Glossary
What is Mistral AI?
In short
Mistral AI is a French company that builds language models, including some with open weights. It also makes the Le Chat assistant and tools for developers.
Companies · Mistral AI
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine three friends who opened a workshop of talking helpers. They give many of their creations to anyone who wants them, together with the drawings: take one, build it at home, improve it. The most complex ones they keep and help others use. This is how their helpers end up in small clubs, in big schools and even in factories.
In plain words
The Paris company appeared in April 2023; its founders are Arthur Mensch, Guillaume Lample and Timothee Lacroix. The first model, Mistral 7B with 7 billion parameters, came out under the Apache 2.0 license. You can download the weights, run them yourself and build them into your product without fees. Larger models followed, then the Le Chat assistant and a coding helper now called Vibe. For example, a coding school puts a small model on its own server to check students’ homework and explain their mistakes.
For business
Customers use the company’s models for assistants, document search, text recognition and coding; its website names Stellantis, HSBC, BNP Paribas and the French Ministry of the Armed Forces among its clients. The official pages give no list of countries where Le Chat and the API are available. The commercial terms of 25 September 2026 require compliance with EU, US and Singapore sanctions and export law. Russia is not named directly, so a lawyer reads the vendor contract first. Models under Apache 2.0, such as Mistral 7B and Mixtral 8x7B, offer a separate path: they run inside your own network, and the data never leaves the company. How many GPUs the right model needs is calculated when you choose a server.
For a specialist
The Mistral 7B paper (Jiang et al., 2023) showed that a model with 7 billion parameters beats Llama 2 13B on all the authors’ benchmarks. Grouped-query attention (GQA) gives speed, and sliding window attention (SWA) makes long sequences cheaper to process. Mixtral 8x7B (2024) is a sparse mixture of experts: each layer has 8 feed-forward blocks, and a router sends each token to two of them. So only 13 billion of its 47 billion parameters work at inference, and the context window is 32,000 tokens. Both models are open under Apache 2.0. On 9 September 2025 the company announced a EUR 1.7 billion Series C round at a post-money valuation of EUR 11.7 billion. The lead investor was ASML, the maker of lithography equipment, which also agreed a strategic partnership with it.
Example
A law firm runs Mistral 7B on its own server and gives it first-draft summaries of contracts; a lawyer checks the final text.
Related terms
Questions and answers
Sources
- Mistral AI, About, checked 16 October 2026mistral.ai
- Mistral AI, Commercial Terms of Service, in force from 25 September 2026legal.mistral.ai
- Jiang et al., Mistral 7B, 2023arxiv.org
- Jiang et al., Mixtral of Experts, 2024arxiv.org
- Mistral AI, Mistral AI raises 1.7B€ to accelerate technological progress with AI, 09.09.2025mistral.ai
Not sure whether your company needs this?
AI solutions map