By Roman Aglaganov
Mistral AI was founded in Paris in April 2023 by three French researchers: Arthur Mensch from Google DeepMind and Guillaume Lample and Timothée Lacroix from Meta. Weeks later, with no product and a short pitch memo, it raised a seed round of about €105 million, then one of the largest in European history.
The launch that was just a link
In September 2023, Mistral released its first model, Mistral 7B, with a post on X containing nothing but a magnet link, the address format used by BitTorrent. No blog post, no keynote, no countdown. The model was small, fast, released under the permissive Apache 2.0 licence, and outperformed larger rivals on many tests. The AI world, used to glossy launch events, was charmed.
In December 2023 it did it again with Mixtral 8x7B, a mixture-of-experts model: instead of using the whole network for every word, it routes each token to a couple of specialised sub-networks, making it much cheaper to run for its size.
Europe’s champion
Mistral has since built a mix of open and commercial models and its own chatbot, launched as Le Chat in 2024 and renamed Mistral Vibe in 2026. It has become the flag-bearer for European “digital sovereignty”, the idea that Europe should not depend entirely on American or Chinese AI. In September 2025, ASML, the Dutch maker of chipmaking machines, invested €1.3 billion in the company.
The lesson: you do not need a stadium and a keynote to launch a model. Sometimes a 2000s-era download link and a good model are enough.
