Mistral has announced the public preview of its latest flagship model, Mistral Medium 3.5, designed to enhance productivity by enabling coding tasks to run in the cloud autonomously. This marks a significant shift from traditional laptop-based coding agents to a more flexible cloud-based system, allowing users to initiate tasks through the Mistral Vibe CLI or via Le Chat without interrupting their workflow. Mistral Medium 3.5 integrates instruction-following, reasoning, and coding capabilities into a single 128 billion parameter model, aimed at performing efficiently over extended periods of coding and productivity work.
The introduction of the new Work mode in Le Chat enhances collaboration for complex, multi-step tasks such as research and analysis, allowing users to work on projects without having to micromanage every detail. Mistral Medium 3.5’s capabilities are backed by its impressive performance metrics, scoring 77.6% on SWE-Bench Verified and demonstrating robust agentic capabilities with a score of 91.4 on τ³-Telecom.
The model’s architecture provides a 256,000 token context window, adaptable reasoning effort per task, and the ability to self-host on as few as four GPUs. This flexibility makes it an attractive option for both small teams and larger enterprises, as it can handle a range of tasks from quick responses to complex agent-based operations. Mistral Medium 3.5 replaces the previous Devstral 2 model in the Vibe CLI and becomes the default model in Le Chat.
Mistral’s cloud computing approach allows coding sessions to continue even when users step away, alleviating the bottleneck often associated with manual coding. Users can initiate coding tasks in Le Chat and monitor ongoing projects in real-time, with capabilities to inspect file changes, tool usage, and other progress indicators. Notably, local CLI sessions can be seamlessly transferred to the cloud, ensuring continuity and preserving session history.
The integration of Mistral Vibe with popular project management tools like GitHub, Linear, and Jira enhances its utility, especially for developers managing high-volume tasks. Each coding session operates within a secure sandbox environment, allowing for extensive modifications and installations without affecting system integrity. Once tasks are completed, agents can open pull requests automatically, streamlining the review process for developers.
The new Work mode in Le Chat is designed to facilitate complex workflows, enabling the agent to traverse various tools simultaneously for comprehensive task execution. This mode allows for extended sessions that surpass conventional chat limits, permitting the agent to engage in trial and error until a task is completed. Current capabilities include cross-tool workflows that integrate email, messaging, and calendars, as well as research tasks that can synthesize information and produce structured reports.
Users can maintain visibility over the agent’s actions, with a detailed log of tool calls and decision-making processes provided throughout each task. Furthermore, explicit user approval is required for sensitive actions, ensuring that tasks are carried out in accordance with user permissions.
Mistral Medium 3.5 is now available for users on the Pro, Team, and Enterprise plans, with pricing set at $1.5 per million input tokens and $7.5 per million output tokens through API. The model’s open weights are accessible on Hugging Face under a modified MIT license, and it can be prototyped on NVIDIA GPU-accelerated environments.
This launch marks a significant advancement in agentic systems, and Mistral is actively seeking talent in research, engineering, and product development to further innovate in this field. As the landscape of coding and productivity tools continues to evolve, Mistral’s cloud-based solutions aim to streamline workflows and enhance efficiency for developers and teams alike.
See also
Germany”s National Team Prepares for World Cup Qualifiers with Disco Atmosphere
95% of AI Projects Fail in Companies According to MIT
AI in Food & Beverages Market to Surge from $11.08B to $263.80B by 2032
Satya Nadella Supports OpenAI’s $100B Revenue Goal, Highlights AI Funding Needs
Wall Street Recovers from Early Loss as Nvidia Surges 1.8% Amid Market Volatility
