Mistral AI, a France-based startup, has announced a new large-scale language model (LLM) called Mistral Large that it claims can compete with some of the top AI systems on the market.
Mistral AI said Mistral Large outperformed most leading LLMs, except for OpenAI's recently released GPT-4, in language understanding tests, and also performed well in math and coding assessments.
Co-founder and chief scientist Guillaume Lampre said Mistral Large represents a big step forward compared to previous Mistral models. The company also launched a chatbot interface called Le Chat to allow users to interact with the system, similar to ChatGPT.
The proprietary model is fluent in English, French, Spanish, German and Italian, with a vocabulary of over 20,000 words. While Mistral's first model was open source, Mistral Large's code remains closed, as are systems from OpenAI and other companies.
Mistral AI raised about $500 million in funding late last year from backers including Nvidia and Andreessen Horowitz, and recently partnered with Microsoft to provide access to Mistral Large through its Azure cloud service.
Microsoft's €15 million investment in Mistral AI will likely come under scrutiny from European Union regulators, who are already analyzing Microsoft's relationship with OpenAI, the developer of market-leading models such as GPT-3 and GPT-4. The European Commission said on Tuesday it would review Microsoft's contract with Mistral, which could lead to a formal investigation that could jeopardize the partnership.
Microsoft has focused most of its AI efforts on OpenAI, investing about $13 billion in the California company, and those ties are currently under investigation in both the EU and the UK over potential anti-competitive concerns.
Pricing for the Mistral Large model starts at $8 per million input tokens and $24 per million output tokens. Mistral AI and Microsoft have also partnered on AI research, and the system will leverage Azure's computing infrastructure for training and deployment needs.
While Mistral Large has yet to be fully evaluated in third-party rankings, the company's previous Mistral Medium ranked sixth out of more than 60 language models. With its latest release, Mistral AI seems positioned to challenge major players in the increasingly competitive AI field.
(Photo by Joshua Golde on Unsplash)
reference: Stability AI Previews Stable Diffusion 3 Text-to-Image Conversion Model
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California and London, a comprehensive event taking place alongside other major events like BlockX, Digital Transformation Week and Cyber Security & Cloud Expo.
Find out about upcoming enterprise technology events and webinars hosted by TechForge here.
