Mistral Releases Downloadable Large 4 AI Model With Coding and Cybersecurity Focus
The French AI company's upgraded open-weight model can run on an organization's own systems, following a public preview earlier this month.
French artificial-intelligence company Mistral released the downloadable weights of its Large 4 model on Thursday, offering organizations an upgraded system focused on coding and defensive cybersecurity.
The release follows a public preview earlier this month. Mistral says it used feedback from developers, governments and corporate users, along with three more weeks of training, to improve the version now being made available.
Large 4, nicknamed Le Chonk, has one trillion parameters and is designed to work with more than one type of input. Releasing its weights means organizations can download and run the model on their own infrastructure rather than relying only on a hosted online service.
Mistral says the updated model improves coding performance and can help security teams identify and fix software vulnerabilities. Those performance statements are the company's claims; results on individual benchmarks do not establish that a model will be best for every task or organization.
The model offers three reasoning settings: low for simpler work, medium for everyday tasks and high for more demanding problems. Mistral says the choices let developers balance computing cost and response quality.
The open-weight approach gives organizations more control over deployment and customization, but it also means access cannot be contained once the model is released. Downloadable weights should not be confused with a guarantee that every part of the model's training process is openly available.
Mistral's release page includes technical details, benchmark results and links for developers who want to use the new version.