Mistral AI Updates: May 1, 2026

1. Vibe Goes Cloud: Remote Coding Agents Launch With Le Chat and CLI Triggers

Mistral. Mistral moved its Vibe coding agent off the laptop and into the cloud, letting developers run long-horizon coding sessions asynchronously while integrating with GitHub, Linear, Jira, and Slack. Multiple remote agents can run in parallel, and developers can “teleport” a local Vibe session to the cloud while preserving session history and approval state. Sessions are launchable from the Vibe CLI or directly from Le Chat, which closes the gap between the chat surface and the IDE for teams who prefer a single entry point. Sandboxed execution keeps human oversight in the loop for anything that touches a real repo. Source

2. Le Chat Gets Work Mode for Long-Horizon Multi-Tool Tasks

Mistral. Alongside Vibe, Mistral introduced Work mode in Le Chat, an agentic interface powered by Mistral Medium 3.5 aimed at research synthesis, inbox triage, and cross-tool workflows that stitch together email, calendar, and messaging. Every action is shown in the UI and sensitive operations such as sending a message or modifying data require explicit approval, which is the same human-in-the-loop pattern Mistral baked into Workflows the day before. The framing is deliberate: the chat box is no longer just a generation surface, it is the launch point for jobs that may run minutes to hours. Source

3. Medium 3.5 Pricing and Footprint Confirmed for Agentic Workloads

Mistral. The Vibe and Work mode launch doubled as a public confirmation of Mistral Medium 3.5’s commercial profile: 128B dense parameters with a 256k context window, 77.6% on SWE-Bench Verified, configurable reasoning effort per request, and self-hosting on as few as four GPUs. API pricing lands at $1.50 per million input tokens and $7.50 per million output tokens, with open weights on Hugging Face under a modified MIT license. The combination of a four-GPU minimum and open weights is the lever Mistral is pulling against frontier closed-model competitors for regulated and on-prem agent deployments. Source