Mistral releases a new AI model and a platform to put it to work

Mistral AI has released a new AI model, Mistral Large 4, and a business platform called Mistral Studio to put it to work. The model goes by the nickname Le Chonk. The pitch is simple: companies can use it in the cloud, on their own servers, or a mix of both, and keep control over their data.

What the model does

Mistral says Le Chonk is strong at writing computer code, at defensive cybersecurity, and at reasoning through complex documents and business tasks. It is described as an open model, which Mistral says pushes forward what companies can build. The material does not give test scores, so those claims come from Mistral itself.

What Studio adds

Studio is where a company turns the model into working tools. Mistral describes three kinds. An agent investigates and acts. A workflow coordinates the steps and approvals. An app gives employees a way to use them. All three run on shared infrastructure and connect to company data and systems.

This release adds managed agents and apps. It also adds new features for workflows, company context, agent observability (seeing what an agent did) and evaluations. Developers can also deploy to Studio with one command.

• Agents can carry out multi-step tasks, run code in a sandbox, and hand parts of a job to helper agents.
• Workflows mix fixed rules, AI reasoning and human decisions, and interrupted runs can resume.
• Apps get hosting, sign-in and access controls, so they are easier to ready for company use.

Early results

Mistral points to its own teams. An engineering agent starts investigating production problems, asks for approval before sensitive actions and posts findings in Slack. Mistral says this saves hundreds of hours each month. A product operations workflow sorts internal ideas and customer feedback, which Mistral says saves about 35 hours a week.

Who it is for

Studio is aimed at larger organizations. Mistral says ABANCA, Accenture, Alten, BNP Paribas, Capgemini, CMA CGM and Tata Consultancy Services already use it in production. Studio can run in the cloud, in a company's own VPC (a private, company-controlled area of cloud space), on its own servers, or across several of these.

What is not known

The material does not say what Mistral Large 4 or Studio costs, or how smaller businesses can get access. The savings figures come from Mistral's own teams, not independent testing. A business weighing AI tools can note that running them on its own systems is now a stated option here.