Meta is preparing to release several versions of Llama 4 throughout 2025, with a focus on reasoning capabilities and voice interaction. Meta CEO Mark Zuckerberg says Llama models have been downloaded more than 650 million times and developers have created more than 85,000 variations on the Hugging Face platform, including an Nvidia adaptation called Nemotron.
Meta has released several major updates to Llama throughout 2024. This included Llama 3.1, followed by Llama 3.2, which added the first multimodal capabilities and specialized models for mobile devices. The company then released Llama-3.3-70B, which matched the performance of the larger 3.1-405B while using fewer resources. Llama’s popularity has recently skyrocketed, with twice as many licenses issued compared to six months earlier. This growth has been supported by partnerships with major tech companies, including AWS, AMD, Microsoft Azure, Databricks, Dell, Google Cloud, Groq, NVIDIA, IBM watsonx, Oracle Cloud, and ScaleAI.
In 2025, Meta plans to accelerate Llama’s development with multiple releases of Llama 4. According to Zuckerberg, training Llama 4 could require nearly ten times the computing power used for Llama 3.
While the company is looking to improve the model across the board, it’s particularly interested in advanced reasoning and voice interactions — capabilities that are becoming increasingly important in the AI community. The company is already testing business-focused AI agents that can handle customer conversations, provide support, and process transactions. Meta believes these same advances could make AI assistants more capable for everyday users, helping them tackle more complex tasks on their own.
Due to regulatory uncertainty, Meta announced in the summer of 2024 that Llama 4 would not initially be available to companies in Europe, despite its open nature. Meta sees voice as the future of AI interaction. As language models become more conversational, the company expects users to shift from typing to interacting with AI assistants. Meta took the first step in this direction last fall by adding voice features to Meta AI, with major updates planned for early 2025.
The numbers show that Meta’s approach is working. Even without access in the European Union, Meta AI has attracted 600 million monthly users, making it one of the most widely used AI assistants in the world. By comparison, OpenAI reports that ChatGPT reaches around 300 million users each week.
In addition to text and speech, Meta is also making inroads into video AI. In October, the company unveiled Meta Movie Gen, a set of research models designed to generate and edit videos using generative AI.
