France's Mistral AI drew €600 million ($640 million, £510 million) in its latest funding round, bringing its valuation to €5.8 billion ($6.2 billion, £4.9 billion).

The LLM development firm is one of the bigger players in the model training industry. OpenAI with its $80 billion to $90 billion valuation is by far the biggest figure in the field, but at a valuation of $6.2 billion, Mistral AI is relatively close to its smaller peers, ahead of Cohere which was recently valued at $5 billion, somewhat behind Anthropic at $15 billion to $18.4 billion.

The French startup has grown its valuation substantially since the start of 2024, when it was estimated to be worth $2 billion, just a third of its current valuation.

Among Mistral AI's latest investors are tech firms like IBM, Nvidia, and Cisco, as well as Silicon Valley venture capital company Andreessen Horowitz. Nvidia and Cisco are also investors in Cohere, which raised $450 million just a week ago.

Microsoft is another key partner of Mistral AI, having invested $16.3 million into the French concern to install its large language models (LLMs) to Azure, including both open and closed-source models. That investment dwarfs the billions Microsoft has pumped into OpenAI. However, OpenAI's GPT AI models are all quite large with tens of billions of parameters, while Mistral AI offers both a larger LLMs as well as a small one that has just seven billion parameters.

Despite the investment being relatively small, it hasn't stopped regulators from being skeptical. The European Commission (EC) announced its probe in February, with the UK's Competition and Markets Authority (CMA) following in April. The primary concern has been that these investments could constitute an acquisition or merger, allowing companies like Microsoft and Amazon (Anthropic's main backer) to control the market.

Luckily for Mistral AI and Microsoft, the CMA already decided to drop its investigation last month, which may bode well in respect to the EC's separate examination.

Away from the money stuff, you can find Mistral's models on Hugging Face to download, deploy, and perform inference using various tools and frameworks, such as the quite nice mistral.rs. The firm's models aren't that bad in our experience, relatively speaking. ®

https://www.theregister.com//2024/06/12/mistral_ai_funding/