Mistral AI
The Paris-based AI company: its assistant Le Chat is now Vibe, with Work, Code and Chat modes, plus open-weight models you can run yourself. A free mini-course from basics to power moves.
What it nails
- Work mode with Skills, Canvas and 100+ connectors
- Open-weight models (Apache 2.0 / modified MIT)
- Pro plan from $14.99
Watch out for
- Old Le Chat features are being retired
Mistral AI is a Paris-based AI company. Its assistant, formerly Le Chat, is now Vibe: "Mistral's unified agent for professional productivity and coding tasks", still at chat.mistral.ai, with three modes: Work, Code and Chat. Mistral also publishes open-weight models, from the tiny Ministral 3 to Mistral Medium 3.5.
Start here: the basics
- Open chat.mistral.ai or the mobile app. Your old Le Chat conversations move with you.
- Choose a mode: Work for documents, research and tasks; Code for developers; Chat for quick questions (being phased out).
- Use the Fast / think toggle to switch "between quick conversational responses and deeper, agentic workflows".
- Treat outputs as drafts. Mistral's own docs ask you to check facts, dates and sources.
Mistral's prompting rules
- Give a purpose: "You are a <role>, your task is to <task>."
- Structure it: "Organize instructions hierarchically or with a clear structure".
- Format with Markdown or XML tags, which are "Readable" and "Parsable".
- Show a few examples of the output you want.
- Avoid blurry words like "too long", "many" or "few", and do not rely on exact word counts.
You are a senior HR advisor. Your task is to turn the notes below into a clear onboarding checklist for a new remote employee. <notes> laptop shipped day -3, accounts day 1, buddy assigned, first 1:1 end of week 1, security training within 14 days </notes> Format: a Markdown checklist grouped by week.
Example by Mistral AI · source ↗ · A teaching prompt built on Mistral's advice: "You are a
Models and plans
| Model | What it is | Licence |
|---|---|---|
| Mistral Medium 3.5 | Frontier multimodal model for agents and coding; 128B dense, 256k context | Open weights, modified MIT |
| Mistral Small 4 | Reasoning, vision and coding in one model; 119B total, 6B active (16 March 2026) | Apache 2.0 |
| Mistral Large 3 | Open-weight multimodal model (2 December 2025) | Apache 2.0 |
| Ministral 3 | Small 3B, 8B and 14B models; the 3B even runs in a web browser | Apache 2.0 |
| Plan | Price | What you get |
|---|---|---|
| Free | $0 | Web and mobile, limited messages and searches, image generation, 100+ connectors |
| Pro | $14.99 / month ($5.99 for students) | More messages, searches and images, all-day coding, $15 a month in API credits |
| Team | $24.99 per user / month | Up to 30GB storage per user, domain verification, data export |
Power moves
1. Work mode: multi-step tasks with approvals
Work runs multi-step jobs across your tools and "shows progress and asks for confirmation before sensitive actions".
Research this topic and create a one-page brief I can share with my team.
Example by Mistral AI · source ↗ · An official example. Add your topic.
Summarize this PDF and extract owners, deadlines, and action items.
Example by Mistral AI · source ↗ · An official example. Attach your PDF first.
2. Skills, Deep Research and Canvas
Skills are "reusable instructions and resources" you call by typing / or /skill-name. /deep-research runs the Deep Research skill, and /canvas opens the built-in editor for text, data, code and presentations.
3. Connectors
More than 100 connectors bring your company tools and knowledge into Vibe, on every plan.
4. Images with Black Forest Labs models
"Work uses Black Forest Labs models for image generation." Create a picture, then edit it in the same chat. Daily limits depend on your plan.
Create a flat illustration of a modern office interior with natural light and plants.
Example by Mistral AI · source ↗ · An official example.
Add a whiteboard on the back wall. Keep the lighting and color palette unchanged.
Example by Mistral AI · source ↗ · The official follow-up edit, sent in the same chat.
5. Voice in 13 languages
Tap the microphone to talk to Vibe on web and mobile, on every plan. It runs on Mistral's Voxtral models.
6. Run Mistral models yourself
Medium 3.5 can be self-hosted "on as few as four GPUs"; Ministral 3 even runs in a web browser.
write a haiku about this
Example by Simon Willison's Weblog · source ↗ · Made with Ministral 3 3B running locally in the browser on a webcam feed.
Generate an SVG of a pelican riding a bicycle
Example by Simon Willison's Weblog · source ↗ · A famous informal test. Made with Mistral Large 3.
Generate an SVG of a pelican riding a bicycle
Example by Simon Willison's Weblog · source ↗ · Made with Mistral Small 4.
A NewsGuard audit reported in April 2026 found Le Chat repeating state-sponsored false claims, especially when prompts were leading or malicious. The product is changing fast: Memories, Agents and the old Chat tab are being retired in favour of Work and Skills. On Free and Pro, voice may be used for model improvement unless you opt out.
FAQ
Is Le Chat gone?
Le Chat has become Vibe. It is still at chat.mistral.ai and your conversations and plan carry over.
Is Mistral free?
Yes. The Free plan includes web and mobile access, image generation and 100+ connectors, with limited messages.
What is the difference between Work, Code and Chat?
Work handles documents, research and multi-step tasks; Code is for developers; Chat is the legacy quick-answer mode being phased out.
Are Mistral models open source?
Many are open-weight: Small 4, Large 3 and Ministral 3 under Apache 2.0, and Medium 3.5 under a modified MIT licence.
Where are Magistral and Devstral?
They are deprecated; their reasoning and coding skills are now built into Mistral Small 4.
Sources: Le Chat is now Vibe · Mistral pricing · Models overview · Mistral Medium 3.5 · Prompting guide