
Independent company · Paris, France
Mistral AI
A Paris AI company that pairs open-weight models under permissive licenses with a commercial platform, positioning itself as Europe's frontier lab.
Paris, France
- Founded
- 2023
- Type
- Independent company
- Based in
- Paris, France
- Key papers
- 4
- Featured projects
- 8
About
Mistral AI was founded in April 2023 in Paris by Arthur Mensch, Guillaume Lample and Timothée Lacroix, who had met at École polytechnique and worked at Google DeepMind and Meta. Its first model, Mistral 7B, was released on 27 September 2023 under the Apache 2.0 license.
The company builds language, coding, speech, document and forecasting models, and sells them through its Studio and Vibe products and through sovereign deployments for enterprises and governments. Recent open-weight releases include Mistral Small 4 under Apache 2.0, Voxtral TTS and the one-trillion-parameter Mistral Large 4.
Mistral is venture-funded and commercial. ASML took an 11 percent shareholding in 2025, and the firm is often cited as a leading beneficiary of European digital sovereignty policy.
Mission and ambitionsMistral says its goal is to put frontier AI in everyone's hands, keeping it open and giving organizations control and ownership of their AI systems.
Known for
- 01
Mistral 7B and Mixtral open-weight models
- 02
Apache 2.0 licensed releases
- 03
Europe's largest AI startup
- 04
Sovereign AI deployments for enterprises and governments
Milestones
- 2023
Founding in Paris
Mensch, Lample and Lacroix established the company in April 2023.
- 2023
Mistral 7B
A 7.3-billion-parameter model released under Apache 2.0 that outperformed Llama 2 13B on the benchmarks Mistral reported.
- 2025
Magistral Small
The first open reasoning model from Mistral, released in June 2025.
- 2025
Mistral Large 3 and Ministral 3
A 675-billion-parameter sparse mixture-of-experts model and three small dense models released in December 2025.
- 2026
Mistral Small 4
A multimodal hybrid model with a 256,000-token context window released under Apache 2.0 in March 2026.
- 2026
Mistral Large 4
A one-trillion-parameter mixture-of-experts model whose weights were released on 29 October 2026.
Open source
Official inference library for Mistral models (10,814 stars)
Minimal CLI coding agent by Mistral (5,401 stars)
Open-source repository (3,089 stars)
Open-source repository (2,367 stars)
Model: 1,815,007 downloads, 3843 likes.
Model: 1,546,100 downloads, 1062 likes.
- mistralai/mmlu_speechdataset
Dataset: 483 downloads, 20 likes.
Space: 225 likes on the Hub.
Research papers
- Mistral 7B2023
- Pixtral 12B2024
- Voxtral2025
Notable releases
Openness
Many models are released as open weights with Apache-2.0 licences, plus open inference and finetuning code; some flagship models are API-only.
People
- Arthur MenschCo-founder and CEO
- Guillaume LampleCo-founder and Chief Science Officer
- Timothée LacroixCo-founder and CTO
Listed from the lab's own pages and public profiles.
Contact and community
Press address printed on the website; Discord community for developers.
Official Discord community for developers using Mistral models and API.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



