Paris-based OpenAI competitor Mistral AI closed its $415 million funding round

Mistral AI reached its Series A funding goal in France. Bloomberg reports that the company raised €385 million, or $415 million at today’s exchange rate, worth $2 billion. Today, Mistral AI unveils its commercial platform.

Google’s DeepMind and Meta alumni founded Mistral AI to create open-technology foundational models. It raised $112 million in seed funding less than six months ago.

Andreessen Horowitz (a16z) and Lightspeed Venture Partners are the lead investors in the AI company’s most recent funding round. Not only that, Salesforce, BNP Paribas, CMA-CGM, General Catalyst, Elad Gil, and Conviction are also investing.

Mistral AI co-founder and CEO Arthur Mensch said, “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 decentralised approach to technology.”

Mistral 7B was Mistral AI’s first model, released in September. Trained on a “small” dataset of 7 billion parameters, this large language model isn’t meant to compete with GPT-4 or Claude 2.

Instead of offering APIs, the company offered the Mistral 7B model as a free download for developers to run on their devices and servers.

An open-source Apache 2.0 license allows unlimited use and reproduction of the model with attribution. While anyone can run the model, it was developed privately with a proprietary dataset and weights.

Mistral AI’s influence on EU AI Act discussions was significant. In order to exempt foundational models, the French AI startup has advocated for regulation to apply only to use cases and companies developing products for end users.

EU lawmakers reached a political deal two days ago. Foundational model companies must share technical documentation and dataset summaries.

API access is required to access Mistral AI’s best model.
Foundational models will still generate revenue for the company. So Mistral AI is beta-launching its developer platform today. Companies can pay for Mistral AI models via APIs on this platform.

Both the Mistral 7B (“Mistral-tiny”) and Mixtral 8x7B (“Mistral-small”) models will be available to developers. This model processes input tokens using “a router network” to select the best parameters to answer.

This method increases model parameters while controlling cost and latency by using only a fraction of the total set of parameters per token. In detail, Mixtral has 45B parameters but uses 12B per token. The company wrote in a blog post that it processes input and outputs at the same speed and cost as a 12B model.

The Apache 2.0 license makes Mixtral 8x7B free to download. The Mistral developer platform offers Mistral-medium, a third model. No download link is available for Mistral AI’s superior model, which is only available through the paid API platform.