The French startup Mistral AI has officially completed its much-anticipated Series A funding round. The company has raised 385 million euros or 415 million USD at today's exchange rate – according to Bloomberg, the company is valued at around 2 billion USD. Mistral AI is also launching its commercial platform today.
As a reminder: Mistral AI raised a 112 million USD seed round less than six months ago to establish a European competitor to OpenAI. Mistral AI was co-founded by graduates from Google's DeepMind and Meta and is working on foundational models with an open technology aspect.
Andreessen Horowitz (a16z) is leading the latest funding round, with Lightspeed Venture Partners investing in the AI company again. That's not all, as a long list of investors is also participating in the round, including Salesforce, BNP Paribas, CMA-CGM, General Catalyst, Elad Gil, and Conviction.
Since the founding of Mistral AI in May, we have been following a clear path: the creation of a European champion with a global focus in the field of generative artificial intelligence, based on an open, responsible, and decentralized technology approach,” said Mistral AI founder and CEO Arthur Mensch in a statement.
In September, Mistral AI released its first model called Mistral 7B. This large language model is not intended to compete directly with GPT-4 or Claude 2, as it was trained on a 'small' dataset of around 7 billion parameters.
Instead of enabling access to the Mistral 7B model via APIs, the company made it available as a free download so developers could run it on their devices and servers.
The model was released under the Apache 2.0 license, an open-source license that imposes no restrictions on use or reproduction beyond attribution. Although the model can be run by anyone, it was developed behind closed doors with a proprietary dataset and undisclosed weights.
Mistral AI also played an important role in shaping discussions around the EU AI Act. The French AI startup has advocated for a complete exemption for foundational models, stating that regulation should apply to use cases and companies working on products directly used by end-users.
EU lawmakers reached a political agreement just a few days ago. Companies working on foundational models will face some transparency requirements and must share technical documentation and summaries of the content contained in the datasets.
The best model from Mistral AI is now only accessible via an API.
The company continues to plan to make money with its foundational models. For this reason, Mistral AI is opening its developer platform in beta today. With this platform, other companies can use Mistral AI's models via APIs for a fee.
Besides the Mistral 7B model ('Mistral-tiny'), developers can access the new Mixtral 8x7B model ('Mistral-small'). This model uses a 'router network' to process input tokens and select the most suitable group of parameters to provide a response.
This technique increases the number of parameters of a model while controlling costs and latency, as the model only uses a fraction of the total parameter set per token. Specifically, Mixtral has a total of 45B parameters but only uses 12B parameters per token. Therefore, it processes inputs and generates outputs at the same speed and cost as a 12B model, the company wrote in a blog post.
Mixtral 8x7B was also released under the Apache 2.0 license and is available as a free download. A third model, Mistral-medium, is available on Mistral's developer platform. It allegedly offers better performance than the other models from Mistral AI and is only available through the paid API platform – no download link available.
