Meta took a significant step in AI development by launching the first models from its Llama 4 family — Llama 4 Scout and Llama 4 Maverick. These advanced language models are now accessible through Meta’s platforms like WhatsApp, Messenger, Instagram Direct, and Meta.ai. They also mark Meta’s initial open-source offerings under the Llama 4 umbrella, with two more models expected in the near future.
In a video shared on Instagram, Meta CEO Mark Zuckerberg described Llama 4 as a breakthrough for both Meta AI and the open-source community. He emphasized the company’s goal to make powerful AI tools—across all sizes—open and accessible.
Llama 4 Scout, one of the two released models, is a compact yet highly capable multimodal system. It features a 10 million token context window and is powered by 17 billion parameters using a Mixture of Experts (MoE) architecture with 16 experts. Designed to operate on a single Nvidia H100 GPU, Scout excels in performance and efficiency, reportedly surpassing models like GPT-4o, Gemini Flash 2, and DeepSeek V3 in benchmark comparisons.
The Llama 4 series doesn’t stop here. Meta is planning to release Llama 4 Reasoning next month. Meanwhile, Llama 4 Behemoth, currently in training, aims to become Meta’s largest AI model yet with an astounding 2 trillion parameters.
Both Llama 4 Scout and Maverick can be downloaded freely through Meta and Hugging Face, although certain usage restrictions might apply due to licensing conditions.
This latest release reinforces Meta’s ambition to push the boundaries of AI while making cutting-edge tools widely available for developers and researchers around the world.
