Mistral AI has moved coding agents from local laptops to the cloud with Mistral Medium 3.5, enabling parallel execution and notifications upon completion. This new flagship model merges instruction-following, reasoning, and coding into a single 128B dense model that can self-host on as few as four GPUs. The model becomes the default in both Mistral Vibe and Le Chat, replacing Devstral 2 in the coding agent.

Remote coding agents can be initiated from the CLI or directly within Le Chat, allowing developers to offload tasks without leaving conversations. These sessions run in isolated cloud environments with full visibility into progress, tool calls, and file diffs. The system integrates with GitHub, Linear, Jira, and other development tools, automatically creating pull requests when work completes.

The new Work mode in Le Chat handles complex multi-step tasks powered by Mistral Medium 3.5's 256k context window and configurable reasoning effort. Priced at $1.5 per million input tokens and $7.5 per million output tokens, the open-weight model excels at well-defined development work that consumes time but not judgment. The system transforms how developers approach refactoring, test generation, and bug fixes across their workflows.