Remote agents in Vibe. Powered by Mistral Medium 3.5.

On May 22, 2026, Mistral announced the migration of coding agents from local terminals to the cloud with the launch of Vibe Remote Agents. These agents run independently in the cloud, support parallel execution, and allow users to start a task and leave, receiving notifications upon completion. Users can launch cloud agents from Mistral Vibe CLI or Le Chat and view file diffs, tool calls, progress status, and issues in real time. Local CLI sessions can be 'teleported' to the cloud, migrating session history, task status, and approval information.

Each coding session runs in an isolated sandbox, including extensive editing and installation operations. Upon task completion, the agent can open a pull request on GitHub and notify the user, who only needs to review the results rather than each step. Vibe integrates with GitHub, Linear, Jira, Sentry, and apps like Slack or Teams, making it suitable for high-frequency, well-defined tasks such as module refactoring, test generation, dependency upgrades, CI investigation, and bug fixes.

Mistral uses Workflows in Studio to bring Vibe to Le Chat, initially for internal coding environments, then opened to enterprise customers, and now to all users. Developers can launch coding tasks from the web without binding to a local terminal and run multiple tasks in parallel.

Mistral Medium 3.5 is Mistral's first flagship fusion model, currently in public preview. It is a 128B dense model with a 256k context window, combining instruction following, reasoning, and coding capabilities into a single set of weights. Reasoning effort can be configured per request, allowing the same model to handle quick chat replies as well as complex agent runs. The vision encoder is trained from scratch, supporting variable image sizes and aspect ratios.

The model scores 77.6% on SWE-Bench Verified, leading over models like Devstral 2 and Qwen3.5 397B A17B; and scores 91.4 on τ³-Telecom, demonstrating strong agent capabilities. Designed for long-horizon tasks, the model reliably calls multiple tools and generates structured outputs, making asynchronous cloud agents practical in Vibe.

Mistral Medium 3.5 becomes the default model for Le Chat and replaces Devstral 2 in Vibe CLI. Pricing is $1.5 per million input tokens and $7.5 per million output tokens, available immediately in Pro, Team, and Enterprise plans. The model is released with open-source weights under a modified MIT license and can be self-hosted on as few as 4 GPUs.

Le Chat adds Work mode (preview), a powerful new agent mode driven by a new agent framework and Mistral Medium 3.5. The agent becomes the execution backend of the assistant itself, enabling Le Chat to read and write, use multiple tools simultaneously, and handle multi-step projects to completion.

Work mode currently supports cross-tool workflows (e.g., handling email, messages, and calendar in one run), research synthesis (deep dives across web, internal documents, and connected tools to generate structured briefs or reports), and classification and action (e.g., categorizing support tickets and executing follow-up steps).