Inside Mistral's AI Now Summit: Vibe Replaces Le Chat, a New 128B Model, and Deals with Airbus & BMW

Vibe is structured around two primary modes:

• Vibe Work is the productivity hub. It runs on web and mobile, connecting to tools like Google Workspace, Outlook, SharePoint, Slack, and GitHub to handle multi-step professional tasks—summarizing emails, drafting reports, analyzing spreadsheets, and scheduling recurring workflows . It includes Skills, Workflows, Connectors, Libraries, and the ability to run tasks on a schedule .
• Vibe Code is the developer-facing mode. It encompasses the Vibe CLI, a VS Code extension, and Vibe Code Web for remote coding sessions that run asynchronously in managed cloud sandboxes . The remote agents can work in parallel, open pull requests on their own, and notify developers when tasks are complete .

A third legacy mode, simply called Chat, preserves the old Le Chat experience for workflows that haven't yet migrated .

This consolidation is a direct shot at GitHub Copilot, Cursor, and other AI-assisted development tools, but with a distinctly European pitch around data sovereignty and enterprise compliance.

Mistral Medium 3.5: The 128B Brain Behind the Platform

Powering the new Vibe experience is Mistral Medium 3.5, a dense 128-billion-parameter model with a 256,000-token context window . Released in public preview on April 29, 2026, it is Mistral's first "flagship merged model," unifying instruction-following, reasoning, and coding capabilities into a single set of weights .

This single model replaces three predecessors: Mistral Medium 3.1, Magistral, and Devstral 2 . The architectural unification eliminates the need to maintain separate models for different task types and simplifies deployment.

Key technical details:

• Dense 128B parameters, not a mixture-of-experts (MoE) architecture .
• 256K context window, supporting long-horizon tasks across large codebases and document sets .
• Multimodal input: accepts both text and images, with text output .
• Configurable reasoning mode that can be toggled on or off depending on the task .
• Ships natively in FP8 precision, which allows the full model to fit on a single H200 node or two H100 nodes—reducing hardware costs compared to comparably capable MoE systems .

The model was released on Hugging Face under a modified MIT license, continuing Mistral’s open-weight distribution strategy . On the SWE-Bench Verified benchmark, it scores 77.6%, competitive with top-tier coding models though not surpassing Claude Sonnet 4.6 .

Pricing Tiers Carried Over to Vibe

Mistral did not announce new pricing at the summit. The subscription structure transitioned directly from Le Chat into Vibe, based on previously established tiers :

All tiers from Le Chat, including Pro and Teams, remain fully intact within the Vibe platform .

Developer Tooling: VS Code Extension and Remote Agents

Developers gained several new tools alongside the rebranding:

• VS Code Extension: Vibe now integrates directly into Visual Studio Code, allowing developers to generate, edit, and review code without leaving their editor .
• Vibe CLI: A command-line interface for terminal-based coding workflows, already familiar to early adopters .
• Remote Coding Agents: Cloud-based sessions that run asynchronously in isolated sandboxes, can open pull requests automatically, and notify developers upon completion. This feature, powered by Mistral Medium 3.5, represents one of Mistral's most direct attacks on the GitHub Copilot ecosystem .

The remote agents are designed for complex, multi-file software engineering tasks that require long-running compute—ideal for refactors, test generation, and dependency upgrades .

Industrial AI: Airbus, BMW, and "Mistral for Industrial Engineering"

A major strategic signal at the summit was Mistral's aggressive push into physical industries. The company announced Mistral for Industrial Engineering, a new vertical offering that combines large language models with physics simulation capabilities, designed to accelerate design, eliminate simulation bottlenecks, and optimize asset performance in manufacturing environments .

Two landmark partnerships were confirmed:

• Airbus signed a five-year contract to integrate Mistral AI across its commercial aircraft, helicopter, defense, and space divisions. The partnership covers everything from initial design and simulation to onboard cockpit capabilities .
• BMW Group will apply Mistral's AI to car crash simulations and broader manufacturing workflows, focusing on complex engineering problems in automotive production .

The choice of Airbus and BMW as inaugural industrial partners is deliberate: both are sovereign European industrial champions with massive R&D footprints, security-sensitive data, and a strong incentive to avoid American cloud providers .

The assistant's draft mentioned a partnership with ASML and a "Search Toolkit" for fake news detection with CMA CGM, but these claims could not be verified from the available sources. No independent reporting from the summit confirms ASML as a named partner announced at this specific event, and the alleged open-source Search Toolkit framework with Voxtral does not appear in any of the source materials.

Infrastructure: A New Data Center in Les Ulis

Mistral confirmed it will open a new inference-focused data center in Les Ulis, Essonne (south of Paris) in the second half of 2026 . The facility will deliver 10 megawatts of computing power dedicated to running Mistral's models .

This data center is part of a broader infrastructure roadmap:

• Existing facilities include a 44-megawatt data center outside Paris and another under construction in Sweden .
• The company targets 200 megawatts of total computing capacity by the end of 2027 .
• Long-term ambition: 1 gigawatt by 2030, roughly equivalent to a nuclear reactor's output .

The investment is part of a €4 billion strategy to build sovereign European AI infrastructure that keeps sensitive data within the continent .

The Bigger Picture: From Model Maker to Full-Stack Platform

Taken together, the AI Now Summit announcements reveal a company executing a deliberate pivot. Mistral's CEO Arthur Mensch framed the moment bluntly: "The most important use cases for AI are located in research and development and the creation of physical objects" .

Mistral now covers the full AI value chain:

• Infrastructure: Own data centers for inference compute.
• Models: Open-weight distribution via Hugging Face, with enterprise deployment options.
• Applications: Vibe for productivity and coding, replacing Le Chat.
• Vertical solutions: Mistral for Industrial Engineering, with Airbus and BMW as anchor clients.

The company is positioning itself as the enterprise AI provider for organizations that refuse to hand sensitive data to American hyperscalers—a sovereignty-first alternative that leverages open-weight model distribution while monetizing through commercial platform offerings and industrial partnerships .