Mistral AI Announces Partnership with Google Cloud and Releases New Open-Source LLM
-
Paris-based AI startup Mistral AI has announced a strategic partnership with Google Cloud, marking a significant breakthrough. Under the non-exclusive agreement, Mistral AI will leverage Google Cloud's AI-optimized infrastructure to distribute its open-source LLMs and optimized proprietary language models.
Isabelle Fraine, General Manager of Google Cloud France, stated: "In just a few months, Mistral AI has become a leader in generative AI LLM development based on open-source principles." Fraine added that leveraging Google Cloud's AI-optimized infrastructure will enable Mistral AI to transform its already impressive achievements into trustworthy, accessible, and enterprise-ready solutions.
This collaboration follows Mistral AI's integration of its 7B model into Google Cloud's Vertex AI Model Garden. Mistral-7B is the company's foundational model built using customized training, tuning, and data processing methods. Despite having fewer parameters, this optimized model allows for easy deployment on any organization's existing cloud due to its compact size. Under the agreement, the French AI company will use Google Cloud's AI-optimized infrastructure to further test, build, and scale its LLMs while benefiting from Google Cloud's security and privacy standards.
Now, businesses of all sizes can test, refine, and launch AI-based products and services while maintaining full control over their data with security and privacy as priorities, leveraging Mistral AI's open-source models and those from other providers. Arthur Mensch, CEO and co-founder of Mistral AI, said: "Google Cloud's flexible hardware and software stack will help us progress along our trajectory of building larger, more complex open-source AI models spanning more languages, making these models more accessible to organizations worldwide." He added: "Google Cloud's open-source support, responsible AI principles, robust and reliable infrastructure capabilities, and commitment to privacy and security align perfectly with our mission to develop publicly available AI models."
Meanwhile, Mistral AI's new hybrid expert model will be launched later on Google Cloud Marketplace. This strategic partnership comes just days after Mistral AI raised $415 million in a Series A funding round led by A16z, valuing the company at $2 billion.
Additionally, the company quietly released its latest LLM, sharing it via Torrent links. According to the company, Mixtral-8x7B is the most powerful open-source model with a permissive license and offers the best cost/performance trade-off, matching or outperforming competitors (including ChatGPT's GPT3.5 and Meta's Llama2 series) in most standard benchmarks.
The model's features include: processing 32k token contexts, supporting English, French, Italian, German, and Spanish, excelling in code generation, and being fine-tunable into an instructional model scoring 8.3 on MT-Bench.
Mistral AI was founded by Timothée Lacroix, Guillaume Lample, and Arthur Mensch, focusing on generative artificial intelligence. The company aims to develop new generative AI models for enterprises, combining scientific excellence, an open-source approach, and a vision of social responsibility in technology.