Mistral releases remote coding agents and Mistral Medium 3.5 model
Mistral has moved coding agents from laptops to the cloud, letting developers start tasks in the command line or chat interface and have them run in parallel while work continues elsewhere. The company released Mistral Medium 3.5, a 128-billion-parameter model that powers the new remote agents and a Work mode feature in Le Chat.
Coding agents have traditionally required local compute and developer attention at each step. Remote agents in Vibe, Mistral's coding assistant, now run in isolated cloud sandboxes and notify developers when finished. A developer can spawn multiple agents in parallel from the Mistral Vibe CLI or directly from Le Chat, then step away.
Mistral Medium 3.5 performance
Mistral Medium 3.5 is a dense model with a 256,000-token context window. It scores 77.6% on SWE-Bench Verified, a benchmark for software engineering tasks, ahead of Mistral's previous Devstral 2 model. The model handles instruction-following, reasoning, and code generation in a single set of weights.
The model can run self-hosted on as few as four GPUs, making it practical for organizations running their own infrastructure. Reasoning effort is configurable per request, so the same model handles quick chat replies and extended agent runs.
Mistral released the weights under a modified MIT license on Hugging Face. Through the API, the model costs $1.5 per million input tokens and $7.5 per million output tokens.
Remote agents in Vibe
Developers can start coding tasks from the CLI or Le Chat. Sessions run on the same remote infrastructure regardless of where they're initiated. A local CLI session can be teleported to the cloud mid-task, carrying over session history and approval state.
Each agent runs in an isolated sandbox with broad edit and install permissions. When work finishes, the agent can open a pull request on GitHub and send a notification, so developers review the result rather than every intermediate step.
The system integrates with GitHub, Linear, Jira, Sentry, Slack, and Teams. Developers stay in the loop where needed, with visibility into file diffs, tool calls, and progress states.
Remote agents suit high-volume, well-defined work: module refactors, test generation, dependency upgrades, CI investigations, and bug fixes.
Work mode in Le Chat
Work mode is a new agentic interface in Le Chat powered by Mistral Medium 3.5. The agent becomes the execution backend, able to read, write, and use multiple tools in parallel across multi-step projects.
Developers can use Work mode to:
- Build cross-tool workflows across email, messages, and calendar
- Research topics across the web, internal documents, and connected systems, then produce structured reports
- Triage inboxes and draft replies, create issues in Jira, and send summaries to Slack
Connectors are enabled by default in Work mode, giving agents access to documents, mailboxes, calendars, and other systems. Le Chat displays each tool call and reasoning step. The system asks for explicit approval before sensitive actions like sending messages or modifying data.
Availability and pricing
Mistral Medium 3.5 is available in Mistral Vibe and Le Chat on Pro, Team, and Enterprise plans. The model is also available on NVIDIA GPU-accelerated endpoints and as NVIDIA NIM, a containerized inference service.
Developers working with coding tasks or AI Agents & Automation may find the remote agent capability relevant to their workflows.
Your membership also unlocks:
