Mistral AI and its founders, including Cédric O, are not fans of the European AI Act, ratified this week by Parliament. However, this does not prevent the French startup from continuing its development and its alliances.
After Microsoft and Snowflake, the Frenchman formalizes an agreement with another giant of American Data & AI platforms: Databricks. The publisher takes a stake in the capital of Mistral AI as part of its series A.
Nearly 1,000 Mistral LLM users in 2024
Databricks is taking advantage of this above all to expand its offering by integrating the young company’s models into its Data Intelligence Platform offering. This includes open source models, now natively integrated into Databricks.
Generative AI models ship in Databricks Marketplace. For platform customers, this means “interacting with these models in the Mosaic AI Playground, using them as landing points for optimized models through Mosaic AI Model Serving, and personalizing them using their own data through adaptation .”
For the American publisher, it is simply a matter of responding to a request from its users. Databricks specifies that since the start of the year, nearly 1,000 companies have used Mistral AI models, generating millions of inferences.
The firm therefore intends to simplify these uses thanks to “ready-to-use integrations”, accelerating the use of the startup’s models for generative AI applications. And this, “without compromising the security, data privacy and governance that are at the heart of the Databricks platform.”
Portability, openness and accessibility of IAGen
For Arthur Mensch, co-founder and CEO of the young unicorn, this partnership is part of the company’s strategy aimed at democratizing artificial intelligence. Through “this strategic alliance with Databricks”, Mistral AI declares to reaffirm its commitment “in favor of the portability, openness and accessibility of generative artificial intelligence for all.”
Unlike the agreements with Microsoft and Snowflake, the one concluded with Databricks does not cover the use of Mistral Large, the latest model launched by the startup and which, unlike its predecessors, is not offered as open source.
Only Mistral 7B and Mixtral 8x7B are mentioned in the press release from the American platform – which has increased fundraising since its creation, but without making an IPO. The last operation dates back to September 2023 for an amount of 500 million dollars and a valuation of 43 billion.
Databricks, model portal and native integration
With Mistral AI, as with other foundation models available on Databricks, access to LLMs can be achieved through APIs. The publisher offers its customers the opportunity to consume and customize these different models from its services.
“Whether it’s starting with a direct comparison of pre-trained models or consuming models by paying tokens, there are several options to get started quickly,” highlights Databricks.
When will we now see partnerships with French platforms to offer sovereign generative AI solutions, another of the initial promises of Mistral AI and its leaders?