Paris-based Mistral AI, an artificial intelligence startup, announced on Wednesday that it has signed a partnership with Google Cloud.
Under this non-exclusive agreement, Mistral AI will distribute its open weights and optimised proprietary language models utilising Google Cloud’s AI-optimised infrastructure.
“In just a few months Mistral AI has already established itself as a leader in the development of generative AI LLMs built on open source principles,” says Isabelle Fraine, Managing Director, of Google Cloud in France.
“Leveraging Google Cloud’s AI-optimised infrastructure and technical expertise will allow Mistral AI to translate its already-impressive achievements into trusted, accessible, and enterprise-ready solutions that directly address the diverse and rapidly growing global demand for generative AI,” adds Fraine.
The agreement comes after Mistral AI integrated its 7B model with Google Cloud’s Vertex AI Model Garden.
Mistral-7B is the company’s foundational model that is based on customized training, tuning, and data processing methods. This optimised model allows for the compression of knowledge and deep reasoning capacities despite having a small number of parameters.
According to Mistral AI, the 7B model can already be easily deployed on any organisation’s existing cloud due to its compact size.
As part of the agreement, the French AI company will use Google Cloud’s AI-optimised infrastructure to further test, build, and scale up its LLMs, all while benefiting from Google Cloud’s security and privacy standards.
Now, businesses, regardless of their size, can test, perfect, and launch their AI-based products and services by utilizing Mistral AI’s open-source models and those of other providers.
They can do so securely and with privacy in mind while retaining complete control over their data claims the company.
“Google Cloud offers a flexible hardware and software stack that will help us continue on our trajectory towards building larger, more sophisticated open source AI models across more languages, as well as make those models easier to access for all types of organizations around the world,” says Arthur Mensch, CEO and co-founder of Mistral AI.
“Google Cloud’s open source support and responsible AI principles, extensive and reliable infrastructure capabilities, and commitments around privacy and security align strongly with our mission to develop openly available AI models,” adds Mensch.
In addition, Mistral AI’s brand new mixture-of-expert model will be made available in the Google Cloud Marketplace at a later stage.
Drops ChatGPT 3.5’s rival
The partnership comes a couple of days after raising $415M in a Series A funding round led by A16z and is now estimated to be worth $2B.Â
Additionally, the company has released its latest large language model (LLM), without explanation, in a nondescript Torrent link posted to X over the weekend without much fanfare.
On Monday, the company published a blog post that provided additional information about the program, Mixtral-8x7B, following its initial release.
According to the company’s claims, “it is the strongest open-weight model with a permissive license and the best model overall regarding cost/performance trade-offs.”
In particular, it matches or outperforms its competitors, including ChatGPT’s GPT3.5 and Meta’s Llama 2 family on most standard benchmarks.
Mixtral capabilities include:
- It handles a context of 32k tokens.
- It handles English, French, Italian, German and Spanish.
- It shows strong performance in code generation.
- It can be finetuned into an instruction-following model that achieves a score of 8.3 on the MT-Bench.
Mistral AI: Developing generative AI
Founded by Timothée Lacroix, Guillaume Lample, and Arthur Mensch, Mistral AI specialises in generative artificial intelligence.
The company aims to develop new models of generative artificial intelligence for companies, combining scientific excellence, an open-source approach, and a socially responsible vision of technology.
01
Job board for modern workforce: How Remote Talent helps jobseekers find truly remote, distributed work