Meta's Llama 4 Models Lead U.S. in AI Race with 17 Billion Parameters

Meta's recent launch of its fourth-generation open-source Llama 4 models, Llama 4 Scout and Llama 4 Maverick, has positioned the U.S. as a leader in the AI race, according to David Sacks, the U.S. AI and crypto czar. Sacks emphasized the importance of open-source AI in maintaining the U.S.'s competitive edge in the global AI landscape. He stated that Llama 4 puts the U.S. back in the lead in the AI race, highlighting the significance of open-source models in this technological competition.

Meta's announcement on April 6 marked a significant milestone in the AI race. The Llama 4 models are described as the company's "most advanced models yet" and are noted for their superior multimodality capabilities. These models are currently available for download and use on Meta applications such as WhatsApp and Instagram, making them accessible to a wide range of users.

Multimodal AI systems, like the Llama 4 models, are capable of processing various types of data simultaneously, including text, image, audio, and video. This capability enables the AI to comprehend complex scenarios and generate comprehensive responses, making them highly versatile and effective in various applications.

Llama 4 Scout and Llama 4 Maverick are the first open-source Meta AI models built using a mixture of experts (MoE) architecture. In an MoE, multiple smaller models or specialized experts collaborate to make the larger AI model work. This means that experts focus on solving the parts of the problem they are designed to handle, enhancing the overall efficiency and effectiveness of the AI system.

Llama 4 Scout has 17 billion active parameters and 16 experts, while Llama 4 Maverick has the same number of parameters but is designed with 128 experts. Llama 4 Scout can fit in a single NVIDIA H100 GPU, whereas Llama 4 Maverick requires an H100 host. Meta claims that Llama 4 Scout outperforms other models like Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1 across a broad range of widely reported benchmarks. Llama 4 Maverick, on the other hand, provides results comparable to DeepSeek v3 on reasoning and coding despite having less than half the active parameters. Meta also asserts that Llama 4 Maverick beats GPT-4o and Gemini 2.0 Flash across a number of benchmarks.

Meta also unveiled Llama 4 Behemoth, which is still in training, as one of the “world’s smartest” large language models (LLMs). This model is expected to further enhance Meta's position in the AI race, showcasing the company's commitment to innovation and technological advancement.

Meta launched its first Llama model in February 2023, marking the beginning of its journey in the AI race. The company's continuous development and improvement of its AI models have positioned it as a key player in the global AI landscape. The launch of Llama 4 models is a testament to Meta's dedication to pushing the boundaries of AI technology and maintaining its competitive edge in the industry.