Mark Zuckerberg says that Meta’s Llama models have hit 1B downloads
In a brief message Tuesday morning on Threads, Meta CEO Mark Zuckerberg said the company’s “open” AI model family, Llama, hit 1 billion downloads. That’s up from 650 million downloads as of early December 2024 — a 153% increase over a roughly-three-month period.
Llama, which powers Meta’s AI assistant, Meta AI, across the tech giant’s various platforms, including Facebook, Instagram, and WhatsApp, is a part of Meta’s bid to foster a wide-ranging AI product ecosystem. The company makes the models and tools required to fine-tune and customize them available for free under a proprietary license.
Some developers and companies have taken issue with the Llama license terms, which are somewhat restrictive. Yet Llama has achieved widespread success in spite of this. Companies including Spotify use Llama models in production today.
That’s not to suggest that Meta hasn’t faced setbacks. Llama is at the center of an AI copyright lawsuit that accuses Meta of training a number of Llama models on copyrighted ebooks without authorization. In another challenge to Meta’s Llama ambitions, several EU countries have forced the company to postpone — and in some cases cancel altogether — its model launch plans over data privacy concerns. And Llama’s performance has been leapfrogged by models like Chinese AI lab DeepSeek’s R1.
Meta is said to have scrambled to set up war rooms to decipher how DeepSeek lowered the cost of running and deploying models, so that it could apply those learnings to Llama’s own development. And Meta recently said that it would spend as much as $80 billion on projects related to AI this year, including AI hires and the construction of new AI data centers.
Meta plans to launch several Llama models over the next few months, including “reasoning” models along the lines of OpenAI’s o3-mini and models with natively multimodal capabilities. Zuckerberg has also hinted at “agentic” capabilities, suggesting that future Llama models will be able to take certain actions autonomously.
Some of these may be announced at LlamaCon, Meta’s first generative AI developer conference, which is scheduled to take place on April 29.
“I think this very well could be the year when Llama and open source become the most advanced and widely used AI models,” Zuckerberg said during Meta’s Q4 2024 earnings call in January. “[O]ur goal for [Llama this year] is to lead.”
Kyle Wiggers is TechCrunch’s AI Editor. His writing has appeared in VentureBeat and Digital Trends, as well as a range of gadget blogs including Android Police, Android Authority, Droid-Life, and XDA-Developers. He lives in Manhattan with his partner, a music therapist.
