March 19th.Meta The company's CEO, Mark.ZuckerbergAnnounced on Threads, Meta's family of "open" AI models Llama has reached 1 billion downloads, up from 650 million downloads in early December 2024, an increase of approximately 53% in just about three months.
The Llama model is the core technology underpinning Meta AI, Meta's artificial intelligence assistant.Widely used across Meta's platformsThe Llama model and its associated fine-tuning and customization tools are available under a proprietary license to developers and businesses, including Facebook, Instagram, and WhatsApp. As a key component of Meta's multi-year commitment to building a broad ecosystem of AI products, the Llama model and its associated fine-tuning and customization tools are available free of charge to developers and businesses under a proprietary license.
Nonetheless, Llama's license terms have been questioned by some developers and companies due to certain commercial restrictions. However, since its launch in 2023, Llama has been widely successful. CurrentlySeveral high-profile companies, including Spotify, AT&T and DoorDash, are already using the Llama model in their production environments.
However, Meta has encountered some challenges in developing Llama, which is currently facing an AI copyright lawsuit alleging unauthorized use of copyrighted e-books to train multiple models. In addition, data privacy concerns have forced Meta to delay or even cancel Llama's local release in some EU countries. Meanwhile, in terms of performance, Llama is also facing competitive pressure from other models, such as the R1 model developed by Chinese AI lab DeepSeek, which has already outperformed Llama in some areas.
In the face of these challenges, Meta is said to have set up a "war room" to apply DeepSeek's advances to Llama's own development. In addition, Meta recently announced that it will spend up to $80 billion (note: 578.428 billion yuan at current exchange rates) on AI-related projects this year.
Looking ahead, Meta plans to release a number of Llama models in the coming months, including a "reasoning" model similar to OpenAI's o3-mini, as well as a model with native multimodal capabilities. Zuckerberg also hinted at the introduction of "agents" in Llama, meaning that some of the models will be able to act autonomously. During Meta's fourth-quarter earnings call in January 2024, Zuckerberg said, "I think this is likely to be the year that Llama and open-source models become the most advanced and widely used AI models. Our goal this year is to make Llama the leader."
More updates and future plans for Llama are expected to be revealed at LlamaCon, Meta's first generative AI developer conference on April 29th.
