Mistral Opens Le Chonk, China Still Leads
Mistral released Le Chonk's weights. It leads open models at coding but scores 26 to the 32 of DeepSeek V4.1 Flash, now on Mistral's cloud.
Read the storyCo-founder and CEO, Mistral AI
Arthur Mensch leads Mistral AI, a European company building models, developer tools, applications, and AI infrastructure. His research background includes machine learning for brain imaging and co-authorship of the Chinchilla compute-scaling paper. Since co-founding Mistral, he has developed a business spanning open-weight releases and enterprise services, with an emphasis on customers retaining control over their models, data, and deployments.
Get AI Secret’s free daily briefing, including Arthur Mensch and the wider AI industry.

Arthur Mensch leads Mistral AI, a European company building models, developer tools, applications, and AI infrastructure. His research background includes machine learning for brain imaging and co-authorship of the Chinchilla compute-scaling paper. Since co-founding Mistral, he has developed a business spanning open-weight releases and enterprise services, with an emphasis on customers retaining control over their models, data, and deployments.
Sources checked . Selected career background and milestones.
Selected research, leadership and career milestones, newest first.
Mistral introduced Forge, a system for enterprises to train models on proprietary knowledge, extending its offering from general-purpose models toward organization-specific training and customization.
Source: Mistral AI ↗Mistral announced an infrastructure offering spanning GPUs, orchestration, APIs, and services, describing a plan to give enterprises and research organizations greater control over their AI stack.
Source: Mistral AI ↗Mistral released its flagship language model through its own platform and Microsoft Azure, expanding commercial distribution alongside the company’s separate open-weight model offerings.
Source: Mistral AI ↗Mistral released its first widely distributed language model under Apache 2.0, making weights available for local deployment and adaptation rather than limiting access to an API.
Source: Mistral AI ↗Mistral’s official company history dates its founding to April 2023. Mensch leads the company alongside co-founders Guillaume Lample, chief science officer, and Timothée Lacroix, chief technology officer.
Source: Mistral AI ↗Mensch co-authored research on compute-optimal language-model training, showing how model size and training data should be balanced under a fixed compute budget and introducing Chinchilla.
Source: arXiv / paper authors ↗Mensch’s Paris-Saclay doctoral thesis studied learning representations from functional MRI data, establishing a research foundation in statistical learning, optimization, and the analysis of brain-imaging datasets.
Source: HAL / author research record ↗Mensch and collaborators presented compressed online dictionary learning for fMRI decomposition, exploring ways to reduce the computational demands of extracting structure from large brain-imaging datasets.
Source: arXiv / paper authors ↗Briefs, articles and relevant sections, newest first. Explore older coverage below.
More coverage
Mistral released Le Chonk's weights. It leads open models at coding but scores 26 to the 32 of DeepSeek V4.1 Flash, now on Mistral's cloud.
Read the storyDiscover OpenAI's former CTO's $100 million valued all-star team.
How would you feel if we all hired AI civil servants?