French Startup Mistral AI Secures €385 Million in Series A Funding to Enhance Generative AI Development
French startup Mistral AI has successfully concluded its highly anticipated Series A funding round, raising €385 million (approximately $415 million) based on current exchange rates. According to Bloomberg, this funding round values the company at around $2 billion. Additionally, Mistral AI is launching its commercial platform today.
Notably, Mistral AI had previously secured €112 million in a seed round less than six months ago, positioning itself as a European competitor to OpenAI. Co-founded by alumni from Google’s DeepMind and Meta, Mistral AI focuses on developing foundational AI models from an open technology perspective.
Leading the latest funding round is Andreessen Horowitz (a16z), joined by returning investor Lightspeed Venture Partners. A diverse group of other investors is also participating, including Salesforce, BNP Paribas, CMA-CGM, General Catalyst, Elad Gil, and Conviction.
"Since our inception in May, Mistral AI has remained committed to creating a European leader in generative artificial intelligence, driven by an open, responsible, and decentralized technological approach," stated Arthur Mensch, co-founder and CEO of Mistral AI.
In September, Mistral AI launched its first model, Mistral 7B. This large language model, while not directly competing with GPT-4 or Claude 2, was trained on a relatively smaller dataset consisting of around 7 billion parameters.
Instead of restricting access through APIs, Mistral AI freely released the Mistral 7B model for developers to download and deploy on their own devices and servers. Released under the Apache 2.0 license, this open-source framework imposes no usage restrictions beyond the requirement for attribution. Although the model is accessible to all, it was developed using a proprietary dataset kept confidential.
Mistral AI has also been influential in discussions surrounding the EU’s AI Act. The startup has advocated for a complete exemption for foundational models, arguing that regulations should focus on the applications and companies directly serving end users.
Just days ago, EU lawmakers reached a political agreement regarding transparency requirements for companies working with foundational models, which will include sharing technical documentation and summaries of their datasets.
Mistral AI's leading model is now available only through an API, with plans to monetize its foundational models. To support this, the company is launching its developer platform in beta today, allowing businesses to access Mistral AI's models through paid APIs.
In addition to the Mistral 7B model (also known as “Mistral-tiny”), developers will also be able to utilize the new Mixtral 8x7B model (“Mistral-small”). This innovative model employs a "router network" to effectively process input tokens and select the most appropriate set of parameters to generate responses.
"This technique enables the model to maintain a high parameter count while optimizing cost and latency; it only utilizes a fraction of the total parameters for each token. Specifically, Mixtral has 45 billion total parameters but employs 12 billion parameters per token, processing inputs and generating outputs at speeds and costs comparable to a 12 billion parameter model," the company explained in a recent blog post.
Like the Mistral 7B, the Mixtral 8x7B has also been released under the Apache 2.0 license and is available for free download. A third model, known as Mistral-medium, is obtainable through Mistral’s developer platform and is reported to outperform the other models, although it is only accessible via the paid API platform.
Ranking
