Mistral, French A.I. Start-Up, Is Valued at $2 Billion in Funding Round


Mistral AI, a Paris start-up founded seven months ago by researchers from Meta and Google, has raised 385 million euros, or about $415 million, in yet another sign of feverish interest in a new kind of artificial intelligence that drives online chatbots.

The deal values the 22-person company at about $2 billion, two people familiar with the deal said. Investors include the Silicon Valley venture capital firms Andreessen Horowitz and Lightspeed Venture Partners.

The start-up’s value has increased more than sevenfold in just six months. In the summer, it raised a seed funding round of €105 million (about $113 million) that valued the company at about $260 million.

Mistral builds technology that other businesses can use to deploy chatbots, search engines, online tutors and other A.I.-driven products. It is among a small group of companies — including the tech industry’s giants and a handful of start-ups — that are building A.I. that could rival technology under development at OpenAI, the San Francisco start-up that kicked off the A.I. boom last fall with the release of the ChatGPT chatbot.

Mistral is also among the companies that believe in sharing this technology as open-source software — computer code that can be freely copied, modified and reused — providing outsiders with everything they need to quickly build chatbots of their own. Rival companies like OpenAI and Google argue that the open-source approach is dangerous and that the raw technology could be used to spread disinformation and other harmful material.

Mistral’s fate has taken on considerable importance in France, where leaders like Bruno Le Maire, the finance minister, have pointed to the company as providing the nation a chance to challenge U.S. tech giants. Europe has not produced many meaningful tech companies dating back to the dot-com boom and sees artificial intelligence as a field where it can gain ground.

Investors are pumping money into other start-ups that believe in the open-source approach. Perplexity, founded last year by another group of top researchers, has raised a new $70 million funding round that values the company at $500 million, a person familiar with the deal said. Investors include IVP and Bessemer Venture Partners.

“We just believe A.I. should be open,” said Anjney Midha, a general partner with Andreessen Horowitz who led the new investment in Mistral. Most of the major technologies that drive modern computing are open source, he added, including computer operating systems, programming languages and databases.

Mistral was founded by Timothée Lacroix and Guillaume Lample, who previously worked as researchers in the Paris A.I. lab of Meta, the owner of Facebook and Instagram, and Arthur Mensch, who was a researcher at DeepMind, an A.I. lab that Google acquired in 2014 for $650 million.

Company employees like to joke that the first letters of their last names of the founders are “L.L.M.” — which is also short for large language model, the A.I. technology that the company is building.

Companies like OpenAI, Microsoft and Google are leading the A.I. race, after spending hundreds of billions of dollars on this kind of technology. By analyzing enormous amounts of digital text culled from the internet, a large language model can learn to generate text on its own. This means it can answer questions, write poetry and even generate computer code.

Companies like OpenAI and Google believe that this technology is so powerful, they can release it to the public only in the form of an online chatbot after spending months applying digital guardrails that prevent it from spewing disinformation, hate speech and other toxic material.

But many A.I. researchers, tech executives and venture capitalists believe the A.I. race will be won by companies that build the same technology and then give it away for free — without any guardrails.

Meta, the former home of two Mistral founders, has been at the forefront of companies encouraging this open-source approach. This year, the tech giant built a large language model called LLaMA and essentially gave it away as open-source software.

On Sunday, Mistral also released its latest technology as open-source software, saying it performs at a level on a par with Meta’s technology.

Widely sharing the underling code for A.I., Mr. Midha said, is the safest path because more people can review the technology, find its flaws and work to remove or mitigate them.

“Now a single engineering team can find every bug,” he said. “Large communities of people building cheaper, faster, better, safer software.”

Adam Satariano contributed reporting from London.