Mistral AI
French company making open-weight and commercial AI models
Profile
- Country
- France
- Headquarters
- Paris
- Founded
- April 2023
- Founders
- Arthur Mensch, Guillaume Lample, Timothée Lacroix
- Type
- Private simplified joint-stock company (SAS)
Mistral AI is an artificial intelligence company headquartered in Paris, France. It was founded in April 2023 by Arthur Mensch, Guillaume Lample and Timothée Lacroix. Mensch had worked on AI at Google DeepMind, and Lample and Lacroix at Meta (Facebook's parent company). Its legal form is a French simplified joint-stock company (SAS), and it is not listed on a stock exchange. In September 2025, the Dutch chip-equipment maker ASML invested in it and became its largest shareholder, with about 11%. As of October 2026, its CEO is Mensch.
Mistral sells AI models, and services built on them, to businesses and other organisations. It names finance, manufacturing, defence, energy and the public sector as its key customer sectors. Developers use its models through an API (a channel that lets programs call an AI model) on ‘Mistral Studio’, and members of the public use the AI assistant ‘Vibe’. Vibe was originally a chatbot called ‘Le Chat’; it was renamed in May 2026 as features such as coding were added, and it has a free plan and paid plans. Mistral also runs a cloud service called ‘Mistral Compute’. In September 2025 it agreed a long-term partnership with ASML to explore using AI models across ASML's chip-making equipment and its research, development and operations.
Mistral makes both open-weight models (models released for anyone to download and use) and commercial models that are not released. Its official documentation lists Mistral Large 3 and Mistral Small 4 under ‘Apache 2.0’ and Mistral Medium 3.5 under a ‘Modified MIT’ licence, while models such as the code-writing model Codestral are labelled ‘Premier’ instead of an open licence. It also has ‘Voxtral’ models, which turn speech into text or read text aloud. In October 2026 it released Mistral Large 4, which understands text and images together, as an API preview. The model has about 1 trillion parameters (the numbers a model adjusts as it learns), and at the end of that month Mistral released the downloadable model files (the weights) on Hugging Face. They come under the 'Mistral Ecosystem Open-Weight' licence, a modified MIT licence; businesses that offer the model to others as a service need a separate commercial licence from Mistral.
Sources for this profile (10) · as of Nov 1, 2026
- About Mistral · Mistral AI
- Legal (company identification) · Mistral AI
- Mistral AI: the French underdog taking on US tech giants as Europe's hope in global generative AI race · Malay Mail (ETX Studio, with AFP)
- Chip giant ASML invests €1.3bn in French AI champion Mistral · Euronews
- ASML, Mistral AI enter strategic partnership · ASML (press release)
- Vibe gets to work. · Mistral AI
- Models Overview · Mistral AI Docs
- Introducing Mistral Large 4 · Mistral AI
- Mistral Large 4 (Le Chonk) is now open weight · Mistral AI
- mistralai/Mistral-Large-4-1T-A52B · Hugging Face
AIMade by AIThis profile was written by AI (Claude) from the 10 sources above (official pages, filings, news reports).How it's made
Records 22
Newest first. Open one to see it with its source.
2026 11
A 1 trillion parameter model has appeared on the open-weight side putting coding and defensive security forward together, and the focus of the company's figures is not the score but how few tokens it takes to reach the s
Company-reported, the 209M model ranks No. 1 in the pretrained zero-shot categories of GIFT-Eval, TIME and fev-bench, and the 28M model beat Chronos-2, which is about four times larger. On an 8-thread server CPU the 5M m
A trend in which the form of contract shifts from "selling a model" to "training a dedicated model on the customer's data", with air-gapped deployment rising to a commercial requirement.
Shows that the battlefield for coding agents has moved from new development to modernizing legacy code, a core corporate asset, and that the combination of checking harnesses and human gates is hardening into a working p
A turning point at which Europe's open-weight, sovereign AI camp secured capital to rival US big tech, notable for the direct participation of semiconductor supply-chain companies such as Samsung Electronics, NVIDIA and
Following xAI, Mistral too partners with HUMAIN — a setup in which Gulf capital secures labs from different camps at the same time. Mostly at the review and planning stage
A financing structure that converts customers' advance purchases into infrastructure equity. 1GW is a target for 2030, not capacity currently in operation
A design that ships a safety classifier as a small open-weight model and injects the policy through the prompt to fit each organization's standards
An LLM lab expanding into replacing industrial simulation. No named model release at the time of the announcement
Brought the trend of merging separate model families into a single set of weights to open weights
A move toward mechanically checking AI-generated code against formal specifications. Shifted the evaluation axis from competition math to contributions to real formalization projects
2025 5
Released a 675B-class frontier model under Apache 2.0. The largest fully open release in the West as of late 2025
Europe's largest AI funding round. That the lead was not a VC but the sole supplier of EUV equipment reveals its industrial-policy character
A model lab expanding into a European sovereign compute provider. The only GPU figure given is a range: tens of thousands
The first open-weight reasoning model family from a European lab. A paper on the training infrastructure and RL algorithm was published at the same time
Mistral recommitted to Apache 2.0 for its general-purpose models while moving away from MRL-licensed ones; company-reported over 81% on MMLU at 150 tokens/s
2024 4
Published the weights of a 123B flagship designed for single-node inference, limited to research and non-commercial use, with a commercial license required for self-deployment
Introduced a non-production license that publishes weights for research and testing only, with commercial use requiring a separate license
Mistral's larger MoE follow-up to Mixtral 8x7B, again under Apache 2.0
An API flagship launched with Microsoft Azure as Mistral's first distribution partner; Mistral claimed it ranked second only to GPT-4 among models available through an API
2023 2
Made Sparse MoE a standard architecture of the open ecosystem and imprinted the concept of total versus active parameters
A truly unrestricted license and a challenger from Europe; later the base for thousands of fine-tunes