French startup Mistral AI has successfully concluded its Series A funding round, raising €385 million, or $415 million. This latest funding round values the company at approximately $2 billion. Mistral AI, which aims to be a European rival to OpenAI, received $112 million in seed funding just six months ago. The startup, co-founded by former employees of Google’s DeepMind and Meta, specializes in developing foundational models with an open technology approach.
Esteemed venture capital firm Andreessen Horowitz (a16z) is leading the funding round, with Lightspeed Venture Partners also investing once again in Mistral AI. In addition to these prominent investors, Salesforce, BNP Paribas, CMA-CGM, General Catalyst, Elad Gil, and Conviction are among the participants in this funding round.
Mistral AI has made significant strides in the field of artificial intelligence (AI). In September, the company unveiled its first model, Mistral 7B. Unlike direct competitors GPT-4 and Claude 2, Mistral 7B was trained on a smaller dataset consisting of approximately 7 billion parameters. In a departure from the industry norm, Mistral AI made this model available as a free download rather than via APIs. The company hopes that by allowing developers to run the model on their own devices and servers, it can foster innovation and broader access to AI technology.
The release of Mistral 7B underlines Mistral AI’s commitment to open-source principles. The model is distributed under the Apache 2.0 license, which permits unrestricted utilization and reproduction, provided proper attribution is given. While anyone can run the Mistral 7B model, it is important to note that the model’s development relied on proprietary datasets and undisclosed weights.
Mistral AI has been influential in shaping legislation regarding AI in the European Union. The startup has advocated for a complete exemption for foundational models, asserting that regulation should be focused on companies working on use cases and products directly impacting end users. Recently, EU lawmakers reached a political agreement on the AI Act, which includes some transparency requirements for companies developing foundational models. These requirements entail sharing technical documentation and summaries of the datasets used.
Despite the regulatory landscape, Mistral AI remains dedicated to monetizing its foundational models. To this end, the company is launching its developer platform in beta. By accessing Mistral AI’s models through an API, other companies will be able to utilize the technology for a fee. The Mistral developer platform will offer developers access to multiple models, including the newly released Mixtral 8x7B model. This model employs a router network to process input tokens and identify the most suitable group of parameters for generating an accurate response.
Mistral 8x7B boasts an impressive 45 billion total parameters, but it operates at the same speed and cost as a 12 billion parameter model. The Mixtral 8x7B model, like Mistral 7B, will be available as a free download under the Apache 2.0 license. However, Mistral AI’s most advanced model, Mistral-medium, which outperforms the company’s other models, will only be accessible through the paid API platform.
In summary, Mistral AI has secured significant funding for its Series A round, highlighting the growing interest and investment in the AI field. The company’s commitment to openness, responsibility, and decentralization has garnered attention from prominent investors. With its release of foundational models and the launch of its developer platform, Mistral AI aims to democratize access to AI technology while positioning itself as a major player in the global AI landscape.