What Is Mistral AI?

Mistral AI is an artificial intelligence (AI) company based in Paris, France, that builds open-weight and commercial foundation models for enterprise and developer use. Arthur Mensch, a former Google DeepMind researcher, founded Mistral AI in spring 2023 (April, per Mistral AI) with former Meta researchers Guillaume Lample and Timothée Lacroix.

Mistral AI offers its models through Vibe, the assistant named Le Chat until May 28, 2026, and through its developer Application Programming Interface (API). Mistral AI also licenses private model deployments and operates dedicated European compute infrastructure for sovereign data processing.

As of October 2026, Mistral AI remains an independent private corporation valued at more than €21 billion following its Series D financing round. Mistral AI shares do not trade on any public stock exchange.

What Mistral AI Is

Mistral AI releases some models as open weights and keeps others API-only.

Mistral AI's About page says its co-founders set out to build "a European leader" in AI that pairs new research with openness, transparency, cost efficiency and responsibility. Mistral AI says it builds tailored AI systems with organizations in finance, manufacturing, defense, energy and the public sector.

TechCrunch reports that Mistral AI sends forward-deployed engineers to help governments and large companies adopt and tailor its AI. In its Regional Endpoints announcement, Mistral AI calls itself "the only European AI lab to offer both: choice of processing region and a committed, SLA-backed service level."

• Founded: Spring 2023 (April 2023 per Mistral AI; first employee hired June 5, 2023).
• Headquarters: Paris, France.
• Founders: Arthur Mensch, Guillaume Lample, and Timothée Lacroix.
• Chief Executive Officer (CEO): Arthur Mensch.
• Corporate ownership: Privately held independent corporation.
• Flagship model: Mistral Large 4 (announced October 6, 2026).
• Consumer application: Vibe (originally launched as Le Chat).
• Reported valuation: More than €21 billion post-money (Series D, September 2026).

Founders and Executive Leadership

Mistral AI's About page lists all three co-founders in executive roles as of October 2026.

Chief Executive Officer (CEO) Arthur Mensch previously worked as a researcher at Google DeepMind. Chief Science Officer (CSO) Guillaume Lample and Chief Technology Officer (CTO) Timothée Lacroix previously conducted machine learning research at Meta. In June 2023, Tech.eu described the three founders as "long-time school mates."

According to TechCrunch, Mistral AI took its name from a wind. Charles Gorintin and Jean-Charles Samuelian-Werve serve as co-founding advisers, and Samuelian-Werve also sits on Mistral AI's board. TechCrunch lists newer executives Johan Bergqvist (Chief Financial Officer), Brian Hall (Chief Marketing Officer) and Kamal Brar (Senior Vice President, Partners and Alliances). Reuters names Pierre Stock as Mistral AI's vice president of science.

Model Lineup and Software Licenses

Mistral AI's model list mixes open-weight models, tagged "Open" or with a license name, with API-only models tagged "Premier" in the docs.

Software licensing terms differ between documentation and commercial guidelines. As shown on the Mistral AI documentation portal, several models carry standard Apache 2.0 tags. The Mistral AI pricing FAQ ties the Apache 2.0 license to "research/individual use" and says commercial deployments "require a Mistral license with separate terms for derivatives and production use."

Mistral AI has retired earlier model lines over time. Devstral 2, Magistral Medium 1.2, and Magistral Small 1.2 were retired on July 31, 2026. The docs point Devstral 2 and Magistral Medium users to Mistral Medium 3.5, and Magistral Small users to Mistral Small 4. Pixtral Large was retired on May 31, 2026, while legacy releases Mixtral 8x7B and Mistral 7B were retired on March 30, 2025. As noted in the Mistral Small 4 announcement, Small 4 unifies earlier specialized reasoning, coding, and multimodal features into a single model.

• Generalist models: Mistral Large 4 (v26.10, preview October 6, 2026, tagged "Open", weights due by the end of October 2026); Mistral Medium 3.5 (v26.04, Modified MIT license, dense 128B parameters, 256k context window); Mistral Small 4 (v26.03, Apache 2.0 license, 119B total parameters, 6B active parameters); Mistral Large 3 (v25.12, Apache 2.0 license); Ministral 3 14B, 8B, and 3B (v25.12, Apache 2.0 license).
• Document processing models: OCR 4.1 (v4.1, tagged Premier, proprietary API service); OCR 3 (also listed as active).
• Audio and speech models: Voxtral TTS (v26.03, Creative Commons Attribution-NonCommercial 4.0 or CC BY-NC 4.0 license); Voxtral Mini Transcribe 2 (v26.02, tagged Premier); Voxtral Mini Transcribe Realtime (v26.02, Apache 2.0 license); Voxtral Small (v25.07, Apache 2.0 license).
• Code completion models: Codestral (v25.08, tagged Premier, proprietary API service).
• Embedding models: Codestral Embed (v25.05, tagged Premier); Mistral Embed (v23.12, tagged Premier).
• Moderation and safety models: Shieldstral 1.0 (Apache 2.0 license, compact multimodal classifier); Mistral Moderation 2 (v26.03, tagged Premier, 128k context window).

Consumer Software and Developer Platforms

Mistral AI delivers software products directly to end users through its Vibe assistant and to software engineers through Mistral AI Studio.

On May 28, 2026, Mistral AI announced that Le Chat became Vibe. Mistral AI said every conversation, setting, and plan carried over from Le Chat. Vibe has a Work Mode for long-running, multi-step tasks and a Code Mode that launches remote coding agents from the web. Vibe runs on the web, on iOS and Android, as a Visual Studio Code extension, and in the Vibe Command Line Interface (CLI).

Developers reach the API through Mistral AI Studio, which launched on October 24, 2025, and offers a free Experiment plan for testing. Studio provides observability metrics, agent runtimes, and an AI model registry. Batch processing provides a 50 percent pricing discount, while prompt caching reduces input token expenses by up to 90 percent. Mistral AI's OCR 4 announcement priced OCR 4 at $4 per 1,000 pages through the API and $2 per 1,000 pages with Batch.

• Free plan: Basic assistant features on web and mobile with limited messages, web queries, coding sessions, and over 100 connectors.
• Pro plan: $14.99 per month excluding taxes, offering all-day coding in the CLI, IDE, or on web and $15 in monthly API credits.
• Student plan: Verified students receive the Pro plan for $5.99 per month.
• Team plan: $24.99 per user per month excluding taxes, adding up to 30 gigabytes of storage per user, domain verification, and 200 daily flash answers compared to 150 on Pro.
• Enterprise plan: Custom pricing providing private deployments, custom models, audit logging, Security Assertion Markup Language Single Sign-On (SAML SSO), custom Service Level Agreements (SLAs), and dedicated support.
• Specialized tools: Forge trains models on a company's own data across pre-training, post-training and reinforcement learning. Mistral Code, a coding assistant launched in June 2025, is labeled "later Vibe" on Mistral AI's About page.

European Infrastructure and Sovereign Computing

Mistral AI builds dedicated data centers and physical computing clusters across Europe to secure independent computing capacity.

Mistral AI announced Mistral Compute on June 11, 2025, offering bare-metal servers, managed Platform as a Service (PaaS) and tens of thousands of Graphics Processing Units (GPUs). Reuters reported in March 2026 that Mistral AI raised $830 million in bank debt, its first, to build a data center at Bruyères-le-Châtel, near Paris, with 13,800 NVIDIA chips. Reuters said Mistral AI unveiled a second facility, in Sweden, in February 2026.

In September 2026, Euronews reported that Mistral AI is investing €4 billion into European data infrastructure. "Long term, the plan is to fully rely on capacity that we are building ourselves," Arthur Mensch told Euronews. Reuters reported in March 2026 that Mistral AI aims to secure 200 megawatts of capacity across Europe by the end of 2027. Mistral AI's Munich hub announcement sets a larger target of one gigawatt of European compute by 2030. Regional Endpoints introduced in August 2026 allow clients to restrict inference to Europe or the United States (US).

Venture Capital Financing and Valuation History

Mistral AI is funded through private venture rounds and a bank loan.

In June 2023, Mistral AI raised a €105 million seed round led by Lightspeed Venture Partners at a €240 million valuation. The Register reported the round as $113 million at a $259 million valuation, while TechCrunch reported $260 million. Backers included JCDecaux Holding, Exor Ventures, Sofina, Bpifrance, Eric Schmidt, Xavier Niel, and Rodolphe Saadé. On December 11, 2023, Mistral AI completed a €385 million Series A round at a €2 billion valuation led by Andreessen Horowitz (a16z). In February 2024, Microsoft invested €15 million ($16.3 million) in convertible debt.

In June 2024, Latham and Watkins reported a €595 million Series B at a €5.8 billion post-money valuation, co-led by General Catalyst and DST Global. CNBC put the round at €600 million ($645 million) with General Catalyst as lead, and the valuation at about $6.2 billion. In September 2025, Mistral AI secured €1.7 billion in Series C funding at an €11.7 billion valuation led by ASML. On September 8, 2026, Mistral AI announced €3 billion in Series D capital at a valuation exceeding €21 billion, confirmed by legal counsel Orrick. VentureBeat said Reuters put the Series D valuation at roughly $24 billion.

Strategic Partnerships and Commercial Contracts

Mistral AI's partners include Microsoft, NVIDIA, Samsung Electronics and France's Ministry of the Armed Forces.

Microsoft became Mistral AI's initial distribution partner in February 2024, hosting models on Microsoft Azure alongside la Plateforme. In March 2026, Mistral AI partnered with NVIDIA as a founding member of the NVIDIA Nemotron Coalition. Samsung Electronics led the September 2026 Series D round. On January 8, 2026, France's Ministry of the Armed Forces awarded Mistral AI a framework agreement overseen by the ministry's Agency for Defense Artificial Intelligence (AMIAD). No financial terms were disclosed.

In Saudi Arabia, Mistral AI signed a sovereign-AI deal with HUMAIN that Mistral AI values in the hundreds of millions of euros. Mistral AI models also run on Cloudera's hybrid data platform, including in air-gapped sites. Mozilla uses Mistral AI models for Firefox Smart Window, a beta feature, with zero data retention. Named enterprise clients include Airbus, BNP Paribas, AXA, Stellantis, IBM, SAP, CMA CGM, Snowflake, Orange, France Travail, Cisco, TotalEnergies, and Ericsson.

Ownership Structure and Public Stock Availability

Mistral AI is an unlisted private corporation, meaning retail investors cannot purchase Mistral AI stock on public exchanges.

Mistral AI's investors include Samsung Electronics, ASML, Lightspeed Venture Partners and Andreessen Horowitz. At Davos in January 2025, Arthur Mensch said Mistral AI is "not for sale," TechCrunch reported. "Of course, [an IPO is] the plan," Mensch said when asked about an initial public offering (IPO). In March 2025, Mensch told Fortune that Mistral AI intended to remain independent and was not pursuing an immediate IPO.

As of October 2026, Mistral AI has announced no public listing filings, bank underwriting mandates, or formal IPO prospectuses. ASML and Samsung Electronics shares trade on public markets, but owning them gives you no ownership of Mistral AI.

The Mistral Large 4 Launch in October 2026

On October 6, 2026, Mistral AI launched a public preview of Mistral Large 4, which it calls its "largest and most capable model to date."

According to the Mistral Large 4 announcement, Mistral Large 4 has 1 trillion total parameters and 52 billion active parameters. Mistral AI trained Large 4 from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European data centers. TNW reported that the training run used about 10 megawatts. Mistral AI announced that weights will be released by the end of October 2026, with reports from TNW and VentureBeat citing October 27, 2026.

The Mistral Large 4 documentation lists $1.36 per million input tokens and $4.18 per million output tokens. A launch sale cuts those rates to $0.68 and $2.09. At an event in Abu Dhabi, Mensch said Large 4 is "above the Chinese models on certain aspects, including cyber," Reuters reported.