Overview
Le Chat is a conversational assistant product launched by French AI company Mistral AI, powered by its proprietary models. On April 30, 2026, Mistral released the flagship model Medium 3.5 (128B dense architecture), merging instruction following, reasoning, and coding capabilities into a single weight set, achieving a SWE-Bench Verified score of 77.6%. It also introduced the Vibe cloud-based programming Agent system, moving coding agents from local to the cloud.
Key Features
- Medium 3.5 flagship model: 128B dense parameters, non-MoE architecture, instruction/reasoning/coding three-in-one, 256K context, open-source weights
- Vibe remote programming Agent: Cloud-based coding agent, enabling code writing, debugging, and deployment without a local environment
- Multilingual advantages: Developed by a French team, with outstanding multilingual processing capabilities in Europe, balanced experience in Chinese, English, French, German, etc.
- Canvas mode: Supports collaborative document editing, code generation, and visual content creation
- Web search: Real-time web search for the latest information, with source citations in responses
Use Cases
- Daily conversation and knowledge Q&A
- Code writing and debugging (with Vibe Agent)
- Multilingual content translation and creation
- Document analysis and long text processing (256K context)
- Data compliance scenarios for European enterprises (GDPR-friendly)
Pros
- Medium 3.5 open-source weights, SWE-Bench 77.6% strong coding capability
- Vibe cloud Agent enables programming without a local environment
- European company, data compliance superior to US competitors
- Free version is feature-rich, allowing experience of the flagship model without payment
- 256K ultra-long context support
Pricing
Le Chat's free version allows use of Mistral's latest models. API is billed per token, Medium 3.5 is available via La Plateforme. Enterprise version and Vibe Agent require contacting sales.
Summary
With the release of Medium 3.5, Le Chat has leaped from a second-tier product to a first-tier competitor. The coding capability of the 128B open-source model is approaching GPT-5.5 and Claude Opus. For users who value data compliance (GDPR) and multilingual capabilities, it is a highly recommended choice.