The artificial intelligence startup Mistral AI, based in Paris, revealed today that it has raised €600 million ($640 million) in fresh funding, headed by General Catalyst, at a valuation of $6 billion.
Following revelations last month that the business was aiming for $600 million at a roughly $5 billion value, the funds are from a Series B investment. The announcement comes after a16z and Lightspeed Ventures led the company’s $2 billion Series A fundraising round six months prior.
Current investors Andreessen Horowitz, Lightspeed, BNP Paribas, and Bpifrance participated in the round. Notable companies including Nvidia Corp., Samsung Venture Investment Corp., Salesforce Ventures LLC, and others were among the other backers.
The leadership experience of Mistral AI’s founders—who were also researchers at other businesses doing AI work—has drawn attention to the company. Arthur Mensch is from IBM Corp.’s DeepMind AI research center, while Timothée Lacroix and Guillaume Lample both came up through the ranks of Meta Platforms Inc.’s Paris AI center.
Despite being a relatively recent addition to the AI industry, having been established in April 2023, Mistral AI is renowned for its potent multilingual AI. French, English, German, Spanish, and Italian are among the five languages that its most recent model, the Mistral Large, speaks with ease. According to the business, Mistral Large only lags 10% behind OpenAI’s GPT-4 in reasoning benchmarks. Mistral offers two additional models, Small and Medium, to address varying workloads.
Although the firm faces competition from very large competitors like Meta Platforms Inc. and OpenAI, which have their own strategically positioned massive language models, Mistral Chief Executive Arthur Mensch says that the company’s current investor focus is a very clear signal.
“We were told when we started … that this is a market that is never going to be disrupted,” Mensch told the Financial Times. “We showed that this wasn’t the case and we effectively disrupted the OpenAI business model.”
In order to train and implement the company’s AI models, Mistral’s engineers now have access to Azure’s supercomputing cloud architecture thanks to a distribution partnership agreement that the two companies signed in February. Microsoft made a simultaneous investment of €15 million ($16.3 million) in the venture, which will be converted into equity at the conclusion of this funding round.
At the end of May, the business also unveiled Codestral, a potent open-source coding LLM that can comprehend more than 80 programming languages. Despite being a lightweight LLM with 22 billion parameters, Codestral can compete with some of the most potent LLMs available on the market, such as Meta Platform Inc.’s Llama 3 70B, which has three times the parameters. Additionally, it can process up to 32,000 tokens in prompts—more than twice as much as Llama 3 70B can handle.