Mistral AI
AI Lab
French frontier lab, open-weight and commercial models.
Recent activity
-
Studio provides a system of record for AI prompts and skills—versioned, owned, and traceable. Iterate fast, ship with control, and ensure consistent AI behavior.
Read more → -
Introducing Robostral Navigate: 8B model achieving 76.6% on R2R-CE with just a single RGB camera. No depth sensors, LiDAR, or multiple cameras needed.
Read more → -
Leanstral 1.5, a free Apache-2.0 licensed model with 6B active parameters, delivers a major performance upgrade in formal verification, achieving state-of-the-art results on FATE-H (87%) and FATE-X (34%) and solving 587/672 PutnamBench problems, and uncovering 5 previously unknown bugs across 57 repositories tested. AI summary
Read more → -
Mistral AI has introduced new admin controls to manage connector access per workspace or org, while also providing API keys with scoped permissions and multi-account connectors to enhance security for automated AI workloads. The Connectors Debugger also provides detailed root cause analysis for broken connections. AI summary
Read more → -
Mistral OCR 4 introduces a new, state-of-the-art (SOTA) optical character recognition (OCR) system that outperforms leading OCR systems in human evaluations and benchmarks, supporting 170 languages and running in a single container for self-hosted deployments. The model provides structured outputs with bounding boxes, block classification, and inline confidence scores, enabling downstream systems to access both the extracted text and the context of each element. AI summary
Read more → -
Mistral unveiled a new AI stack for industrial engineering, integrating advanced physics models, engineering expertise, and robotics to optimize design, simulation, and production while ensuring data security. Their agentic AI tool, Vibe, now handles long-horizon tasks like coding and research, while a new 10 MW data center in France will enhance compute control and security. AI summary
Read more → -
Vibe is Mistral's unified AI agent that handles long-running tasks like inbox management, research, and document drafting, as well as coding workflows from feature development to pull requests, with features like Work Mode for multi-step tasks and Code Mode for remote coding sessions. AI summary
Read more → -
Search Toolkit is a composable framework for building production search pipelines for AI applications, providing a shared interface for ingestion, retrieval, and evaluation to reduce engineering time spent on plumbing. It brings together separate tools into a single framework, allowing teams to focus on improving search quality instead of maintaining integrations. The open-source framework runs on any infrastructure, including cloud, on-premises, and edge. AI summary
Read more → -
Mistral has introduced physics AI, a class of AI models that learn from physics solver outputs and predict physical behavior directly from geometry and boundary conditions, enabling faster and more efficient engineering design and optimization. This technology can explore thousands of design variants in seconds, rather than hours, and can be integrated with existing AI models and tools to form a single stack spanning the engineering lifecycle. Physics AI has the potential to accelerate engineering acceleration in industries such as manufacturing, energy, and aerospace. AI summary
Read more → -
Mistral AI has made significant breakthroughs in physics AI research, including datasets for 3D wing simulations, a new framework for CFD simulations, and neural surrogates for aerodynamics and plasma turbulence. These advancements aim to accelerate product development and improve operational performance in industries such as aerospace, automotive, and energy. AI summary
Read more →