Mistral Unveils Medium 3.5 Model Enabling Cloud-Based Remote Coding Agents and Advanced Workflows

Mistral Technologies has introduced a groundbreaking update to their AI platform with the release of the Medium 3.5 model—a 128-billion parameter dense model designed specifically for extended coding tasks and productivity enhancements. This advancement marks a shift as coding agents, traditionally run locally, are now operated remotely in the cloud, enabling asynchronous, parallel execution of programming assignments through Mistral Vibe and Le Chat interfaces.

The new Medium 3.5 model, now the default engine for Mistral Vibe and Le Chat, supports a novel "Work mode" that empowers users to tackle intricate, multi-step operations such as comprehensive research, analysis, and workflows integrating several tools simultaneously. Rather than replacing human judgment, this system facilitates automated task handling while ensuring user oversight remains integral for sensitive procedures.

Available on Pro, Team, and Enterprise subscription tiers, Medium 3.5 is accessible for $1.5 per million input tokens and $7.5 per million output tokens. Notably, Mistral offers the model's weights openly under a modified MIT license, enabling experimentation and integration by interested developers.

Contrasting earlier approaches where coding assistants resided on individual laptops, Mistral’s approach leverages cloud infrastructure, allowing multiple coding sessions to operate concomitantly without tying up the user. Developers can launch these remote agents either from the Vibe CLI or within the chat environment of Le Chat, with the agents autonomously advancing through assigned tasks and alerting users upon completion.

This cloud-hosted functionality extends to workflow continuity—local command line interface sessions can be seamlessly transitioned to the cloud to maintain active processing and preserve task progress and approvals. Integration with existing software development tools simplifies the transition, as Vibe connects to version control systems like GitHub, issue trackers such as Jira and Linear, incident platforms including Sentry, and communication channels like Slack or Microsoft Teams.

Each coding session is encapsulated within isolated environments to safeguard the overall system, supporting wide-ranging code modifications and software installations safely. Upon task completion, the agent can open a pull request automatically, minimizing the need for developers to micromanage every code change and freeing them to focus on critical decisions.

The Work mode introduced in Le Chat represents a significant enhancement by utilizing the Medium 3.5 model as an autonomous agent capable of orchestrating cross-application tasks. For example, it can consolidate information from emails, messages, calendars, and company documents to prepare meeting briefs, triage communications, generate task tickets within team management tools, and distribute concise summaries to relevant stakeholders.

Sessions in this mode are persistent and designed to progress through multiple interactions and trial phases until fulfillment of user objectives. Actions taken by the agent are transparent, offering visibility into decision-making processes and tool usage. Moreover, for certain operations involving data changes or communication dispatches, explicit user permissions and confirmations serve as safeguards against unintended actions.

By deploying Medium 3.5, Mistral enhances the ability of businesses to automate routine, well-specified development and administrative tasks in a scalable and cost-efficient manner. The platform balances sophisticated agent autonomy with human control, providing a practical solution for organizations aiming to optimize coding workflows without compromising quality or oversight.

This latest offering from Mistral aligns with their vision to simplify enterprise deployment of AI-powered coding assistants, strengthening capabilities while keeping infrastructure requirements manageable—Medium 3.5 can run self-hosted on as few as four GPUs.

Developers and teams interested in this technology can access it immediately, benefiting from the advanced performance and flexibility of the new model. Mistral continues hiring across multiple domains to further elevate agentic systems and AI-driven productivity tools.