Remote agents in Vibe. Powered by Mistral Medium 3.5.
In the rapidly evolving world of artificial intelligence, Mistral AI has introduced a groundbreaking development: remote coding agents in Vibe, powered by Mistral Medium 3.5. This innovative technology marks a significant shift in how coding tasks are managed, enabling users to offload complex tasks to the cloud while maintaining productivity.
Introduction to Mistral Medium 3.5
Mistral Medium 3.5 is the latest flagship model from Mistral AI, designed to merge instruction-following, reasoning, and coding capabilities into a single, dense model. With a size of 128 billion parameters and a context window of 256,000 tokens, it is built for long-duration tasks and can operate efficiently on as few as four GPUs.
This model is now available in public preview and has been released under a modified MIT license, allowing developers to leverage its capabilities for various applications.
Key Features of Mistral Medium 3.5
- High Performance: Mistral Medium 3.5 demonstrates strong real-world performance, scoring 77.6% on SWE-Bench Verified, outperforming other models like Devstral 2 and Qwen3.5.
- Agentic Capabilities: The model excels in handling complex tasks, scoring 91.4 on τ³-Telecom, making it suitable for asynchronous cloud agents.
- Configurable Reasoning: Users can adjust the reasoning effort based on task requirements, allowing for quick responses or in-depth analysis.
- Versatile Training: The vision encoder was trained from scratch to accommodate variable image sizes and aspect ratios, enhancing its adaptability.
Remote Agents in Vibe
With the introduction of remote agents in Vibe, coding sessions can now run asynchronously in the cloud. This feature allows multiple sessions to operate in parallel, freeing developers from being the bottleneck in the coding process. Users can initiate coding tasks either from the Mistral Vibe Command Line Interface (CLI) or directly within Le Chat.
While these remote agents execute tasks, users can monitor their progress through various indicators, such as file diffs, tool calls, and task states. Additionally, ongoing local CLI sessions can be “teleported” to the cloud, preserving session history and task states.
Integration with Development Tools
Mistral Vibe seamlessly integrates with popular development tools and platforms, enhancing its utility for coding tasks. It connects with:
- GitHub: For code management and pull requests.
- Linear and Jira: For issue tracking and project management.
- Sentry: For incident reporting and monitoring.
- Slack and Teams: For team communication and updates.
This integration ensures that coding sessions remain organized and efficient, allowing developers to focus on high-value tasks while the agents handle routine operations.
Work Mode in Le Chat
The new Work mode in Le Chat is a powerful feature that enhances the capabilities of remote agents. It allows users to manage complex, multi-step tasks more effectively. Powered by Mistral Medium 3.5, Work mode enables:
- Cross-Tool Workflows: Users can manage tasks across multiple platforms, such as email, calendars, and messaging apps, all within a single workflow.
- Research and Synthesis: The agent can gather information from various sources, synthesize it, and produce structured reports or briefs.
- Inbox Management: Triage emails, draft replies, and create issues in project management tools based on discussions.
In Work mode, the agent operates with connectors enabled by default, allowing it to access necessary documents and systems without manual selection. This feature enhances the agent’s ability to perform tasks efficiently and accurately.
Visibility and Control
One of the standout features of Mistral Vibe and Work mode is the transparency it offers users. Every action taken by the agent is visible, allowing users to track tool calls and the rationale behind decisions. Moreover, explicit approval is required for sensitive tasks, ensuring that users maintain control over critical actions.
Getting Started with Mistral Medium 3.5
Mistral Medium 3.5 is now available in Mistral Vibe and Le Chat, accessible through Pro, Team, and Enterprise plans. The pricing structure for API usage is set at $1.5 per million input tokens and $7.5 per million output tokens. Additionally, open weights are available on Hugging Face under a modified MIT license, allowing for further experimentation and development.
For those interested in prototyping, Mistral Medium 3.5 can be hosted on NVIDIA GPU-accelerated endpoints and as a scalable containerized inference microservice via NVIDIA NIM.
Conclusion
The introduction of remote agents in Vibe, powered by Mistral Medium 3.5, represents a significant advancement in the field of AI and coding productivity. By enabling asynchronous coding tasks and integrating seamlessly with existing tools, Mistral AI is paving the way for a more efficient and collaborative development environment. As the demand for agentic systems continues to grow, Mistral AI is committed to pushing the boundaries of what is possible in this exciting field.
Note: This article is based on information available as of October 2023 and may be subject to change as new developments occur.
