Get started
Five steps to become productive on Le Chat
Follow these steps in order. Each one unlocks the next. A free account is enough to test everything.
Step 1: create an account
Sign up at chat.mistral.ai in 60 seconds
Open chat.mistral.ai/signup. Three options: email, Google, Microsoft, Apple ID. No card required. Automatic email verification. Free account active immediately.
Students: verify with a .edu email to unlock the $6.99/month Pro tier (vs $14.99 standard). Annual re-verification required.
Step 2: enable Memories
Memories to persist context across sessions
Settings -> Privacy -> Memories. Enable (opt-in by default). Memories retains your preferences: professional role, tech stack, writing style, project context. Free up to 500 memories.
Difference from Projects: Memories is global across all conversations. Projects is local to a specific folder. Memories does not feed training, independent of the No Telemetry toggle.
Speed
Flash Answers
Streaming at 1,100 tok/s on Cerebras WSE-3.
Enable via the lightning icon in the model bar. Responses arrive near real-time thanks to Cerebras + Mistral speculative decoding (official announcement, February 6, 2025). Available on Free and Pro with different caps. Most visible on long replies, translations, and streaming code.
Research
Deep Research
Multi-source reports with clickable citations.
Enable Deep Research in the tools menu. Give an open topic (market, technology, regulation). Le Chat chains 10-30 web queries, reads sources, structures a note with citations. Takes 5-15 minutes. Output: introduction, sections, data tables, summary, conclusion.
Editing
Canvas
Side-by-side editor for text, code, tables.
Tools -> Canvas. Opens an editor panel on the right. Ideal for rewriting a contract, polishing an essay, debugging code section by section. You edit directly or ask Le Chat to suggest edits. No real-time multi-user collaboration (May 2026).
Organization
Projects
Task containers: chats + files + tools.
Sidebar -> New Project. Each Project has its own Library (active tools, system prompt, context). Ideal for a client project or research dossier. Limitation: files are attached to a chat inside the Project, not to the whole Project (Mistral acknowledges this, working on it).
Integrations
MCP Connectors
22 official connectors, free on Free tier.
Settings -> Connectors. GitHub, GitLab, Snowflake, Databricks, Atlassian (Jira/Confluence), Linear, Sentry, Stripe, PayPal, Box, Notion, Asana, Monday.com, Zapier, Brevo, Pinecone, Prisma Postgres, DeepWiki, Cloudflare Development Platform, Plaid, Square. OAuth 2.1 auth. User confirmation for each action.
Creation
Image generation
Black Forest Labs FLUX inside Le Chat.
Type 'generate an image of...' in chat. BFL models (FLUX Ultra was the first announced in February 2025). Resolutions 16:9, 4:3, 3:2, 1:1. Free 40 per day, Pro 40x that volume with 15-burst over 3 hours. Censorship: violent/copyright prompts may be blocked.
Going further
Four power-user techniques
Once the basics are in, these techniques separate Le Chat power users from casual ones.
Turn on No Telemetry on Pro for sensitive data
Pro / Team / Enterprise only. Settings -> Privacy -> No Telemetry. Your prompts and responses are not stored or used for training. Memories and Connectors already have their own opt-out by default, but No Telemetry covers everything else.
Combine Memories and Projects
Memories for global preferences (your role, style, stack). Projects for context on a specific subject (client X, product Y). Together, you avoid repeating yourself. Do not pile everything into Memories: 500-entry cap.
Use Canvas for long documents
To rewrite a contract, essay, or article: Canvas + section by section beats asking for a full rewrite. The model keeps more context, you keep control over structure.
Run Deep Research in the background
Deep Research takes 5-15 minutes. Launch it, do other work in another chat, come back for the report. Good for market briefs, literature reviews, product comparisons before a decision.
Pick the right model
Which Mistral model for what
Le Chat lets you switch between several models. Here is when to use which.
Fast everyday chat
- Mistral Small 4 (March 2026) or Mistral Medium 3.5 without reasoning.
- With Flash Answers enabled: near real-time response.
- Enough for 80% of tasks: email, translation, brainstorming.
Deep reasoning and complex code
- Mistral Medium 3.5 with reasoning_effort=high.
- Or Mistral Large 3 (December 2025) for long contexts (256K tokens).
- Magistral Medium 1.2 also keeps its place for complex reasoning chains.
Code and IDE completion
- Codestral 2508 (256K context) for code via API.
- Devstral Small / Medium for lighter code agents.
- In the Le Chat UI, Medium 3.5 reasoning is enough for most code tasks.
Using the API
Migrate or start with the Mistral API
OpenAI-compatible, Python SDK, JavaScript SDK, Free tier 1B tokens/month for evaluation.
Create a console account, generate an API key. Free tier: 1 billion tokens per month, no credit card. Enough to evaluate and prototype. Beyond that: Scale plan, pay as you go.
All endpoints: chat completions, embeddings, FIM (fill-in-the-middle), agents, OCR, audio transcriptions, files, fine-tuning. OpenAPI schema available.
pip install mistralai. asyncio compatible. Covers all endpoints. Full type hints.
npm install @mistralai/mistralai. Side-effect-free, supports Node, Bun, Deno. Validated schema.
Base URL: https://api.mistral.ai/v1. Chat Completions format 99% compatible. Differences: tool_choice='any' equivalent to OpenAI 'required', random_seed instead of seed.
Large 3: $0.50 / $1.50 (in/out per 1M). Medium 3.5: $0.40 / $2.00. Small 4: $0.15 / $0.60. Codestral, Voxtral, OCR have their own grids.
Common traps
Three mistakes new users make
Avoid these to save time.
Memories is global (all conversations). Projects is local (specific folder). Do not pile everything into Memories - it becomes unmanageable past 50 entries. Use Projects for client / project context.
Large 3 is slower and more expensive (in API). For everyday chat and simple questions, Mistral Small 4 or Medium 3.5 without reasoning are nearly as good and much faster. Reserve Large 3 for long 256K contexts.
Many Pro users have never launched it. It is the tool with the most added value versus ChatGPT and Claude. Test on a topic you know, judge source quality and analysis depth.
Official documentation
8 official references to dig deeper
All links below are official Mistral or Cerebras. For anything beyond this tutorial, start here.
Le Chat (official product)
chat.mistral.ai - the official Le Chat product from Mistral. A Free account is enough to test everything.
Open Le ChatMistral Help Center
Official FAQ for key features: tiers, image gen, Voice, Code Interpreter, Deep Research, Memories, Projects, student discount.
Help CenterComplete Mistral documentation
docs.mistral.ai: everything for developers - API, models, agents, fine-tuning, evaluation, deployment.
Mistral docsCerebras x Mistral: Flash Answers
Cerebras blog from February 2025 explaining the speculative decoding architecture and speed figures.
Read Cerebras blogMCP Connectors + Memories (announcement 2025-09-02)
Official announcement of connectors and Memories. Full list of the 22 initial connectors.
MCP announcementLe Chat dives deep (announcement 2025-07-17)
Launch of Deep Research, Voice (Voxtral), Projects, image improvements. Official source for Deep Research Pro GA.
Deep Research launchPricing and tiers (official)
mistral.ai/pricing - canonical source for Free, Pro, Team, Enterprise tiers and the API price grid.
Official pricingLe Chat capabilities overview
Structured overview of Le Chat features with caps and behaviors by tier.
Capabilities overviewPut it into practice
Start with a real prompt now
The best tutorial: a real test on a real task. No signup, no card. Five minutes is enough to see if Le Chat fits your workflow.