Meta Announces Its Most Advanced AI Model, Llama 4

These are advanced multimodal AI systems, capable of processing and integrating multiple types of data such as text, video, images and audio, allowing flexible conversion between these formats.

Llama 4 includes Meta's most advanced models.

Meta says the Llama 4 Scout and Llama 4 Maverick are its most advanced models, leading the way in multimodal AI, and will be released as open source. Meta also previewed the Llama 4 Behemoth, which is described as one of the smartest and most powerful LLMs in the world, serving as a “teacher” for future models.

The launch of Llama 4 was previously delayed because the model failed to meet Meta's technical standards, particularly on tasks involving reasoning and mathematics. The company was also concerned that Llama 4 was less effective than OpenAI's models at conducting natural-sounding speech conversations.

To solidify its position in the AI ​​field, Meta plans to invest $60 billion to $65 billion in 2025 to expand its AI infrastructure, including building large data centers and expanding its use of Nvidia AI chips. Meta aims to have more than 1.3 million graphics processors by the end of the year.

Meta's move comes as major tech companies ramp up their investments in AI following the success of ChatGPT, a game-changer in the tech industry that spurred interest in machine learning. Meta believes that offering open-source AI models like Llama 4 will attract talent and spur innovation in the developer community.

Additionally, Meta has authorized the use of the Llama AI model by U.S. national security agencies and defense contractors to support the development of security and defense-related applications. This decision comes after reports of Chinese researchers using an earlier version of Llama for military purposes.

Cao Phong (according to TC, The Guardian)

Source: https://www.congluan.vn/meta-cong-bo-mo-hinh-ai-tien-tien-nhat-mang-ten-llama-4-post341666.html