Mistral AI Now Summit: Vibe Tutorial (2026 Notes)

The Mistral AI Now Summit wrapped at the Carrousel du Louvre on 28 May 2026, and the headline isn’t a new model – it’s a rename. Le Chat is now Vibe. Conversations, settings, and plans carry over automatically. No migration step.

The question in every community thread since: how do you actually use the thing? Two approaches exist, and one wastes your time.

Two ways to try Vibe (one is wrong for you)

Option A: straight to the API. Spin up Medium 3.5, pay per token, prototype an agent. Makes sense if you’re already building.

Option B: log in to the web app, toggle Work Mode. Free tier. No setup. You see what the rebrand actually changed in under five minutes.

Option B first – always. The Vibe subscription and the Mistral API are two completely separate billing lines. (More on that in the pitfalls section.) Start in the browser, decide if it’s worth it, then go to the API.

What changed at the summit

Three things you can actually touch today:

• Le Chat → Vibe. Same login, same history, new name, new top-left mode switcher.
• Work Mode. An agent that plans multi-step tasks across connected apps and waits for your approval before running anything.
• Code Mode + VS Code extension. Remote coding sessions that keep running when your laptop sleeps.

Chat, Work, Code – that’s the new structure, per Mistral’s product page. Chat is the old Le Chat. Work and Code are where the product actually changed.

Hands-on: Work Mode in 10 minutes

Go to chat.mistral.ai and log in. If you had a Le Chat account, it’s already a Vibe account.

1. Switch to Work Mode. Top-left corner, toggle from Chat to Work. Available on the free tier.
2. Connect one app first. Not everything – just Gmail or Google Calendar. Single data source makes the plan output readable.
3. Give it a long task. Something like: “Read my last 20 unread emails, group by sender domain, draft one-line replies to anything from a client.”
4. Read the plan before approving. Work Mode maps out every tool call and gets your sign-off before it runs. According to Mistral’s press materials, it “works across your connectors to carry the task through” only after that approval. That’s the actual product change – not the rename.
5. Save the workflow as a Skill. Mistral’s name for templated automations you can run on a schedule. Toggle individual Skills on or off from the sidebar.

Before connecting Slack or SharePoint: run one task against a throwaway Gmail account. See how the plan UI surfaces tool calls. Then decide if you trust it with your real inbox.

Here’s an honest question worth sitting with: if Work Mode can automate the email-triage task you do every morning, does that replace a Zapier workflow you’re already paying for – or does it just add another tab? Run one real task before you cancel anything.

The VS Code extension

Lowest-friction path into the coding side. Open the VS Code marketplace, search “Mistral Vibe”, publisher: mistralai. Sign in with your Vibe account. Done.

"Refactor this file to use async/await instead of .then() chains.
Ask before touching any other file."

Permission prompt before every write – that’s the default, and it’s the right call. Files get read and edited via prompt, but the extension requests access before each change. It also pulls context from GitHub, GitLab, Jira, or Linear if you connect them, per heise’s hands-on review. Claude Code users will recognize the pattern.

The CLI adds a /teleport command: move a coding task between terminal, editor, and cloud without restarting. Session history, task state, and pending approvals carry across surfaces. Slack integration for Code Mode is scheduled for June 2026 (as of the summit announcement – confirm current status on the product page).

Three traps. Nobody mentioned these on stage.

The Pro vs Team pricing is a trick question. Pricing tiers as of May 2026: Free, Pro at €14.99/month, Team at €24.99/user/month (€19.99 annual), Enterprise at custom pricing. The trap – Pro and Team have identical usage limits. That €10/user premium buys more storage, domain verification, data export, and admin tools for shared workspaces. Not more messages, not higher quotas. Solo users upgrading to Team thinking they’ll get more throughput: you won’t.

The quotas aren’t published. Mistral hasn’t disclosed concrete usage limits for any tier – only that paid plans offer “multiples” of the free plan. Anthropic publishes theirs; OpenAI publishes theirs. Mistral doesn’t yet (as of June 2026). You’re guessing at headroom until you hit a wall.

The CLI rate-limit error tells you nothing. Community reports and an open GitHub issue document the same problem: “Rate limit exceeded” with no indication of which limit was hit or by how much. Some developers report the CLI freezing on “generating” with no error message and no timeout. Workaround that users report works: wait 60 seconds, then prompt “continue from where you stopped.” It usually resumes mid-task. This is an open issue as of late May 2026 – worth checking the GitHub tracker for current status.

Subscription ≠ API credits. Using the coding agent inside a CI pipeline or script bills tokens separately at API rates – it’s not covered by your Vibe seat. If you’re running agents in automated pipelines, the API billing is entirely separate. Either go API-only, or stay in the web UI where the subscription covers you.

Medium 3.5 – the model underneath

Dense 128B model. 256k context window. According to Mistral’s announcement, it handles instruction-following, reasoning, and coding in a single set of weights – no separate reasoning model to route between.

77.6% on SWE-Bench Verified is a real number for a model at this price point – $1.50/M input is well below what you’d pay for comparable frontier models. Reasoning effort is configurable per request too, so the same model can handle a quick chat reply or a long agentic run without you spinning up a separate endpoint.

What’s less clear: how Medium 3.5 compares on complex multi-file refactors across large codebases. Mistral hasn’t published that comparison directly, and the benchmark scores don’t capture it well. If that’s your primary use case, test it yourself before committing – the weights are open, self-hosting is possible, and you don’t need a Vibe subscription to evaluate the model.

When NOT to use Vibe

• CI pipeline agents. The Vibe seat doesn’t cover API usage – you’d pay twice. Use the API directly.
• Predictable capacity planning. Until Mistral publishes concrete quotas, you can’t reliably model headroom. Anthropic and OpenAI both publish theirs.
• Heavy multi-file refactoring at scale. Medium 3.5 is competitive but Mistral hasn’t published direct comparisons for this workload. Evaluate against your actual codebase.

For everyone else – European teams with data residency concerns, students who qualify for the €5.99/month plan, developers who prefer open weights – the summit’s output is a cleaner single-license product than existed a month ago.

FAQ

Did anything actually new launch, or was it just a rename?

Mostly consolidation. Work Mode and the VS Code extension are real additions. Medium 3.5 shipped earlier in May – the summit packaged it into the Vibe product story.

I’m a student. Is the €5.99/month plan worth it over the free tier?

For regular coding work, probably yes – free tier limits coding sessions and excludes remote agents. The student plan is €5.99/month (60% off the €14.99 Pro price), valid up to 12 months, requires student verification, and requires that you’ve never used a Vibe or Le Chat account before on that email. Already have an old Le Chat account? The discount won’t apply to it. One more thing to check: the Vibe subscription doesn’t include API credits – if you’re planning to use the coding agent programmatically rather than through the web UI, you’ll pay for tokens on top of the subscription.

Can I run Medium 3.5 myself instead of paying for Vibe?

Yes – weights are on Hugging Face under a modified MIT license. You lose Work Mode and the connectors, but you get full inference control. Worth doing if you want to evaluate the model before committing to a seat.

Next step: open chat.mistral.ai, switch to Work Mode, run one Skill against a real inbox. That’s the only honest test of whether Vibe replaces something you already pay for – or just adds another tab.