Mistral Medium 3.5 and Vibe Remote Agents Release

Mistral AI has introduced Mistral Medium 3.5, a 128B open-weight model designed for long-horizon coding and productivity, alongside the launch of remote agents in Mistral Vibe and a new "Work mode" in Le Chat. These updates shift coding agents from local environments to the cloud, enabling parallel, asynchronous execution of complex tasks while maintaining human oversight.

Mistral Medium 3.5: Technical Specifications and Performance

Mistral Medium 3.5 is a dense 128B model that integrates instruction-following, reasoning, and coding capabilities into a single set of weights. It is released under a modified MIT license with open weights available on Hugging Face.

Key Technical Capabilities

  • Context Window: The model features a 256k context window.
  • Hardware Efficiency: The model can be self-hosted on as few as four GPUs.
  • Configurable Reasoning: Reasoning effort is configurable per request, allowing the model to toggle between quick chat responses and complex agentic runs.
  • Vision Integration: The vision encoder was trained from scratch to support variable image sizes and aspect ratios.

Benchmarks

  • SWE-Bench Verified: Mistral Medium 3.5 scores 77.6%, outperforming Devstral 2 and Qwen3.5 397B A17B.
  • τ³-Telecom: The model scores 91.4, demonstrating strong agentic capabilities.

Mistral Vibe Remote Agents

Remote agents move coding sessions from the local laptop to the cloud, removing the developer as the bottleneck for every step of the agent's execution. These agents are available via the Mistral Vibe CLI and Le Chat.

Asynchronous Coding Workflows

  • Parallel Execution: Developers can spawn multiple cloud agents to run tasks in parallel.
  • Session Teleportation: Local CLI sessions can be "teleported" to the cloud, carrying over session history, task state, and approvals.
  • Integration: Vibe integrates with GitHub (code and pull requests), Linear and Jira (issues), Sentry (incidents), and communication tools like Slack and Teams.
  • Sandbox Environment: Each session runs in an isolated sandbox allowing for broad edits and installations.
  • Outcome-Based Review: Agents can open GitHub pull requests and notify the user upon completion, allowing developers to review the final result rather than individual keystrokes.

Le Chat Work Mode (Preview)

Work mode is an agentic execution backend for Le Chat, powered by Mistral Medium 3.5, designed for complex, multi-step projects that require trial-and-error and long-term persistence.

Agentic Capabilities

  • Cross-Tool Workflows: The agent can simultaneously access email, messages, and calendars to prepare for meetings or triage inboxes.
  • Research and Synthesis: The agent can synthesize information from the web, internal documents, and connected tools to produce structured reports.
  • Administrative Automation: The agent can create Jira issues from discussions or send summaries to Slack.
  • Operational Transparency: Every tool call and thinking rationale is visible to the user. Sensitive tasks (e.g., sending messages or modifying data) require explicit user approval based on permissions.

Availability and Pricing

Mistral Medium 3.5 is the new default model in Le Chat and replaces Devstral 2 in the Vibe CLI. It is available for Pro, Team, and Enterprise plans.

API Pricing

  • Input Tokens: $1.5 per million tokens.
  • Output Tokens: $7.5 per million tokens.

Deployment Options

  • Hugging Face: Open weights available under a modified MIT license.
  • NVIDIA: Available via GPU-accelerated endpoints on build.nvidia.com and as a scalable containerized inference microservice via NVIDIA NIM.

Sources

Related

  • Dispatch
    Mistral AI Vibe Release Notes

    Mistral AI has launched Vibe, a unified AI agent that integrates work and coding workflows across web, mobile, IDE, and CLI interfaces.

  • Dispatch
    Mistral AI Le Chat Release

    Mistral AI has released Le Chat, a multilingual conversational assistant that serves as a demonstration of Mistral models' capabilities for both individual users and business environments.

  • Dispatch
    Mistral Vibe 2.0 Release Notes

    Mistral AI has released Mistral Vibe 2.0, a terminal-native coding agent powered by the Devstral 2 model family that introduces custom subagents, slash-command skills, and unified agent modes.

  • Dispatch
    Mistral AI Devstral 2 and Mistral Vibe CLI Release

    Mistral AI has released Devstral 2 (123B) and Devstral Small 2 (24B), open-weight coding models that achieve state-of-the-art performance on SWE-bench Verified with high cost-efficiency, alongside the open-source Mistral Vibe CLI for autonomous code automation.

  • Dispatch
    Mistral Small 3.1 Release Notes

    Mistral AI has released Mistral Small 3.1, an Apache 2.0 licensed model featuring multimodal understanding, a 128k token context window, and performance that surpasses GPT-4o Mini and Gemma 3.