French startup Mistral AI has successfully concluded its highly anticipated Series A funding round, securing €385 million (approximately $415 million at today’s exchange rate). According to Bloomberg, this places the company’s valuation at roughly $2 billion. Concurrently, Mistral AI is introducing its commercial platform today.
It’s worth noting that just under six months ago, Mistral AI secured a $112 million seed round with the aim of establishing a European rival to OpenAI. Founded by former Google DeepMind and Meta professionals, Mistral AI focuses on developing foundational models with an open technology approach.
The most recent funding round is spearheaded by Andreessen Horowitz (a16z), with Lightspeed Venture Partners reinvesting in the AI firm. An array of other notable investors are also participating in the round, including Salesforce, BNP Paribas, CMA-CGM, General Catalyst, Elad Gil, and Conviction.
Mistral AI’s co-founder and CEO, Arthur Mensch, stated, “Since the creation of Mistral AI in May, we have been pursuing a clear trajectory: that of creating a European champion with a global vocation in generative artificial intelligence, based on an open, responsible, and decentralized approach to technology.”
In September, Mistral AI unveiled its inaugural model, Mistral 7B. This substantial language model is not designed to directly compete with GPT-4 or Claude 2, as it was trained on a “small” dataset of approximately 7 billion tokens as parameters. Instead of offering access to the Mistral 7B model via APIs, the company made it available as a free download for developers to utilize on their devices and servers. The model was released under the Apache 2.0 license, allowing unrestricted use and reproduction with proper attribution, even though it was developed using proprietary and undisclosed data.
Mistral AI also played a pivotal role in shaping discussions surrounding the EU’s AI Act. The French AI startup advocated for a complete exemption for foundational models, contending that regulation should apply to use cases and companies developing products directly used by end users. Recently, EU lawmakers reached a political agreement, imposing transparency requirements on companies working with foundational models, necessitating the sharing of technical documentation and dataset summaries.
While Mistral AI aims to profit from its foundational models, it is introducing its developer platform in beta today. This platform enables other companies to pay for access to Mistral AI’s models via APIs. In addition to the Mistral 7B model (“Mistral-tiny”), developers can now access the new Mixtral 8x7B model (“Mistral-small”). This model employs a “router network” to process input tokens and select the most suitable group of parameters for generating a response. This technique enhances the model’s parameter count while maintaining cost efficiency and low latency. Specifically, Mixtral has a total of 45 billion parameters but uses only 12 billion parameters per token, ensuring it processes input and generates output at the same speed and cost as a 12 billion parameter model. Mixtral 8x7B is also released under the Apache 2.0 license and is available for free download.
Lastly, Mistral-medium, a third model, is accessible exclusively through Mistral’s paid API platform, with no downloadable option. This model is purportedly superior in performance to Mistral AI’s other models.