Mistral AI is renaming Le Chat to Vibe, repositioning the former chatbot as a work tool that handles emails, writes reports, and ships code all the way to the pull request stage. Existing conversations and settings carry over, but the pricing is new.
Work mode is the main event. Vibe hooks into Google Workspace, Outlook, SharePoint, Slack, or GitHub and runs tasks from there, scanning inboxes, pulling numbers from a spreadsheet, building a report, and pushing it to Notion or SharePoint.
Before the agent kicks off, it lays out its plan and waits for a thumbs-up. Every step can be expanded and checked. Tasks can repeat daily, weekly, or monthly. So-called skills save recurring workflows as templates, similar to what Anthropic rolled out for Claude last year.
Code Mode drops programming agents into isolated sandboxes. They build features, fix bugs, write tests, and open a pull request when they're done. Sessions run in parallel, survive a closed laptop, and should be launchable from Slack starting in June. To go with it, Mistral is shipping a new VS Code extension and a CLI update with a /teleport command that moves running sessions and their history between the terminal and the cloud.
Vibe comes in four tiers: Free costs nothing, Pro runs 14.99 euros a month, Team is 24.99 euros per user per month, and Enterprise is priced on request. Mistral tacks on tax for all plans. Students get half off Pro, and Team drops to 19.99 euros per user with annual billing.
What each tier actually gets you is hard to pin down. Mistral shares numbers, but only as multiples of the free plan: up to six times as many messages, up to 40 times as many image generations. The company never says what the free plan's actual limits are. Six times an unknown number is still a mystery. Hard figures exist for only a handful of items, like 15 gigabytes of storage instead of "limited."
Pro and Team have identical usage limits. The ten-euro-per-user premium doesn't buy higher quotas. It adds more storage, domain verification, data export, and admin tools for shared workspaces. Whether that's worth it depends less on how much you use the tool and more on whether you need team management features.
The line between terms like "complex tasks" and "in-depth reasoning" is fuzzy, too. There's a technical reason, though: Mistral Medium 3.5 can toggle between a fast mode and a more thorough one through reasoning_effort. There's a free experiment plan in Mistral Studio.
The name "Vibe" already existed, by the way. Until the rebrand, it was Mistral's coding tool, available as a paid feature on Pro and Team since Vibe 2.0. Now the name has jumped from the coding agent to the whole assistant.
Vibe lives at chat.mistral.ai, as an app in the App Store and on Google Play. The code interface is at code.mistral.ai.
Over the past year, Le Chat gradually picked up Deep Research, a voice mode, image editing, MCP integrations, and a memory function. Work Mode and cloud-based coding agents followed in early May. Vibe bundles all of it under one brand and puts Mistral in more direct competition with the agentic versions of ChatGPT, Gemini, and Claude.
The coding setup with parallel sandbox agents isn't unique. OpenAI has Codex, Anthropic has Claude code in the browser, and Cursor runs entire agent fleets in version three. Adding to the confusion, the Vibe name used to mean Mistral's coding assistant Vibe 2.0. Now it covers the whole product, work mode and all.
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。