Meta's open source artificial intelligence model family, Llama, has recently ushered in an important milestone, with more than one billion downloads. The news was announced by company CEO Mark Zuckerberg on social media Threads. Compared with the 650 million downloads in December 2024, Llama's downloads surged by 153% in just three months.

Originally launched for researchers, Llama has gradually attracted the attention of developers and AI enthusiasts around the world over time. Ahmed Al Dahler, vice president of Generative Artificial Intelligence at Meta, shared this achievement on LinkedIn, saying that over the past two years, Llama has grown from a research project to the most widely used open source AI model.
Al Dahler was proud of the team's efforts, mentioning that Meta's generative AI organization celebrated its second anniversary on February 27. He said he felt extremely excited and humbled to see the hard work the team has put in for developers, startups and businesses around the world. He believes that the download volume of Llama represents trust in open source artificial intelligence and also reflects everyone's belief in building the future together.
To further promote Llama, Meta plans to release multiple Llama models in the coming months, including an "inference" model similar to OpenAI's o3-mini, as well as a multimodal version. Zuckerberg also hinted that models with "agent" capabilities may be launched, meaning that some models may have the ability to operate autonomously.
Last October, Zuckerberg confirmed on the company's third-quarter earnings call that Llama4 will be released in early 2025. He expects the smaller Llama4 model to be ready first and believes that this will be a major advance, involving new modalities, capabilities, stronger inference capabilities, and faster processing speeds.
The Llama3 model has become an important turning point in the industry, but Zuckerberg expressed greater enthusiasm for the development of the Llama4. He mentioned that the training scale of Llama4 exceeds 100,000 H100 chips, far exceeding the scale currently known by other companies.
Llama has powered Meta's Facebook, Instagram, and WhatsApp, and plays a key role in the company's long-term AI strategy.
Key points:
Llama downloads exceeded 1 billion, an increase of 153% in three months.
Meta plans to release a variety of new models, including inference and multimodal capabilities.
Zuckerberg is confident in the development of the Llama4 and is expected to bring stronger performance.