QWE AI Academy If you logged into Le Chat last week and got bounced to a screen called Vibe, that’s the AI Now Summit hitting your account. Mistral’s first-ever flagship event dropped on May 28 in Paris, and the headline wasn’t a new model – it was a repackaging. Three products collapsed into one. The question this article answers: now that the dust has settled, what do you actually click first?
Most write-ups from the summit focus on the Airbus and BMW partnerships. Useful if you run a factory. The official announcement tells you what Vibe is. What it doesn’t tell you: which mode to open, where the billing breaks, and what’s quietly broken at launch. That’s what this covers.
The takeaway in one paragraph
Le Chat is now Vibe – Mistral’s unified agent for productivity and coding tasks, available across web, mobile, your code editor, and your terminal. Per Mistral’s official docs, it runs in three modes: Work (multi-stage tasks across your apps), Code (development in terminal, VS Code, or cloud sessions), and Chat (quick conversations). All existing Le Chat conversations, settings, and subscription plans migrated automatically. Nothing to reinstall – just a new layout and two new modes to figure out.
Why this rename matters more than it looks
Rebrands are usually marketing noise. This one isn’t, because it bundles billing too. Before the summit, the coding agent and the chatbot lived under different names and different URLs. Now one licence covers both work and code.
The underlying model matters here. 77.6% on SWE-Bench Verified – that’s Mistral Medium 3.5, which now powers Vibe by default. Add a 256K context window and configurable reasoning effort, and you get an agent stack that the previous Le Chat couldn’t run. That said, a capable model only matters if you’re in the right mode for your task, which brings us to the actual decision.
Work Mode or Code Mode: pick one before you open the app
This is the choice that traps new users. They’re not interchangeable, and the docs bury the difference in a comparison page.
| If you want to… | Use | Why |
|---|---|---|
| Triage email, draft reports, pull data from SharePoint or Slack | Work Mode | Connects to business apps and runs multi-step plans across them |
| Edit a repo, write tests, open a pull request | Code Mode | Has filesystem access, a shell, and PR tooling |
| Ask a quick question or use Deep Research | Chat | Legacy Le Chat surface – fastest for one-shot prompts |
Work Mode is where most readers should start. Available on Free in the browser, no install required. Code Mode is where Vibe earns its keep for developers – but it carries a billing gotcha that the setup docs skip entirely.
Worth sitting with for a moment: Mistral is betting that one tool can cover both a product manager’s Monday-morning inbox triage and a developer’s 3am debugging session. That’s an unusual scope for a single interface. Whether the unified login is a convenience or a source of confusion will probably depend on which half of that pairing you use more.
Getting Work Mode useful in 5 minutes
Open chat.mistral.ai. Switch to Work in the top-left corner – it’s available even on Free.
- Connect one tool first. Don’t connect everything at once. Pick the app where your actual bottleneck lives. According to the official announcement, Vibe reaches across Google Workspace, Outlook, SharePoint, Slack, GitHub, and custom MCP connectors. Start with email or your docs store.
- Write the outcome, not the steps. “Catch me up on what I missed in the customer-support channel this week, group complaints by theme, draft a Monday update.” The agent plans, then acts.
- Read the plan before it runs. The agent shows its intended steps before executing. If a step looks wrong, edit it before clicking go – this is the safest habit to build early.
- Repeat the workflow a few times before judging it. The first run on a new connector is usually the roughest; the agent gets more useful once it has context about your actual data.
Code Mode: the CLI path
The fastest install:
# install
curl -LsSf https://mistral.ai/vibe/install.sh | bash
# first run - wizard asks for sign-in or API key
vibe
# one-shot prompt in any repo
cd ~/projects/myrepo
vibe -p "Find TODO comments in this project"
Python 3.12+ required for manual installation. The GitHub repo (Apache 2.0) has full config docs, including the ~/.vibe/config.toml setup. One note on platform support: the repo states Vibe works on Windows but officially targets UNIX environments only. If you’re on Windows and hit weird path bugs, that’s why – WSL is the cleaner option.
Code sessions run in isolated cloud sandboxes, can run in parallel, and persist while your laptop is off. The /teleport command moves an active local CLI session up to the cloud – start a refactor at 5pm, teleport it, close the lid, check the PR in the morning. This is the one capability where Vibe does something Claude Code and Codex don’t replicate cleanly.
The edge cases other write-ups skipped
Here’s the part you won’t find in the summit recaps.
The API-vs-subscription billing trap. Your Vibe subscription does not include API credits. The €5.99 student plan – and the Pro plan – only cover use inside the Vibe app and interface. Use the coding agent through an API call (inside your IDE, a CI pipeline, or an automated script) and you’re billed separately at token rates: $1.50 per million input tokens and $7.50 per million output tokens for Medium 3.5 via API. Run agents in CI without checking and you’ll watch the bill add up fast.
The undisclosed quota. Mistral has not published concrete usage limits for any Vibe tier – the official position is only that paid plans offer “multiples” of the free plan’s allowances. Competitors like Anthropic and OpenAI publish specific rate limits. With Vibe, you find your ceiling by hitting it.
Team is admin, not capacity. Upgrading from Pro (€14.99/mo) to Team (€24.99/user/mo) for more usage headroom won’t help. The two tiers share identical usage limits – the €10 gap buys team management, domain verification, and admin tools, not higher quotas. Pay for Team only if you need the admin layer.
The internal-comms skill can’t see your data yet. As of May 29, 2026, internal-comms cannot access connected connectors or independently retrieve information from them. Ask it to draft a newsletter using last quarter’s numbers from SharePoint and it won’t pull those numbers – it only fills templates. For anything requiring live data from connected apps, use Work Mode’s general agent instead.
Slack-triggered Code sessions are coming, not here yet. The official announcement notes Slack-triggered Code sessions are scheduled for June 2026. Don’t build a workflow that depends on it until it ships.
FAQ
Do I lose anything from my old Le Chat account?
No. Conversations, settings, and your active plan all carried over automatically at launch.
Is Vibe worth €14.99/month over ChatGPT Plus?
For pure prose writing, ChatGPT and Claude have deeper consumer-feature ecosystems right now. But if your day involves stitching together emails, documents, and a codebase – the combined Work + Code modes under one login is a genuinely different value proposition than either competitor currently offers in a single subscription. The real risk: Mistral is opaque about exact usage limits, so heavy users should budget for one month of testing before committing to the annual plan.
Can I run Vibe Code fully offline?
The web app and VS Code extension need Mistral’s hosted services. The CLI can be pointed at a compatible API key you provide, but Mistral’s official documentation does not explicitly describe offline/local-model operation – check the GitHub repo for the latest on self-hosted configurations, as this may have changed since launch.
Next action: open chat.mistral.ai, switch to Work in the top left, and connect one tool – the one tied to your most repetitive task this week. Ship one automation tonight. Decide on Code Mode after that.