Skip to content

Pick your style: Denkbaar

Book a conversation

Articles

History

Mixtral 8x22B under Apache 2.0

·

Mistral AI releases an open mixture-of-experts model that you may use without restrictions.

A technician examines eight modules, two of them purple and connected to a screen.
Mixtral uses only part of the model at each step. Illustration: Denkbaar.

On 17 April 2024, Mistral AI publishes its announcement of Mixtral 8x22B, an open model.

What happened

  • Mixtral 8x22B is a sparse mixture-of-experts model. At each step, it uses only part of its parameters: 39 billion of 141 billion.
  • Mistral releases it under Apache 2.0. That licence lets you use the model anywhere, also in business.
  • The model speaks English, French, Italian, German and Spanish, and has a context window of 64,000 tokens.
  • It can call functions by itself. With that, you connect it to your own software.
Blue lines run through two purple modules on a board with eight modules.
A mixture-of-experts model selects part of its parameters at each step. Illustration: Denkbaar.

Why it mattered

A strong model with a real open source licence is a different thing from a model with only open weights. With Apache 2.0, you can put it on your own server without a separate agreement.

A licence booklet lies beside a model block and a server connected to a laptop.
Apache 2.0 also allows business use on your own server. Illustration: Denkbaar.

What you can do with it now

Put the licence of every model you consider in your comparison. A good score does not help you if you may not use the model in your business.

Written afterwards on 12 November 2026, from the source below.

Sources

  1. Mistral AI: Cheaper, Better, Faster, Stronger

Topics: open-sourcemodelsmistral