Mistral AI

European foundation model lab producing the Mistral and Codestral model families with competitive coding performance and a strong focus on sovereignty and on-premise deployment for EU enterprises. Launched Agents API with MCP support in May 2025 alongside OpenAI and Anthropic; raised approximately $640M at $6B valuation in 2024; Codestral models compete directly with Codex and Code Llama.

Mistral is the leading European alternative to US/Chinese model providers, and Codestral specifically targets code generation with an emphasis on data-residency compliance for EU enterprises.

mistral.ai Foundation Models

Franceprivate

At a glance

Total raised

$4.0B

last round undisclosed

Headcount

201-500

approximate · public range

Headquarters

Paris

France

Founded

2023

3 years active

Status

Private

Foundation Models

Subsector

Foundation Models

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Founding

50% off

/month

94 of 100 spots left

Full

$10

/month

Available once all 100 Founding Member spots are claimed.

Get full access

Already subscribed? Sign in →

Funding history

Total raised

$4.0B

Rounds tracked

Latest stage

Undisclosed

Date	Stage	Amount	Valuation	Lead investors
Mar 30, 2026	Undisclosed	$830M	—	—
Sep 7, 2025	Series C	$2B	$13.7B	ASML
Jun 11, 2024	Series B	$640M	$6B	General Catalyst
Feb 26, 2024	Convertible	$16M	—	Microsoft
Dec 11, 2023	Series A	$415M	$2B	Andreessen Horowitz
Jun 13, 2023	Seed	$113M	$260M	Lightspeed Venture Partners

Founder backgrounds

Arthur Mensch

Co-founder & CEO

Trained as a mathematician and machine learning researcher, Arthur Mensch graduated from École Polytechnique (2011), earned a Master's from Télécom Paris, and completed a PhD in machine learning at Université Paris-Saclay, followed by a postdoc at ENS Ulm. In late 2020, he joined DeepMind's Paris office, where he spent nearly three years working on large language models and contributed to an early version of Google's Gemini. He left in 2023 to co-found Mistral AI alongside Guillaume Lample and Timothée Lacroix, raising a record €105M seed round. The École Polytechnique community has described his public posture as "simple, direct, detached, but with a clear and uncompromising vision" — a tone he has carried into his advocacy for European AI sovereignty and open-weight models.

Guillaume Lample

Co-founder & Chief Scientist

Trained at École Polytechnique (X2011) and Carnegie Mellon University, with a PhD on unsupervised machine translation from Université Paris-Saclay/LIP6. At Facebook AI Research (FAIR) in Paris, Lample authored seminal papers on unsupervised machine translation using monolingual corpora only and co-created the XLM and XLM-RoBERTa cross-lingual language models. He was part of the core team that built Meta's Llama large language model before co-founding Mistral AI in April 2023 as Chief Scientist. His deep expertise in multilingual modeling and efficient training directly informs Mistral's focus on sovereign, performant models for European enterprises.

Timothée Lacroix

Co-founder & CTO

A graduate of École Normale Supérieure, Lacroix spent roughly eight years at Meta's FAIR lab in Paris, first as a PhD student and later as a Research Engineer, where he worked on link prediction and large-scale language-model optimization. He was a key architect of Meta's original open-source Llama models alongside co-founder Guillaume Lample — making the two of them central figures in the Llama talent exodus that later fed Mistral's early team. As CTO, Lacroix leads the company's research and technical direction, overseeing the Mistral and Codestral model families that compete on coding performance while emphasizing European sovereignty and on-premise deployment for enterprise clients.

AI bull / bear

Bull case

The EU regulatory environment is Mistral's structural moat. As GDPR enforcement intensifies and the EU AI Act phases in, enterprises face mounting pressure to keep model inference and fine-tuning within European data centers. Mistral is the only scaled foundation model lab that can offer on-premise deployment with full data residency, giving it a captive addressable market that OpenAI and Anthropic cannot easily serve. The $830M raised in March 2026 brings total funding to $4.0B, providing ample runway to close performance gaps without the distraction of chasing consumer adoption. Codestral's competitive coding benchmarks create a wedge into the developer tooling market, where EU banks, defense contractors, and government agencies represent billions in procurement spend that US providers are structurally disqualified from. The Agents API launch with MCP support in May 2025 placed Mistral in the same release cadence as OpenAI and Anthropic, signaling that the company can keep pace on developer experience even while navigating the Brussels regulatory stack. If the sovereignty premium proves durable—and every signal from EU industrial policy suggests it will—Mistral has a clear line to $20B+ enterprise value without needing to beat GPT on raw capability.

Bear case

Mistral has raised $4.0B but operates in a market where that figure barely registers. OpenAI, Anthropic, and Google are each spending multiples of Mistral's entire capital base annually on compute and talent. In foundation models, capability gaps compound: a narrow lead in coding benchmarks can evaporate in one training run, and the frontier labs are scaling inference-time compute at rates that make $830M rounds look like table stakes. The EU sovereignty thesis is real but thin. Large enterprises may mandate European hosting, but that requirement gets satisfied by Microsoft and AWS deploying US models into Frankfurt and Paris data centers—no Mistral needed. Meanwhile, Meta's Llama models are open-weight, free, and increasingly competitive on code generation, giving enterprises a compliance-friendly escape hatch with zero vendor lock-in. Mistral's Codestral faces direct pressure from Code Llama on the open side and Claude Code in the premium tier, squeezing it from both directions. The Agents API parity moment from May 2025 proves catching up is possible, but sustained parity against labs with 10x the compute budget requires flawless execution and a little luck—neither of which the market should price at a $6B valuation from 2024, let alone whatever mark the 2026 round implies.

Generated by deepseek-v4-pro · May 2, 2026Refreshes quarterly · see disclosures

In the news

Watch

Leadership

Arthur Mensch

Co-founder & CEO

in 𝕏

Guillaume Lample

Co-founder & Chief Scientist

in 𝕏

Timothée Lacroix

Co-founder & CTO

in 𝕏

Products

Mistral Large
Mistral Large is Mistral AI's flagship frontier-grade large language model, designed for complex enterprise reasoning, multilingual tasks, and high-throughput production deployments. It excels at nuanced instruction following, code generation, and long-context processing across dozens of languages. Deployable on-premise or via cloud APIs, it gives European enterprises a sovereign alternative to US-based models while matching or exceeding benchmark performance on key reasoning and coding evaluations.
Le Chat
Le Chat is Mistral AI's conversational AI assistant available on web, iOS, and Android. It offers multi-model support (Mistral Large, Small, and Next), web search with citations, document and image understanding, canvas-based ideation, and image generation via Flux Pro. Designed for both personal productivity and enterprise use, Le Chat Enterprise adds advanced customization, security controls, and on-premise deployment options for organizations requiring data sovereignty.
Codestral
Codestral is Mistral AI's open-weight generative AI model purpose-built for code generation across 80+ programming languages. At 22B parameters, it balances performance and low latency for tasks like code completion, test generation, and interactive coding assistance. Developers can integrate it into VS Code and JetBrains via Continue.dev and Tabnine, and it supports both instruction-style prompts and traditional code completion workflows for research and non-commercial use.
Mistral Small
Mistral Small is a compact, cost-efficient language model optimized for high-throughput, latency-sensitive applications where tier-1 reasoning is not required. It delivers strong performance on classification, summarization, structured extraction, and lightweight conversational tasks at a fraction of the computational cost of larger models. Ideal for enterprises deploying AI at scale on modest infrastructure or in edge environments, it can run on-premise or via Mistral's API.

Open roles

View all careers ↗

Mistral AI

At a glance

Subscribe to see the full profile

Funding history

Founder backgrounds

Arthur Mensch

Guillaume Lample

Timothée Lacroix

AI bull / bear

In the news

Cloud code: Conductor joins the rush toward remote coding agents

TanStack Npm Packages Compromised Inside The Mini Shai Hulud Supply Chain Attack

Mistral, Europe’s answer to OpenAI and Anthropic, pushes its coding agents to the cloud

Watch

Leadership

Products

Mistral Large

Le Chat

Codestral

Mistral Small

Open roles

Compare Mistral AI with