📋About Mistral AI
Updated September 12, 2026Mistral AI is a French AI company founded in 2023 by Arthur Mensch, Guillaume Lample, and Timothée Lacroix — former researchers at Google DeepMind and Meta. The company has raised more than $6 billion. The September 2025 Series C took 1.7 billion euros, led by ASML; the September 2026 Series D took 3 billion euros, led by Samsung Electronics with co-leads Scaleup Europe Fund, managed by EQT, and existing investor PSG Equity — a round Mistral describes as the largest ever raised by a European technology company. With $400 million-plus in annualized recurring revenue, Mistral is Europe's most valuable AI startup. Its two lead investors are a semiconductor-equipment maker and a chipmaker rather than software investors, which is the sovereignty pitch expressed as a cap table.
Mistral develops a range of foundation models: Mistral Large 3 (675 billion MoE flagship, 256K context, top open-source coding model), Devstral 2 (72.2% SWE-bench Verified coding specialist), Mistral Small 4 (119 billion MoE, Apache 2.0, March 2026), Mistral Medium 3.5 (128 billion dense, 77.6% on SWE-Bench Verified, self-hosts on as few as 4 GPUs at $1.50 / $7.50 per million input/output tokens), and the Ministral family for edge deployment. In March 2026, Mistral released Voxtral TTS — its first voice product, an open-source 4 billion parameter text-to-speech model supporting 9 languages that runs on consumer hardware. In June 2026 Mistral shipped OCR 4, a document-intelligence model that extracts structured text, tables, and layout from scanned files across 170 languages and runs self-hosted in a single container — positioned as the ingestion layer for enterprise search and retrieval pipelines, and benchmark-leading on document extraction.
Mistral's consumer and enterprise platform is Vibe, the unified Work + Code agent that replaced Le Chat in May 2026. Vibe consolidates chat, productivity, and coding agents under a single license — Work mode plans multi-step tasks across Google Workspace, Outlook, SharePoint, Slack, and GitHub, while Code mode runs coding agents in isolated cloud sandboxes launched from the web, the Vibe CLI, the Vibe VS Code extension, or third-party apps. Pricing starts free, with Pro at $14.99 per user per month and Team at $24.99 per user per month — a unified offering priced below ChatGPT Plus or Claude Pro alone.
Mistral has positioned itself as Europe's answer to US AI labs, combining frontier capabilities with a strong open-source commitment and EU data sovereignty. The company runs its own compute: an existing 40 megawatt Paris data center, a new 10 megawatt inference facility under construction in Les Ulis south of Paris (opening in the third quarter of 2026), and a 1.2 billion euro Swedish facility planned for sovereign EU inference. The Koyeb acquisition added AI cloud infrastructure, and Mistral operates an 18,000 Grace Blackwell GPU compute cluster for training and inference. Defense contracts with Helsing and the French military demonstrate government trust in European AI sovereignty.
Mistral has also expanded into industrial engineering through the acquisition of Emmi AI, a Linz-based Austrian startup specializing in physics-based simulation models. The deal launched Mistral Physics AI — a class of data-driven foundation models that learn from physics-solver outputs and predict the behavior of physical systems in seconds on a single GPU, replacing CAE simulations that traditionally take hours to weeks. Disclosed production customers include Airbus (commercial aircraft, helicopters, defense, and space), ASML (semiconductor parts design and surrogate modeling), BMW Group (the carmaker's Large Industry Model initiative for multimodal reasoning on engineering data), Safran, and Siemens Energy. The combined pitch reframes Mistral's roadmap beyond chat and coding into real-time simulations and digital twins for high-stakes manufacturing customers — layering Emmi's physics-AI on top of the Large 3 and Medium 3.5 language stack to compress R&D cycles in domains like jet-engine design, semiconductor process modeling, and battery chemistry. The acquisition also establishes Linz as Mistral's first deep technical bench outside its Paris, London, Amsterdam, Munich, San Francisco, and Singapore footprint.
In August 2026 Mistral extended its sovereignty pitch beyond Europe for the first time, announcing a collaboration with HUMAIN — the Saudi artificial-intelligence company majority-owned by the Public Investment Fund — worth hundreds of millions of euros. It spans infrastructure, model development and deployment across the Middle East: Mistral will run on HUMAIN data centers in the region, the two intend to build frontier models that perform strongly in Arabic, and the initial focus is cybersecurity and voice, with a joint go-to-market aimed at financial services, manufacturing, telecommunications and the public sector. The significance is that the argument Mistral built for European buyers — keep data, compute and operations inside your own jurisdiction — turns out to travel to any state willing to fund its own stack, which widens Mistral's addressable market well beyond the EU while tying it to a sovereign backer.
🛠️Products & Tools (11)
Open-source terminal coding agent powered by Devstral 2 with custom subagents (TOML config), slash-command skills, MCP support, and on-premise deployment.
Mistral's 3 billion parameter open-weights safety classifier. Judges text and images against moderation policies written in plain language at inference time, under Apache 2.0.
Mistral's most capable model. 675B total MoE (41B active), 256K context, multimodal. Top open-source coding model on LMArena. Roughly half the cost per token of GPT-5.1. Modified MIT license.
Mistral's coding specialist. 123B params scoring 72.2% on SWE-bench Verified, claims 7x more cost-efficient than Claude Sonnet. Devstral Small 2 (24B, Apache 2.0) scores 68%. Includes Vibe CLI.
Efficient MoE model — 119 billion total parameters, 128 experts, approximately 6.5 billion active per token, 256,000 token context, Apache 2.0 license
Structure-aware document-intelligence OCR model — extracts text, tables, and layout across 170 languages, self-hostable in a single container, at $4 per 1,000 pages. Leads the OlmOCRBench benchmark.
Mistral's open Apache-2.0 model for formal mathematical proof in Lean 4 — a mixture-of-experts specialist that aces the miniF2F benchmark and finds real software bugs.
Data-driven foundation models for industrial engineering — learning from physics-solver outputs to predict the behavior of physical systems in seconds on a single GPU. Launch partners include Airbus, ASML, Safran, and Siemens Energy.
Mistral AI's consumer AI platform. Powered by Mistral Large 3 (675B MoE). Mobile apps, Deep Research mode, Projects, code interpreter, Flux Ultra image gen. EU data sovereignty. Pro at €14.99/month.
Mistral AI's first embodied-AI model — an 8 billion parameter vision-language-action model that steers robots through unfamiliar environments using only a single RGB camera, with no LiDAR, depth sensors, or pre-built maps. Reports 76.6% success on unseen Room-to-Room continuous-environment navigation benchmarks; hardware-agnostic across wheeled, legged, and flying robots. Announced July 2026.
Mistral AI's open-source text-to-speech model — 4 billion parameters, 9 languages, runs on consumer hardware, $0.016 per 1,000 characters via API
Keep track of the companies you’re watching
- Save the companies you want to follow
- Get ⚡ alerts when your saved companies change
- Every product they ship, cross-linked to 900+ AI tool profiles
- Today’s top AI Stories — the day’s most important AI news, free
Swipe for Recommended for you and My AI Tools
Your AI Hub — sample data. See desktop view example



