What is Mistral AI?
Paris-based frontier AI lab building open-weight and enterprise LLMs — Europe's most valuable AI startup at roughly $14B and reportedly raising at ~€20B.
- Category
- Foundation models / Generative AI
- Headquarters
- Paris, France
- Founded
- April 2023
- Employees
- ~1,000–1,220
- Total funding
- ~$4B (equity + debt)
- Valuation
- ~€11.7B / $14B (Sep 2025); ~€20B round in talks (Jun 2026)
What is Mistral AI?
Mistral AI is a Paris-based artificial-intelligence lab founded in 2023 by ex-Meta and Google DeepMind researchers that builds frontier large language models, releasing many of them as downloadable open-weight models under permissive licenses. It is Europe's most valuable AI startup, valued at roughly €11.7 billion (~$14 billion) after its September 2025 Series C, and is positioned as the leading non-US, 'sovereign' alternative to OpenAI and Anthropic.
Mistral's business spans three layers: open-weight and commercial models (Mistral Large 3, Mistral Small 4, Ministral, Codestral/Devstral, Pixtral, Voxtral) delivered through its La Plateforme API; an AI assistant and agent platform called Vibe (rebranded from Le Chat in June 2026); and developer and enterprise products like Studio, Forge and Mistral Compute for building, fine-tuning and self-hosting agents. The same models are downloadable from Hugging Face for teams that want to run them on-premises or at the edge — a key differentiator versus closed US labs. In 2026 the company also pushed into 'industrial' and physics-grounded AI, extending beyond pure text models.
The company has scaled revenue dramatically: annual recurring revenue crossed roughly $400 million in early 2026, up about 20x from ~$20 million a year earlier, and CEO Arthur Mensch has publicly targeted surpassing $1 billion in revenue by the end of 2026. On the demand side, Vibe (formerly Le Chat) reaches on the order of 45 million monthly active users, and Mistral reported roughly 450,000 total customers with about 1,031 high-value accounts as of mid-2025. Its 100+ enterprise clients include ASML, HSBC, Stellantis and Ericsson, alongside the European Space Agency and government deployments across France, Germany, Luxembourg, Greece, Estonia and Singapore.
Mistral's edge is efficiency, openness, and European data sovereignty: it ships frontier-class models at aggressive prices, lets customers run weights themselves, and pairs that with its own GPU data-center buildout near Paris (funded partly by an ~$830M debt raise). Around 60% of revenue comes from Europe, where regulated buyers are wary of routing sensitive data through US providers. That mix has made it the default 'EU AI champion' for sovereignty-conscious enterprises and the public sector.
What does Mistral AI offer?
Mistral offers a family of open-weight and commercial LLMs plus a platform of developer and enterprise products for building, fine-tuning and deploying AI — anchored by the Vibe assistant (formerly Le Chat) and the La Plateforme API.
- Mistral Large 3 (flagship LLM)· Models
- Mistral Small 4 (open-weight)· Models
- Ministral (edge models)· Models
- Codestral / Devstral (code models)· Models
- Pixtral (multimodal/vision)· Models
- Voxtral (speech/audio)· Models
- Vibe (AI assistant, formerly Le Chat)· Consumer
- Vibe for Code (coding agents)· Developer
- La Plateforme (model API)· Developer
- Studio (build & run agents)· Platform
- Forge (custom model training)· Platform
- Mistral Compute (GPU cloud)· Infrastructure
- On-prem & private deployments· Enterprise
How does Mistral AI make money?
Mistral makes money from usage-based API access (pay-per-token on La Plateforme), Vibe consumer/team subscriptions (formerly Le Chat), and enterprise contracts for private deployments, custom model development and on-premises licensing — the last being the largest revenue driver.
On the API, La Plateforme charges per million tokens, with prices that undercut frontier US rivals: Mistral Small 4 runs about $0.10 input / $0.30 output, Mistral Large 3 about $0.50 / $1.50, and Mistral Medium 3.5 about $1.50 / $7.50. Mistral Large 3 launched at roughly a 75% price cut versus the prior generation, reinforcing the 'frontier performance at aggressive pricing' positioning. This pay-per-token motion pulls open-weight developers who start free on Hugging Face into paid hosting once they need scale or reliability.
Vibe (formerly Le Chat) is sold in tiers — Free ($0, soft-capped around 25 messages/day), Pro at about $14.99/month, Team at about $24.99/seat/month, and custom Enterprise pricing for the Vibe Enterprise tier. This seat-based consumer and SMB motion, reaching roughly 45 million monthly active users, is a top-of-funnel layer beneath the much larger enterprise business and a source of recurring subscription revenue.
The heavy revenue comes from enterprise: private and on-premises deployments, fine-tuned custom models, and multi-year co-development contracts (Mistral has cited on the order of €350M/year in client contracts). Growth is driven by European 'sovereign AI' demand, open-weight adoption that funnels developers into paid hosting, and large public-sector and regulated-industry deals — with about 60% of revenue from Europe. The company is not yet profitable, reinvesting heavily into compute and talent ahead of its $1B revenue target.
Who leads Mistral AI?
Mistral was co-founded by three researchers from Meta and Google DeepMind: Arthur Mensch (CEO), Guillaume Lample (Chief Scientist) and Timothée Lacroix (CTO). In 2026 it added senior commercial leadership including CMO Brian Hall.
- Arthur MenschCo-founder & CEO2023–presentFormer research scientist at Google DeepMind; École Polytechnique alumnus and the public face of Europe's AI sovereignty push. Owns strategy, fundraising and government relations.
- Guillaume LampleCo-founder & Chief Scientist2023–presentFormer Meta AI (FAIR) researcher who led work on the LLaMA models; drives Mistral's model research agenda.
- Timothée LacroixCo-founder & CTO2023–presentFormer Meta AI engineer and a LLaMA co-author; owns Mistral's engineering and training infrastructure.
- Brian HallChief Marketing Officer2026–presentJoined in 2026 from senior product-marketing leadership roles at Microsoft, Amazon (AWS) and Google; signals Mistral's build-out of a US-style commercial GTM engine.
- Victor VuillardChief Information Security OfficerCurrentOwns security and compliance — central to Mistral's sovereign-AI positioning for regulated and public-sector buyers.
How do you contact Mistral AI's leadership?
Mistral's general contact is contact@mistral.ai (press via press@mistral.ai), with a contact form at mistral.ai/contact. Its dominant work-email pattern is first@mistral.ai (e.g. arthur@mistral.ai), though the CEO has also been listed as amensch@mistral.ai; the leadership addresses below follow the published format rather than being individually confirmed inboxes, so treat them as pattern-derived.
first@mistral.aiHow much funding has Mistral AI raised?
Mistral has raised roughly $4 billion in total (equity plus debt). Its latest closed valuation is about €11.7 billion (~$14 billion) from the September 2025 Series C led by ASML, and as of June 2026 it was reported to be raising about €3 billion more at a ~€20 billion valuation (a round still in early talks, not yet closed).
The company moved fast from a standing start. In June 2023 it raised a €105M (~$117M) seed at a ~€240M valuation led by Lightspeed Venture Partners, with Eric Schmidt, Xavier Niel and JCDecaux participating. In December 2023 came a €385M Series A led by Andreessen Horowitz at a ~$2B valuation, with Salesforce and BNP Paribas joining. Microsoft then added a ~$16M strategic investment in February 2024 alongside an Azure distribution deal.
In June 2024 Mistral closed a €600M (~$640M) Series B led by General Catalyst at a $6B valuation (about €468M equity / €132M debt), with a16z, Lightspeed, Nvidia and Bpifrance also joining. The September 2025 Series C raised €1.7B at a €11.7B post-money valuation, led by Dutch chip-equipment giant ASML — which took roughly an 11% stake for ~€1.3B to become the largest shareholder — with Nvidia, a16z, General Catalyst, DST and Bpifrance also participating. Mistral has had no down round to date; every round has stepped up the valuation.
In March 2026 Mistral added an ~$830M debt facility (from a syndicate of seven banks) to buy roughly 13,800 Nvidia GPUs for a new data center near Paris. As of June 2026, Bloomberg, TechCrunch and others reported Mistral was in early talks to raise about €3B ($3.5B) at a ~€20B valuation — nearly double the Series C mark — to fund its sovereign-AI and industrial-AI expansion, though that round had not yet closed and its terms could still change.
How did Mistral AI get here?
From founding to ~$14B valuation in roughly two years, Mistral scaled through rapid model releases and some of Europe's largest venture rounds — and is reportedly headed toward ~€20B in 2026.
- Apr–Jun 2023Founded in Paris, €105M seedArthur Mensch, Guillaume Lample and Timothée Lacroix found Mistral; raise a €105M seed at ~€240M led by Lightspeed.
- Sep–Dec 2023First models + €385M Series AShips open-weight Mistral 7B and Mixtral 8x7B; raises a €385M Series A led by a16z at ~$2B.
- Jun 2024$640M Series B at $6BGeneral Catalyst leads a €600M Series B; Le Chat and Mistral Large bring a consumer + enterprise stack.
- Sep 2025€1.7B Series C at €11.7BASML leads, taking ~11% to become largest shareholder; Mistral becomes Europe's most valuable AI startup.
- Mar 2026$830M debt for GPUs / data centerDebt facility from seven banks funds ~13,800 Nvidia GPUs for a new data center near Paris; ARR crosses ~$400M.
- Jun 2026Vibe launch + ~€20B round in talksRebrands Le Chat as Vibe and pushes into industrial AI; reported to be raising ~€3B at a ~€20B valuation, nearly doubling the Series C mark.
Who are Mistral AI's competitors?
Mistral competes with the major US foundation-model labs and other open-weight providers, differentiating on price, openness and European data sovereignty.
- OpenAIFar larger and better funded with the broadest model + product ecosystem (ChatGPT); closed-weight and US-hosted vs Mistral's open, European stance.
- AnthropicClaude leads on coding, long-context reasoning and safety; closed models vs Mistral's downloadable open weights.
- Google DeepMind (Gemini)Strong multimodal models with huge context windows and Google distribution; Mistral counters on openness and lower price.
- Meta (Llama)The other dominant open-weight family; free and massive-scale, but Mistral pairs open weights with a commercial platform and EU hosting.
- CohereEnterprise- and RAG-focused models with a sovereignty angle; smaller scale and less consumer presence than Mistral.
- DeepSeekChinese lab shipping very low-cost open-weight frontier models; competes hard on price but faces geopolitical/data-trust barriers in Europe.
Mistral AI — frequently asked questions
