- Meta’s Llama 4 AI models aim to put the U.S. back on top in the AI race, according to advisor David Sacks.
- Llama 4 Scout and Maverick outperform global competitors using a unique Mixture of Experts design and multimodal capabilities.
- Open-source innovation is now America’s counter to rising Chinese AI power, with Meta’s AI models integrated into WhatsApp and Instagram.
Meta’s newest Open Source AI models, Llama 4 Scout and Llama 4 Maverick, have created a buzz in the tech industry. David Sacks—a US venture capitalist and AI advisor—expressed his excitement regarding these models over the weekend on X, describing the products as a ‘victory for America’ in the global AI competition. “For the U.S. to win the AI race, we have to win in open source too, and Llama 4 puts us back in the lead,” he wrote.
Meta’s Llama 4 Models Set New AI Benchmark
Meta claims that Llama 4 Scout and Maverick are their most advanced models, releasing them on the 6th of April. These multifunctional models are already integrated into popular applications such as WhatsApp and Instagram, transforming the accessibility of advanced AI.
At the core of Llama 4’s innovation is multimodal AI, which works with text, images, audio, and videos simultaneously. Both models also utilize the Mixture of Experts (MoE) architecture: a system in which specialized AI units, referred to as “experts,” collaboratively work together. Llama 4 Scout comes with 17 billion active parameters alongside 16 experts; Maverick, however, expands that to 128 experts, enhancing reasoning and coding capabilities.
Meta states that Llama 4 Scout outperforms Gemma 3, Gemini 2.0 Flash-Lite, and Mistral 3.1, while Maverick competes with DeepSeek v3 from China in outstanding performance, despite using much fewer parameters. This release comes during savage competition globally, most notably with China’s DeepSeek, which burst onto the AI scene in late 2024 with its astonishingly effective model trained for only $6 million compared to OpenAI’s $100 million ChatGPT-4.
DeepSeek’s success sent shockwaves through the U.S. markets and spurred a fresh sense of urgency, as U.S. tech giants scramble to reclaim dominance. Sacks describing Llama 4 as America’s “secret weapon” underlines the notion touted that open-source AI might actually be the way to get ahead of other challengers in the world.
Meta further announced that Llama 4 Behemoth is “one of the most intelligent large language models ever conceived”— which may solidify the U.S. win the race on the next level of AI development — is currently in training.