Mistral AI

Mistral is bringing together the inference infrastructure, open models, and long-term commitments Europe needs to control its AI future, and setting a roadmap for the world. ⚡ A coalition that secures long-term AI capacity - Compute capacity in Europe remains difficult to secure on the scale and timelines serious AI programs require. We’re bringing together a group of enterprises whose multi-year compute commitments, European Compute Units, can support Mistral infrastructure at a scale no participant could secure alone. By aggregating these long-term commitments, the group helps determine what capacity is built, where it’s located, and whom it serves. 🌍 Regional control and production-grade reliability - Enterprises need to have the ability to keep their data in-region and rely on local capacity when demand peaks. Mistral Compute Regional Endpoints, now in public preview, are dedicated access points that give customers control over where their data is processed (either in Europe or the U.S.) to align with their data residency, regulatory, and latency requirements. 🎯 Sovereign intelligence built on model choice - Different jobs call for different models: deep reasoning, high-volume production, and work shaped around a company's own data. Open weights let customers see inside a model, adapt it, and keep the intelligence they build with it. To that end, we’re expanding our platform to third-party open models, starting with Z.ai’s GLM-5.2. This gives enterprises a broader foundation on which to build so they can match workloads to the right models. The world needs an open-source platform - and that’s exactly what we’re building to give our customers more choice and the flexibility to choose the right model for the right task. As part of this, we’ll continue to innovate in frontier, efficient open models across modalities. Read our blog to learn more about the announcement and what customers like Amadeus, ASML, CMA CGM, Factory, and Groupe Caisse des Dépôts are saying about this announcement. https://lnkd.in/gXB7C358

🛡️Introducing Shieldstral: Mistral's 3B open-weights content moderation model that lets teams define what counts as safe or not in the AI models they use. An open-source content safety classification model is ideal for enterprises: it blends privacy, flexibility and governance so brand standards and regulatory requirements can be more uniquely reflected. Shieldstral’s size and interoperability is key to faster and efficient content evaluation, and the model can be deployed on-device to reduce latency and keep sensitive data private. Rather than returning a yes or no label for potentially unsafe content, Shieldstral supplies a numerical score so teams can rank content by confidence. One interface handles text and image prompts too, which means the same model evaluates and operates. As an inaugural partner in the Open Secure AI Alliance, which focuses on improving AI safety and security through open-source frameworks, Mistral is making Shieldstral available today under Apache 2.0 — putting production-grade content classification tools in the hands of every team building on open models. Read more: https://lnkd.in/g-H4_cAs

Mistral is at the #AI42026 Conference this week in Las Vegas. Come and meet our team at booth #900 and tune into our session at 9:30am PT Tuesday, August 4th. We’ll be talking about open source AI, our latest innovations in audio models and more.

🚀 Today we’re announcing an expanded global strategic partnership with Microsoft to give enterprises and regulated industries frontier AI they can control. As Mistral is expanding its AI compute capacity in Europe, Mistral and Microsoft are expanding their strategic partnership with Microsoft’s commitment to leverage part of this capacity. This brings Mistral’s frontier models across Microsoft’s AI platform and gives customers flexible deployment options from cloud to fully disconnected environments. There are several key pillars to the expanded partnership: 🏢Scaling Europe’s AI compute capacity: Microsoft and Mistral are announcing a new agreement to expand AI infrastructure in Europe. Microsoft will leverage Mistral’s expanded Europe-based GPU infrastructure to increase capacity for AI development and to support the delivery of Microsoft’s cloud and AI services. This represents a multibillion dollar commitment from Microsoft and an important way for Microsoft customers to benefit from Mistral’s scientific and compute innovations. 🛠️ Integrating Mistral models into Microsoft enterprise products: Mistral Medium 3.5 and OCR 4 are now available in Microsoft Foundry, and Mistral Medium 3.5 is now in Microsoft Copilot Studio. This brings the benefits of Mistral’s frontier, efficient and multilingual models to Microsoft customers globally, allowing developers to build, customize and operate AI applications. 🗝️ Giving enterprises greater control over AI at scale: Azure enables organizations to deploy Mistral models across cloud, cloud-connected and fully disconnected environments, while maintaining control over data, operations and business continuity. Arthur Mensch and Brad Smith share more details on what this partnership means for enterprises around the world 👇 Watch here: https://lnkd.in/eScXnT9Y

Ever wondered what it’s like to shape how enterprises adopt and deploy AI? You’re in luck: We’re growing the Solutions team at Mistral! We’re hiring globally, and we’re looking for people with an entrepreneurial mindset, who care about building things that matter. Think this could be your next role? Join us: https://lnkd.in/etstGUxN

Introducing Robostral Navigate: Mistral’s state-of-the-art robotics navigation model that lets robots navigate complex environments using just a single camera and plain-language instructions - beating the best single-camera approach by 9.7 points, using 22x fewer tokens, and working across any robot form factor from wheeled to legged robots. Trained entirely in simulation (no real-world data collection), this breakthrough could transform manufacturing, logistics, and delivery. It eliminates the need for costly LiDAR and depth sensors, dramatically reducing deployment costs. Its hardware-agnostic design enables rapid scaling across diverse robot fleets. Plain-language instructions allow non-technical staff to operate robots without specialized training. This puts AI in the hands of workers across the manufacturing sector and puts the power of embodied frontier AI in everyone’s hands. Read more: https://lnkd.in/gGZ8PVrA

Upcoming livestream: OCR4 in Production. We recently released Mistral OCR 4, featuring bounding boxes, block classification, and inline confidence scores alongside extracted text. Join us to see a demo and to find out how OCR4 can be used inside workflows and Search Toolkit (Mistral's open-source, composable search framework) to get production-grade indexing.

We're building frontier AI systems to solve some of the world's hardest problems, from transforming global enterprises to preserving human history. 🌐 Cisco built an AI renewals agent that consolidates data from 50+ sources, automates proposal creation, and targets cutting renewal prep time by up to 20%. 🏦 HSBC is powering agentic workflows across global banking operations, automating financial analysis, multilingual document processing, and risk evaluation at scale. 📄 The European Patent Office co-developed a specialized OCR model to transform complex, multilingual patent documents into structured, AI-ready data, supercharging prior-art search and examiner productivity. 🏛️ The Austrian Academy of Sciences is training Apollo, the world's first LLM for Ancient Greek, enabling researchers to reconstruct and decipher 1 million+ unread papyri in hours, not years. Read their stories: mistral.ai/customers

What does it take to deploy frontier AI inside a global bank? BNP Paribas needed a partner who could run models on their own infrastructure, meet strict security and compliance requirements, and be energy-efficient at scale. Since the partnership began, our collaboration has expanded from models to co-created solutions, research, and products. We’re looking forward to even more co-innovation in the future.

Today, we're releasing Mistral OCR 4 - our state of the art document intelligence model. In order for businesses to deploy AI in the enterprise, they need to take the first step of digitizing decades of documents to ensure their customized AI systems have the full context of their organization. OCR 4 is a critical step in that journey. OCR 4 supports 170 languages across 10 language groups, with real gains on the rare and low-resource documents where most systems fall off. It's small enough to run fully self-hosted on a single container, so sensitive data never leaves your environment. Independent annotators preferred OCR 4 over all other systems tested, with win rates averaging 72%. It also takes the top overall score on a leading public industry benchmark, OlmOCRBench. It’s available today via the API, Document AI in Mistral AI Studio, Amazon SageMaker, Microsoft Foundry, and coming soon Snowflake Parse Document. Read more: https://lnkd.in/gQR36HA8