Mistral AI - French chatbot - 2026
Published: 2026-08-03
Mistral AI is a French artificial intelligence company founded in April 2023 by former Google DeepMind and Meta researchers: Arthur Mensch, Guillaume Lample, and Timothée Lacroix. The company is headquartered in Paris and is widely regarded as Europe's main answer to the American giants OpenAI, Google, and Anthropic. Its flagship consumer product is the chatbot Le Chat, which was renamed Vibe in 2026. The old name Le Chat is still widely used in the media and by the user community.
History and positioning
Founding and philosophy
Mistral AI was founded with the ambitious goal of creating a European alternative to American and Chinese AI labs. The company's key principles:
- Open Weights — most models are distributed with open weights under Apache 2.0, Modified MIT, or CC BY-NC 4.0 licenses, allowing companies to deploy them on their own infrastructure;
- Privacy — support for Zero Data Retention (ZDR), option to opt out of using data for model training;
- Enterprise focus — solutions for the financial sector, public sector, manufacturing, and energy;
- Efficiency — Mistral models have traditionally combined strong performance with moderate computational costs.
Timeline of key events
| Date | Event |
|---|---|
| April 2023 | Company founded by Arthur Mensch, Guillaume Lample, and Timothée Lacroix |
| September 2023 | Release of the first open-weights model Mistral 7B |
| 2024 | Launch of Le Chat — a browser-based chat assistant |
| 2025 | Mobile apps for iOS and Android, introduction of Flash Answers, Canvas, and image generation |
| June 2025 | Premiere of the Magistral (Small/Medium) family — models for multilingual reasoning |
| December 2025 | Release of Ministral 3 (3B/8B/14B) for edge devices |
| February–March 2026 | Voxtral TTS (speech synthesis) and Voxtral Mini Transcribe Realtime (speech recognition) |
| July 2026 | Le Chat renamed to Vibe — a unified agent for productivity and coding |
Le Chat renamed to Vibe
In July 2026, the company announced that Le Chat becomes Vibe — a single product combining:
- Vibe for work (formerly Le Chat) — an AI assistant for chat, documents, and research;
- Vibe for code — a coding agent for the terminal, IDE, and background tasks.
During the rename, all conversations, settings, and subscriptions were automatically migrated. An official article titled "Le Chat is now Vibe" was published in the Mistral Help Center, confirming product continuity. The name Le Chat continues to be used colloquially and in parts of the documentation.
Product ecosystem
Vibe (formerly Le Chat)
Vibe is an AI agent for long-horizon tasks, available as a web app and mobile apps. It is positioned as a counterpart to ChatGPT, Google Gemini, and Claude, with an emphasis on European privacy standards.
Vibe for code
A separate direction — coding agents:
- CLI tool installable from the terminal;
- integration with VS Code, JetBrains, and Zed via the ACP (Agent Client Protocol);
- remote agents — launching coding tasks directly from Vibe, with the ability to continue in the CLI.
Studio
Developer console and Mistral API: API keys, Playground, model evaluation, agents, and SDKs for building applications (client-python, client-ts).
Forge
A platform for training, aligning, and evaluating custom models on proprietary enterprise data — not just fine-tuning, but full custom training.
Compute
Frontier-scale infrastructure for model training and inference.
Admin
Control plane for organization setup, billing, SSO, Workspaces, and access policies.
Vibe capabilities
Main features of the assistant:
- 💬 Natural language conversation;
- 🌐 Web search;
- 📄 Document analysis — PDF, Word, Excel, and other formats;
- 🖼️ Image analysis;
- 🎨 Image generation;
- 🐍 Python code execution (Code Interpreter);
- 🔎 Deep Research — multi-step investigations;
- 📚 RAG — working with your own documents and libraries;
- 👨💻 Code generation and analysis;
- 🗣️ Voice input powered by Voxtral;
- 🔌 100+ external tool integrations (email, calendar, Slack, GitHub, Jira) via connectors;
- 🧩 Full MCP (Model Context Protocol) compatibility;
- ⏰ Task scheduling;
- 🧠 Memories — personalization based on interaction history;
- ✅ Verified News;
- 🌍 Support for dozens of languages.
Most features are available even on the free plan, with limits on the number of requests.
Mistral models (2026)
The model catalog includes several families covering scenarios from cloud inference to on-device deployment.
| Model | License | Purpose |
|---|---|---|
| Mistral Large | Commercial | Flagship model for complex cloud tasks |
| Mistral Medium 3.5 | Modified MIT | Frontier-class, multimodal; optimized for agents and coding (v26.04) |
| Mistral Small 4 | Apache 2.0 | Hybrid model: instructions, reasoning, and coding in one model (v26.03) |
| Magistral (Small/Medium) | Apache 2.0 (Small) | Dedicated multilingual reasoning, 24B (Small) |
| Devstral 2 | Commercial | Autonomous coding agents |
| Codestral | Commercial | Low-latency code completion in IDEs, 32K–256K context |
| Ministral 3 (3B/8B/14B) | Apache 2.0 | Edge and on-device models, 131K–262K context |
| Voxtral TTS | CC BY-NC 4.0 | Text-to-speech, zero-shot voice cloning, multilingual |
| Voxtral Mini Transcribe Realtime | Apache 2.0 | Real-time speech recognition |
| OCR 4 | Premier | Document extraction and understanding, paragraph-level bounding boxes |
| Embedding Models / Codestral Embed | Commercial | Text vector representations |
Since 2023, the company has released new models on average every ten weeks.
Pricing (2026)
Vibe offers several plans — from free to Enterprise. Prices are current as of mid-2026.
| Plan | Price | Key features |
|---|---|---|
| Free | 0 | Limited messages, web searches, and document uploads; up to 5 scheduled tasks; image generation; 100+ connectors |
| Education (Student) | $5.99/month | Pro-level features for verified students |
| Pro | $14.99/month | Up to 6× free-tier messages, up to 5× web searches, up to 40× image generations, 15 GB libraries, unlimited task scheduling, custom MCP connectors (beta), 150 Flash Answers/day |
| Team | on request | Team management, corporate controls |
| Enterprise | $24.99/user/month and up | SSO, Zero Data Retention, custom deployment, SLA |
Pro users can extend limits with PAYG credits (pay as you go).
Free plan limits
- Messages — limited;
- Web search — limited;
- Document upload — limited;
- Task scheduling — up to 5;
- Model training on your data — opt-out available.
Security, privacy, and compliance
Mistral places a strong emphasis on trust and security:
- Encryption of data in transit and at rest;
- Zero Data Retention (ZDR) — a mode in which data is not stored;
- Training opt-out — users can prohibit the use of their data for model training;
- SOC 2 certification;
- Data controls — export, deletion of conversations and accounts;
- Enterprise adoption — models are used in finance (HSBC), manufacturing (BMW), logistics (CMA CGM), and high-tech industry (ASML).
For developers
API and documentation
Official documentation, API Reference, and SDKs are available:
pip install mistralai # Python SDK npm install @mistralai/mistralai # TypeScript SDK
The official Mistral GitHub organization includes:
- client-python — official Python SDK;
- client-ts — official TypeScript/JavaScript SDK;
- cookbook — usage examples;
- mistral-common — shared utilities;
- mistral-finetune — fine-tuning tools;
- mistral-vibe — tools related to Vibe.
Getting an API key
API keys are issued in the Mistral Studio console (console.mistral.ai). Free quotas for experimentation are available immediately after registration.
Comparison with competitors
| Criterion | Mistral (Vibe) | ChatGPT (OpenAI) | Claude (Anthropic) | Gemini (Google) |
|---|---|---|---|---|
| HQ | Paris, France | San Francisco, USA | San Francisco, USA | Mountain View, USA |
| Open weights | Yes (Apache 2.0, MIT) | No | No | No |
| Focus | Privacy, enterprise deployment | Broad consumer | Code, safety | Google ecosystem |
| Free plan | Yes (limited) | Yes (limited) | Yes (limited) | Yes (limited) |
| Distinction | European compliance, edge models | Multimodality, agents | Long context, Claude Code | Workspace integration |
Help Center and support
The Mistral Help Center covers in detail:
- plans and subscription upgrade/cancellation rules;
- Free plan limits;
- Enterprise plan capabilities;
- security and data storage;
- settings (two-factor authentication, password, email);
- Canvas, Web Search, Code Interpreter, Image Generation;
- refund policy;
- child usage rules and ethical practices;
- data export and deletion procedures.
Summary
Mistral AI is Europe's leading AI lab, founded in 2023 by alumni of Google DeepMind and Meta. Its consumer chatbot Le Chat, renamed Vibe in 2026, has become a unified agent for productivity and coding, available free with limits and via a Pro subscription for $14.99/month. The company ships open-weights models (Mistral Small 4, Ministral 3, Magistral, Voxtral), flagship models (Mistral Large, Medium 3.5), and specialized services (OCR 4, Codestral), serves enterprise customers such as HSBC, BMW, and ASML, and provides full SDKs and APIs for developers. Its key differentiators from American competitors are openness, privacy, and European regulatory compliance.
Links
