Technology

France's Giant New AI Is Here, but You Can't Download It Yet

Martin HollowayPublished 2month ago2 min readBased on 8 sources
Reading level
France's Giant New AI Is Here, but You Can't Download It Yet
Photo by Sergei Starostin on Pexels

Mistral AI has released Mistral Large 4, a very large AI system nicknamed Le Chonk that can work with text, code and images. The French company says it aims to move past American and Chinese rivals. TechCrunch

Large 4 cannot be downloaded yet. Developers can only try it through Mistral's controlled test access, with an API preview open. The Next Web It is like a test drive before a sale. Mistral plans to publish the full model files, called weights, on Oct. 27, three weeks after release, once safety testing is complete.

The training detail matters most. Mistral says it trained Large 4 on its own computers using only 4,000 NVIDIA GPUs, the specialized chips used for AI. That is a small number for a model with one trillion settings, or parameters. It suggests careful planning to split up the work and keep a long training run stable, rather than simply using many more machines.

Mistral says Large 4 is built for cybersecurity, finance and chip design. Early test scores are reported to lead in coding and vision, or understanding images. The Next Web Mistral's CEO said the newest model beats Chinese models in some areas, including cybersecurity. Reuters

Mistral presents the release as a European attempt to compete with closed U.S. models and open Chinese models. Mistral and other European AI firms have accused U.S. rivals of using safety concerns to protect their lead. Reuters

Mistral raised 3 billion euros at a value of around 21 billion euros (24 billion dollars). Reuters It said it would build a new data centre in Les Ulis, France, with 10 megawatts of computing power in the second half of 2026. Microsoft agreed to fund Mistral's European AI expansion in a multibillion-dollar deal, and Mistral added Medium 3.5 and OCR 4 models to Microsoft Foundry. Reuters

The broader context here is what efficient training on 4,000 chips would mean if outside checks confirm it. The raw number of settings says little on its own. Experts will still ask how much computer memory it needs to run, how well it handles code and European languages, how reliably it uses outside tools, and how the current test version compares with the downloadable version.

In my view, the step-by-step release is practical. The controlled access gives Mistral time for safety testing while developers try the system for cybersecurity, finance and chip design. The tradeoff to note is that companies cannot yet check the full files, repeat results, or plan hosting costs until Oct. 27. If the early leads in coding and vision hold up, more teams could get high-level help with code review, security checks, money analysis and hardware design, including European users with a system built closer to home.