Mistral’s open source approach: a breath of fresh air to AI… while it lasts?

Mistral.ai is an interesting case study in the nascent AI industry; it was founded by Arthur Mensch, a young French PhD engineer who worked for two and a half years at DeepMind, with two colleagues from the School of Engineering with experience at Google and Meta.

The company is less than nine months old, and raised €105 million at its first funding round, when it was valued at €240 million, before starting its activities: It shows that Europe can be entrepreneurial, and intends to compete directly with Google and OpenAI.

Just three months ago, the company closed another €385 million funding round, which put its value already at around €2 billion, unusual for a nine months old company. How does a company grow like this in less than six months? Timing; it has arrived at the height of the AI gold rush, when all the world’s investors are furiously competing to put their money into this area.

Five months ago, the company introduced its first large language model, Mistral 7B, free to use, open source and capable of working in English and with code. Since then, it has launched other products, some also in open source: Mistral 8x7B, for many the best model in real open source (not like Llama, which isn’t really), able to work in English, French, German, Spanish, Italian…