Mistral Medium 3.5 Model Powers Vibe Agents
Mistral AI announced Mistral Medium 3.5, a new model that supports remote coding agents in Vibe. The company also introduced a Work mode in Le Chat for handling complex tasks. These features move coding agents from local laptops to the cloud, where they operate independently, run in parallel, and alert users upon completion. Users can initiate them via the Mistral Vibe CLI or directly in Le Chat without leaving the chat.
Mistral Medium 3.5 serves as the default model in Mistral Vibe and Le Chat. This public preview version focuses on extended runs for coding and productivity. The model combines instruction-following, reasoning, and coding in one package.
Details on Mistral Medium 3.5
Mistral Medium 3.5 marks the company's first flagship merged model in public preview. It features a dense 128B parameter setup with a 256k context window. The model manages instruction-following, reasoning, and coding using a single weight set. It delivers solid results in practical applications and supports self-hosting on just four GPUs. Users can adjust reasoning effort per request, allowing quick chat responses or in-depth agent operations. The team trained a vision encoder from scratch to process varying image sizes and aspect ratios.
On benchmarks, Mistral Medium 3.5 achieves 77.6% on SWE-Bench Verified, surpassing Devstral 2 and models such as Qwen3.5 397B A17B. It shows strong agent skills with a 91.4 score on τ³-Telecom. Designed for long tasks, it calls tools multiple times reliably and generates structured outputs for code integration. This capability made cloud agents in Vibe feasible. The model now acts as the default in Le Chat and replaces Devstral 2 in the Vibe CLI coding agent.
Mistral AI, founded in 2023 by former DeepMind and Google researchers, specializes in open-source large language models. Previous releases like Mistral 7B gained attention for efficiency on consumer hardware. Medium 3.5 builds on that tradition with open weights under a modified MIT license.
Vibe Remote Agents
Coding sessions in Vibe now handle extended tasks remotely while users step away. Multiple sessions operate in parallel, removing user bottlenecks at each step. Start them from the Mistral Vibe CLI or Le Chat. Monitor progress with file diffs, tool calls, status updates, and queries during execution. Local CLI sessions transfer to the cloud seamlessly, preserving history, state, and approvals.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Vibe integrates with tools developers use, keeping humans involved as needed. It connects to GitHub for code and pull requests, Linear and Jira for issues, Sentry for incidents, and platforms like Slack or Teams for notifications. Each session uses an isolated sandbox for edits and installs. Upon finishing, the agent creates a GitHub pull request and notifies the user for review, skipping per-keystroke checks.
It targets routine tasks that consume developer time but not judgment, such as module refactors, test creation, dependency updates, CI checks, and bug fixes. Workflows from Mistral Studio integrate Vibe into Le Chat. Originally for internal and enterprise use, this now reaches all users. Developers launch tasks from the web, running several in parallel without a local terminal. Chat-initiated sessions use the same remote runtime as CLI and web, returning as branches or draft PRs.
Work Mode in Le Chat
Work mode in Le Chat, available in preview, uses a new agent powered by Mistral Medium 3.5 for complex tasks. The agent acts as the backend, enabling reads, writes, parallel tool use, and multi-step projects to completion.
Current capabilities include cross-tool workflows, like summarizing email, messages, and calendars; or prepping meetings with attendee details, news, and points from sources. It supports research across web, docs, and tools, producing editable briefs or reports. Other functions cover inbox triage with reply drafts, Jira issue creation from discussions, and Slack summaries.
Sessions last beyond standard chats, allowing persistence through turns, trial-and-error, and completion. Connectors activate by default for access to documents, mailboxes, calendars, and systems. All agent actions appear, including tool calls and rationales. Le Chat seeks approval for sensitive actions like messaging, document edits, or data changes, based on permissions.
Availability
Mistral Medium 3.5 appears in Mistral Vibe and Le Chat, driving remote agents and Work mode on Pro, Team, and Enterprise plans. API pricing stands at $1.5 per million input tokens and $7.5 per million output tokens. Open weights reside on Hugging Face under the modified MIT license. It also runs on NVIDIA GPU endpoints at build.nvidia.com and as NVIDIA NIM microservices.
Related on Neura Market