The tech giant Meta has launched the most advanced suite of large language models that support the entire Llama ecosystem.
Meta introduced Llama 4 Scout and Llama 4 Maverick, the first open-weight natively multimodal models with unprecedented context length support and its first built using a mixture-of-experts (MoE) architecture, that is, they are capable of processing and integrating text, video, images, and audio.
Meta is also previewing Llama 4 Behemoth, one of the smartest LLMs in the world and its most powerful yet to serve as a teacher for the new models.
Llama 4 Scout and Llama 4 Maverick are available for download on llama.com and Hugging Face so everyone can continue to build new experiences using our latest technology, Meta said.
“We’ll also make them available via our partners in the coming days. You can also try Meta AI with Llama 4 starting today in WhatsApp, Messenger, Instagram Direct, and on the Meta.AI website,” Meta added.
Llama 4 models are designed with native multimodality, incorporating early fusion to seamlessly integrate text and vision tokens into a unified model backbone. Early fusion is a major step forward, since it enables us to jointly pre-train the model with large amounts of unlabeled text, image, and video data.
Image Via: Meta AI@X | Cropped by GBN
Meta also improved the vision encoder in Llama 4. This is based on MetaCLIP but trained separately in conjunction with a frozen Llama model to better adapt the encoder to the LLM.
Key Features of Llama 4 Models:
Llama 4 Scout
Designed for efficiency and speed, ideal for developers and researchers with limited GPU resources
Utilizes a Mixture of Experts (MoE) architecture with 16 experts and 17B active parameters
Supports a 10 million token context window and runs efficiently on a single H100 GPU
Llama 4 Maverick
Flagship open-weight model for advanced reasoning, coding, and multimodal applications
Employs a Mixture of Experts architecture with 128 routed experts and 17B active parameters
Achieves high performance in benchmark tests, including ELO scores and LMSYS Chatbot Arena
Performance and Applications
Scout excels in efficiency-focused evaluations, such as ARC and MMLU Lite, and is suitable for applications like long-context memory chatbots and educational Q&A bots
Maverick outperforms leading models in knowledge-intensive tasks, code generation, and visual question answering, making it ideal for AI pair programming, enterprise-level document understanding, and educational tutoring systems.
