A solid pick for teams wanting Apache 2.0 open models in every size, from laptop-friendly Ministral to the large mixture-of-experts flagship.

Kenning's take

What it does

Mistral 3 is a family of open-weight models released under the Apache 2.0 licence. It includes three small Ministral 3 models (3B, 8B and 14B) built to run on local hardware, and Mistral Large 3, a mixture-of-experts model with 41B active and 675B total parameters. All of them read images as well as text, and they are available through Mistral Studio, cloud platforms such as Amazon Bedrock and Azure Foundry, and as downloads on Hugging Face.

  • Ministral 3 in 3B, 8B and 14B sizes
  • Mistral Large 3 with 41B active and 675B total parameters
  • Base, instruct and reasoning variants of each Ministral size
  • Image understanding across the family
  • Apache 2.0 licence
  • 256k context window in the API

Who it suits. Developers and companies who want open models they can run themselves or call cheaply through an API.

Key facts

Category
AI models
Made by
Mistral AI
Based in
France
Released
2 December 2025
How to get it
open weights on Hugging Face, and through the Mistral API in Mistral Studio and cloud platforms including Amazon Bedrock, Azure Foundry, IBM watsonx and OpenRouter
Works on
API, self-hosted
In the UK
Yes
Free plan
open weights are free to download under Apache 2.0
API price
Mistral Large 3 $0.50 per million input tokens, $1.50 per million output; Ministral 3 14B $0.20, 8B $0.15, 3B $0.10 per million tokens

FAQ

When was Mistral 3 released?

Mistral announced the Mistral 3 family on 2 December 2025.

Is Mistral 3 open source?

Yes. All Mistral 3 models are released under the Apache 2.0 licence, and the weights are on Hugging Face.

What sizes does Mistral 3 come in?

Ministral 3 comes in 3B, 8B and 14B sizes, and Mistral Large 3 has 41B active and 675B total parameters.

Facts from the maker's own pages and the sources listed. Checked 2026-10-11.