Mistral Moves Coding Agents to the Cloud with Medium 3.5
Mistral AI has shifted coding agents off the developer's laptop and into the cloud. Mistral Medium 3.5, a 128B dense model released under a modified MIT license, now powers remote agents in Mistral Vibe and Le Chat. Agents run in parallel, asynchronously, and notify the user when complete. Pricing is $1.50 per million input tokens and $7.50 per million output tokens.
The move matters because it separates execution from attention. A developer starts a task and walks away. The agent runs on Mistral's infrastructure, not local hardware. Mistral also introduced Work mode in Le Chat for multi-step tasks: research, analysis, and cross-tool workflows. Human oversight is maintained for sensitive actions. Mistral AI reported this directly on its own site.
Watch whether enterprise buyers treat remote agents as a productivity tool or a compliance question. Human-in-the-loop design suggests Mistral is anticipating both. The self-hosted option, four GPUs minimum, keeps sovereign deployment on the table for French and European operators who will not route sensitive workloads through a shared cloud.