29Frontend & UI+83 Elo 31Backend & APIs+13.6 pts 33Distributed Systems+16.8 pts 31Full-Stack Apps+15.2 pts 33Debugging & Repair+16.7 pts 37Agentic Coding+18.7 pts 29Code Review & Security+13.2 pts 32Data Science & SQL+12.8 pts 31Mobile Apps+13.0 pts 35Terminal & DevOps+19.9 pts 32Competitive Programming+439

Hard Reasoning

29Hard Reasoning+14.8 pts 32Frontier Questions+26.4 pts 28Abstract & Novel Reasoning+20.3 pts 32Logic & Puzzles+12.6 pts 33Planning & Strategy+14.6 pts

Math

35Math+16.5 pts 34Competition Math+12.3 pts 35Research Math+20.4 pts 36Applied & Quantitative+15.0 pts

Science

30Science+14.6 pts 27Graduate Science QA+8.6 pts 31Biology & Medicine+12.8 pts 34Physics & Chemistry+13.5 pts 31Research Assistance+13.2 pts

Writing & Language

23Writing & Language+10.8 pts 26Creative Writing+78 Elo 31Technical Writing+11.6 pts 25Editing & Style+9.1 pts 1Multilingual92.9 31Instruction Following+10.1 pts 32Conversation & EQ+84 Elo

Multimodal

30Multimodal+22.8 pts 25Image Understanding+12.0 pts 26Documents & Charts+12.6 pts 31Video Understanding+26.6 pts 31Speech & Audio+29.1 pts

Tool Use & Agents

33Tool Use & Agents+16.6 pts 33Tool Use+12.0 pts 35Computer Use+26.3 pts 30Web Research+17.6 pts 33Long-Horizon+25.1 hrs

Long Context

26Long Context+15.1 pts 271M Recall+12.1 pts 29Long Reasoning+13.8 pts

Speed & Cost

13Speed & Cost+31.6 pts 21Output Speed+419 tok/s 14Latency+0.88s 27Cost-Efficiency+44.7 pts

Trust & Factuality

31Trust & Factuality+14.4 pts 31Factuality+12.5 pts 29Honesty & Calibration+11.3 pts

Overall Intelligence over time

#1 frontierLarge 4GLM-5.3V4 Pro

Best use cases

Multilingual products

Support and content across many European and global languages.

Public sector

Deployments that require EU data residency.

Strengths

Languages
Edges the multilingual race in a photo finish.

Weaknesses

Frontier gap
Clearly behind on hard reasoning and agentic coding.

When to choose it, and when not to

Large 4overGPT-6.1 Sol

EU data residency or rare-language coverage is required.

Pick something else when…

Top-end coding and reasoning tasks.

Update history

The distilled version first; open any release for the deeper breakdown.