David Sacks Claims Meta’s Open-Source Llama 4 Gives the U.S. an Edge in the AI Race

Meta has recently solidified the United States’ position in the artificial intelligence (AI) sector by unveiling two cutting-edge models under the Llama 4 brand—Llama 4 Scout and Llama 4 Maverick. This move has been hailed by David Sacks, the U.S. AI and crypto czar, as essential for ensuring the United States retains its leadership in the increasingly competitive global AI marketplace. In a post on X, Sacks emphasized that for the U.S. to maintain its lead, success in the open-source realm is crucial, with Llama 4 setting the country back on the right path. These fourth-generation models, announced on April 6, 2023, underscore a pivotal moment in the AI race, especially considering the emergence of other formidable competitors.

The AI landscape has become significantly more complex with the introduction of DeepSeek, a Chinese AI startup that launched its first model in December 2024. January 2025 saw DeepSeek unveil a chatbot touted as a contender against industry giants like OpenAI’s ChatGPT. Not only did DeepSeek R1 quickly climb the app store charts, but the startup also challenged prevailing assumptions that substantial financial investments and high-end chips are prerequisites for success in AI development. Remarkably, DeepSeek reported an expenditure of approximately $6 million in training its model, while OpenAI, for comparison, was estimated to have spent around $100 million on ChatGPT-4’s development. This disparity caught the attention of notable venture capitalist Marc Andreessen, who likened the implications of DeepSeek’s launch to an “AI’s Sputnik moment,” signaling a critical juncture for American companies.

In the wake of DeepSeek’s advancements, U.S. tech firms have prioritized efforts to regain the initiative in AI innovation, underscoring Sacks’ assertion that Llama 4 has become a game changer. By positioning itself as a leader with its advanced models, Meta claims that Llama 4 Scout and Maverick are the most sophisticated models in their class, particularly in multimodal AI capabilities. This classification indicates that the models can simultaneously process various types of data, such as text, images, audio, and video. This versatility enables Llama 4 to understand intricate scenarios and offer nuanced, comprehensive responses, enhancing its utility and performance across different applications.

The Llama 4 models are pioneering in that they are built using a Mixture of Experts (MoE) architecture, where multiple expert models come together to enhance the efficiency of the larger AI system. Specifically, Llama 4 Scout features 17 billion active parameters with the backing of 16 specialized experts, while Llama 4 Maverick, although having the same number of parameters, is equipped with 128 experts. The design of these models allows efficient problem-solving, as each expert focuses on specific aspects of the broader issue. Furthermore, Llama 4 Scout has demonstrated superior performance over competitive models like Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1, while Llama 4 Maverick has achieved results rivaling those of DeepSeek v3 in tasks involving reasoning and coding—despite being less resource-intensive in terms of parameters.

In addition to the capabilities of Scout and Maverick, Meta’s Llama 4 models exhibit strong political leanings in their responses, comparable to existing systems like Grok. This characteristic raises intriguing questions about bias and the alignment of AI with user values and expectations. Alongside these models, Meta announced Llama 4 Behemoth, an anticipated language model currently undergoing training, which promises to be one of the most advanced large language models (LLMs) in existence. This continued progression from Meta, which first launched its initial Llama model in February 2023, showcases the rapid development and deployment of transformative AI technologies.

In summary, Meta’s launch of the Llama 4 models marks a critical advancement in the U.S. AI landscape, especially amid rising competition from international players like DeepSeek. The developments underscore the importance of open-source innovations and the necessity for American firms to stay competitive in the race for AI leadership. As these advanced models enhance their functionality and versatility, they represent a significant step forward in the evolution of AI systems and provide a robust foundation for future advancements. The implications of these models extend beyond technology, affecting the dynamics of the AI race globally, while simultaneously challenging industry norms regarding investment and capability disparities within AI development. The ongoing evolution of Llama 4 and its subsequent iterations could very well shape the future of artificial intelligence, making it a pivotal subject for stakeholders across the industry.