Llama 4 Scout

Llama 4 Scout is Meta’s compact yet powerful large language model, designed for high-performance tasks such as long-context reasoning, multi-document summarization, and code analysis. It employs a Mixture-of-Experts (MoE) architecture, activating 17 billion parameters per token from a pool of 16 experts, totaling 109 billion parameters. This design allows for efficient inference while maintaining high accuracy.