Skip to content
ML Atlas

Mistral AI Models

Language models from the Paris start-up Mistral AI, known for efficient open-weight releases such as Mistral 7B and the mixture-of-experts Mixtral 8x7B.

Also Mistral 7B · Mixtral · Mixtral 8x7B · Le Chat

Role · kind
Real systemsFamous system
Level
Beginner

Mistral AI, Paris, 2023

A European AI lab that showed small, cleverly built models can match much bigger ones, and let anyone download several of them.

Where it fits

Large Language ModelLarge Language ModelMixture of ExpertsMixture of ExpertsTransformerTransformerMistral AI Models
+All relationships as a list

Is a kind of / part of

See it on the full map

How it works

Mistral AI was founded in Paris in 2023 by researchers who had worked at Google DeepMind and Meta. Within months it released Mistral 7B (September 2023) under the permissive Apache 2.0 license, simply posting a download link.

Two releases made its name:

  • Mistral 7B: a compact Transformer that outperformed larger open models on many benchmarks, helped by efficient attention variants (grouped-query and sliding-window attention).
  • Mixtral 8x7B (December 2023): a mixture-of-experts model. Each layer has 8 expert networks and a router sends each token to 2 of them, so only about 13B of its roughly 47B parameters work on any given token. That gives the knowledge of a big model at the running cost of a much smaller one.

Like Llama, its open-weight models can be downloaded, fine-tuned and run privately, which matters to companies and governments that want to keep data in-house. Mistral also sells larger proprietary models and runs the Le Chat assistant.

Everyday use cases

  • Case 01

    How you've probably met it

    Through the Le Chat assistant, or inside apps and on-premise deployments that use its downloadable models.

Strengths, weaknesses and when to use it

Strengths

  • Efficient models with permissive licenses for several releases; a major European AI player.

Weaknesses

  • Like other LLMs, can hallucinate; not every model in the lineup has open weights.

When to use itAs an example of efficient open-weight models and of mixture-of-experts in practice.

Compare with

Go deeper

Sources

  1. [1]Jiang et al., Mistral 7B (2023)
  2. [2]Mistral AI, Mixtral of experts (2023)
  3. [3]Jiang et al., Mixtral of Experts (2024)

Facts checked as of 2026-10-12