Llama AI models have been downloaded over a billion times, Mark Zuckerberg boasts

The number of downloads of open-source artificial intelligence models Llama has reached 1 billion, Meta✴ CEO Mark Zuckerberg announced. As of early December 2024, this figure was 650 million, which corresponds to growth of more than 50% in a quarter.

Image source: Stefan Cosma / unsplash.com

The Llama family of models is the basis for Meta✴ AI, an AI assistant that is present on various platforms, including Facebook✴, Instagram✴, and WhatsApp — Meta✴ aims to build a broad system of AI products. The company provides both the models themselves and the tools for fine-tuning and customizing them for free under its own license. These models have some limitations for commercial use, but are actively used in various products, such as by AT&T and the streaming service Spotify.

However, it is not without its challenges: Meta✴ is currently trying to fight off a lawsuit in which it is accused of training AI on copyrighted materials; in the EU, the company was forced to delay or even cancel the release of some models due to concerns about data privacy. In addition, DeepSeek’s models turned out to be better than Llama in a number of ways. Meta✴ is trying to take advantage of DeepSeek’s experience in reducing the costs of training AI.

The American company plans to spend at least $65 billion on AI this year, and it wants to release several Llama models in the coming months, including reasoning ones. Meta✴ is also likely developing AI agents — systems that can independently perform complex sequences of actions. It may present some of these products at LlamaCon, Meta✴’s first generative AI developer conference, scheduled for April 29.