Mistral Le Chat (now Vibe)

Le Chat is now Vibe. Mistral AI renamed its assistant on May 28, 2026, describing it as "one agent and one licence across work and code." Every conversation, setting and plan carried over, and the app still lives at chat.mistral.ai. The Le Chat name has been retired from Mistral's product navigation.

This page describes that product. The privacy posture and the API are unchanged; the model roster underneath has moved on, and is covered below.

Overview

Launched in February 2024 by Paris-based Mistral AI, the assistant evolved from a simple model showcase into a full platform before taking the Vibe name. Mistral AI is the leading European AI lab, known for publishing models that punch far above their size — delivering performance rivaling much larger proprietary models from OpenAI and Anthropic.

As of September 2026 the default model behind Vibe is Mistral Medium 3.5 — a dense 128B multimodal model with a 256K context window, released in April 2026 and the model Mistral put behind Vibe's remote coding agents. Mistral Large 3, the 675B mixture-of-experts model with 41B active parameters that Mistral shipped in December 2025, remains its largest public model and the heavier option for complex reasoning. Both carry the same 256K window.

The platform stands out for its Canvas collaborative workspace, built-in web search, connectors into more than 100 business tools, and strict European data privacy standards (GDPR-compliant by design).

What Changed in the Rename

• Work Mode is the assistant you already knew — web and mobile, multi-step tasks like inbox management, research and document drafting.
• Code Mode is a dedicated surface for remote coding agents, alongside a Mistral Vibe extension for VS Code and a terminal CLI.
• One licence now covers both, rather than separate chat and coding products.
• Chat Mode remains — the turn-based conversational experience Le Chat was known for is still one of the three modes, and all historical Le Chat data is preserved.
• The app URL did not move. vibe.mistral.ai and mistral.ai/vibe both redirect to chat.mistral.ai.

Key Features

• Ultra-Fast Responses: Mistral's models are architecturally optimized for speed, delivering noticeably faster responses than comparable models — critical for high-volume professional use.
• Canvas: A side-by-side collaborative workspace, enabled from the Tools menu, where you co-create and iterate on code, documents and structured content — with version history, a diff view and revert.
• Native Web Search: An integrated search engine that grounds answers in live web sources with inline citations, reducing hallucinations on current events.
• Code Mode: Remote coding agents that run in the cloud, plus a terminal CLI and editor integration — a Mistral Vibe extension for VS Code, and JetBrains and Zed via ACP. Codestral remains available on the API for fill-in-the-middle completion.
• Connectors and MCP: Integrations with 100+ tools including email, calendar, Slack, GitHub and Jira, with full MCP compatibility.
• Image generation and voice: Images are generated in-app through a Black Forest Labs FLUX partnership; voice input is powered by Mistral's own Voxtral.
• Document output: Generates PDFs, decks, briefs, RFP responses, PRDs and reports rather than only chat text.
• European Data Privacy: All data is processed under EU jurisdiction with strict GDPR compliance and a no-training policy for enterprise users.
• Open-Weight Transparency: The models powering Vibe are published openly — Mistral Large 3 under Apache 2.0, Mistral Medium 3.5 under a modified MIT licence — allowing independent security audits and local deployment.
• La Plateforme API: A developer-grade API platform with competitive pricing and full access to the Mistral model family.

How It Works

Technical Architecture:

• Models: Mistral Medium 3.5 (the default), Mistral Large 3 (the largest), Mistral Small 4 (fast and Apache 2.0), Codestral (code completion), Mistral Embed (embeddings).
• Context Window: 256K tokens (262,144) on both Mistral Medium 3.5 and Mistral Large 3.
• Architecture: Grouped-Query Attention (GQA) Transformer with Sliding Window Attention for long contexts.
• API Standard: OpenAI-compatible REST API via La Plateforme.
• Privacy: EU-based data processing, GDPR-compliant, zero data retention for Pro/Enterprise.

Use Cases

Professional Research & Writing

• Deep Research Synthesis: Cross-referencing live web sources to produce cited reports on complex topics.
• Technical Documentation: Using Canvas to draft, review, and iterate on API docs, specs, and whitepapers.
• Multilingual Content: High-quality translation and content creation across 30+ languages, with particular strength in French, Spanish, Italian, and German.

Software Development

• Code Generation: Using Codestral for fill-in-the-middle completion and complex algorithm implementation.
• Code Review & Debugging: Analyzing code for bugs, logic errors, and security vulnerabilities.
• API Development: Using La Plateforme to build Mistral-powered applications.

Privacy-Sensitive Enterprise

• Confidential Document Analysis: Analyzing sensitive business documents without US data jurisdiction.
• On-Premise Deployment: Running Mistral Large on internal infrastructure for full data sovereignty.

Getting Started

Step 1: Access Vibe

1. Visit chat.mistral.ai — old Le Chat links still land here.
2. Create a free account with your email.
3. Start chatting immediately; Mistral Medium 3.5 is the default model.

Step 2: Try Canvas

1. Open the Tools menu at the bottom of the chat window and tick Canvas.
2. Ask Vibe to write or generate code/text.
3. The output appears in its own pane. Edit it directly, ask for a refinement, or use the version arrows and diff view to compare and revert.

Step 3: Enable Web Search

1. In the input bar, toggle the "Web Search" icon (globe icon) to on.
2. Ask a question requiring current information.
3. Vibe will retrieve live sources and include citations in its response.

Step 4: Use the API

1. Visit console.mistral.ai and create an API key.
2. Install the Mistral Python client: pip install mistralai.
3. Make your first API call:

from mistralai import Mistral
client = Mistral(api_key="your-api-key")
response = client.chat.complete(
    model="mistral-large-latest",
    messages=[{"role": "user", "content": "Explain MoE architecture."}]
)
print(response.choices[0].message.content)

Best Practices

• Use Canvas for iterative writing and coding projects.
• Enable Web Search for any question involving current events or recent data.
• Choose the right model: mistral-small-4 for speed and cost, mistral-medium-latest for agentic and coding work, mistral-large-3 for the heaviest reasoning.

Pricing & Plans

Plan names and prices from mistral.ai/pricing, checked 6 September 2026. One licence covers Work Mode and Code Mode.

• Free ($0): Access to the Mistral model family with standard daily limits.
• Pro ($14.99/month, or $5.99/month for verified students): Higher usage limits, full web search, Canvas access, and priority speed.
• Team ($24.99/user/month, $50/month minimum): Collaboration and shared workspaces.
• Enterprise: Custom pricing with EU-based data processing SLAs, SSO, and dedicated support.
• API (Pay-as-you-go) — quoted by Mistral in USD; the pricing page offers a EUR toggle:
  • Mistral Small 4: $0.15 per 1M input tokens / $0.60 per 1M output tokens.
  • Codestral: $0.30 per 1M input tokens / $0.90 per 1M output tokens.
  • Mistral Medium 3.5: $1.50 per 1M input tokens / $7.50 per 1M output tokens.
  • Mistral Large 3: $0.50 per 1M input tokens / $1.50 per 1M output tokens.

Limitations

• Context Window: The models behind Vibe top out at 256K tokens (Mistral Large 3 and Medium 3.5) — large, but smaller than Gemini 3.8 Flash at 1,048,576 or Claude Opus 5 at 1M. Mistral does not publish the context limit Vibe itself exposes.
• Image Generation Is a Partnership: Vibe generates images, but through a Black Forest Labs FLUX integration rather than a Mistral model — so the open-weights and EU-sovereignty argument that applies to the text models does not extend to the image path.
• Agentic Features: Code Mode's remote agents narrowed the gap, but multi-step autonomous coding is still less mature than Claude Code or Cursor, and the model roster underneath is Mistral-only.
• Two Names in Circulation: Mistral retired "Le Chat" in product navigation, but the URL, much of the documentation and most third-party coverage still use it. Expect to search under both names.
• Brand Recognition: Less widely known than OpenAI and Anthropic, which can affect community support availability.

Community & Support

• Documentation: docs.mistral.ai
• La Plateforme Console: console.mistral.ai
• GitHub: github.com/mistralai
• Discord: Official Mistral Discord
• X (Twitter): @MistralAI

Related Tools

• ChatGPT
• Claude
• DeepSeek
• Grok