Mistral Models Explained

Small, Medium, Large

Mistral ships a ladder: small efficient models for chat and edge devices, mid-size all-rounders for most work, and large flagships for the hardest reasoning and code. Bigger is better, smaller is cheaper and faster.

Specialists

Code-tuned variants for programming, embedding models for search and RAG, moderation models for safety filtering, and vision-capable models for images-plus-text. Pick the specialist, not the flagship, for routine jobs.

How to Choose

Prototype on the flagship, deploy on the smallest model that passes your tests. Benchmark on your data: leaderboards predict little about your workload.