Mistral releases remote coding agents and new model for complex tasks
Mistral today introduced Mistral Medium 3.5, a 128-billion-parameter model designed to run coding and productivity work for extended periods. The model is now available in public preview and becomes the default across Mistral's Vibe coding platform and Le Chat assistant.
The release includes remote agents in Vibe that run coding tasks in the cloud without requiring a developer's laptop to stay connected. Multiple agents can execute in parallel, and they notify users when work completes.
What the new model does
Mistral Medium 3.5 combines instruction-following, reasoning, and code generation in a single set of weights. It has a 256,000-token context window and scores 77.6% on SWE-Bench Verified, a benchmark for software engineering tasks.
The model runs on as few as four GPUs for self-hosted deployments. Reasoning effort is configurable per request, allowing the same model to handle quick chat responses or complex multi-step agent work.
Mistral trained a new vision encoder from scratch to handle variable image sizes and aspect ratios.
How remote agents work in Vibe
Developers can start coding sessions from the Mistral Vibe command-line interface or directly in Le Chat. Sessions run on Mistral's servers and continue executing while the developer steps away.
A local CLI session can be "teleported" to the cloud, preserving session history, task state, and any approvals already made. Multiple sessions can run in parallel without blocking each other.
The agent works in an isolated sandbox. When finished, it can open a pull request on GitHub and send a notification, so developers review the final result rather than monitoring each step.
Vibe integrates with GitHub for code and pull requests, Linear and Jira for issues, Sentry for incident tracking, and Slack or Teams for notifications.
Work mode in Le Chat
Le Chat's new Work mode (in preview) uses Mistral Medium 3.5 to handle multi-step tasks that span multiple tools and systems. The agent can read and write across email, messages, calendars, documents, and connected applications.
Users can ask the agent to research topics across the web and internal sources, then produce a structured report. It can triage inboxes, draft replies, create issues in project management tools, and send summaries to team channels.
Every action is visible to the user. Le Chat asks for explicit approval before sending messages, writing documents, or modifying data, based on the user's permissions.
Pricing and availability
Mistral Medium 3.5 is available today through Mistral's API at $1.50 per million input tokens and $7.50 per million output tokens. Remote agents and Work mode are available on Pro, Team, and Enterprise plans.
The model is released as open weights under a modified MIT license on Hugging Face. It's also available through NVIDIA's GPU-accelerated endpoints and as a containerized inference service called NVIDIA NIM.
For developers looking to expand their skills with AI tools, resources on AI coding courses and generative AI and LLM fundamentals can provide foundational knowledge for working with these systems.
Your membership also unlocks:
