Arthur Mensch

Arthur Mensch studied at École polytechnique and École normale supérieure, took a PhD at Inria on stochastic optimization for analyzing brain-imaging data, and worked as a postdoc at ENS before joining DeepMind's Paris office in late 2020. There Mensch worked on large language models and was an author of three of DeepMind's main papers of the period: RETRO, which let a model look up passages from a two-trillion-token database; Flamingo, a model that combined images and text; and Chinchilla (2022), which showed that most large models were undertrained and that parameters and training tokens should grow in equal proportion. The paper's 70-billion-parameter model beat DeepMind's own 280-billion-parameter Gopher.

In spring 2023 Mensch left DeepMind to found Mistral AI in Paris with Guillaume Lample and Timothée Lacroix, friends since their student days who were then researchers at Meta's Paris lab and had worked on its first LLaMA models. Mensch became chief executive. When the company was four weeks old it raised a seed round of 105 million euros, with a pitch built on open models, and in September 2023 it released Mistral 7B under the Apache 2.0 license.

During the final negotiations on the EU AI Act in November 2023, Mensch argued that the law should regulate applications rather than the makers of foundation models, telling TechCrunch that "the deployer should bear the risk." Corporate Europe Observatory, a lobby watchdog, later accused Mistral and its adviser Cédric O, a former French digital minister, of helping France weaken the Act; it quoted Mensch telling Le Monde that in its final form the law was "perfectly manageable for us." Mensch also sat on the French government's generative AI committee.

Under Mensch, Mistral has sold itself to European governments and companies as a sovereign alternative to American labs, while also partnering with Microsoft, which expanded its deal with Mistral in July 2026. In September 2026 Mistral raised 3 billion euros at a valuation above 21 billion euros, and Mensch told CNBC the company would own far more of its own computing capacity.