Le Chonk brings the brains. Studio puts it to work.

Summary

Mistral announced on 2026-11-05 a set of new features in Mistral AI Studio, its platform for building on Mistral Large 4 ("Le Chonk"). The main ones are managed agents with skills, sandboxes, subagents, and hooks; durable workflows that mix rules, agent steps, and human approvals; hosted apps; and a Mistral MCP server so you can build from an existing coding agent and deploy to Studio with one command.

Key points

• Managed agents: multi-step agents with reusable skills, code sandboxes, subagents, and hooks for custom logic, with limits on which tools, data, and actions each agent can use. This matches the managed-agent design of Anthropic and OpenAI.
• Workflows: fixed rules, agent reasoning, and human approvals in one graph; runs resume after interruption; event triggers from enterprise systems and a visual graph.
• Apps: hosted apps with authentication and access controls that can call agents, workflows, and enterprise search.
• Mistral MCP server: build with Mistral Vibe or another coding agent, test locally, and deploy to Studio with one command. The post does not give the install command.
• Observability and evals: OpenTelemetry traces for model calls, tool use, and approvals; evaluations against your own datasets.
• Deploys in Mistral's cloud, your VPC, on-premises, or a mix. No prices or availability tiers are given in the post.

Why it matters

Mistral now offers the same managed-agent building blocks (skills, subagents, hooks, MCP) as the US providers, with on-premises deployment as its main difference.