Llama 4

Meta's open-weight flagship model family using Mixture-of-Experts (MoE) architecture. The Llama 4 lineup ranges from the lightweight Scout (17B active parameters, 109B total) to the massive Behemoth (2T total parameters). Scout supports a groundbreaking 10-million-token context window. As open-weight models, the Llama 4 family can be self-hosted, fine-tuned, and deployed without vendor lock-in, making them foundational infrastructure for the open AI ecosystem.