Llama 4 Scout

Overview

Meta's 109B parameter MoE model featuring an unprecedented 10M token context window. Llama 4 Scout is optimized for processing extremely long documents and codebases while maintaining strong general capability.

Why it matters

The 10M token context window is not a typo — Scout can process roughly 30 average-length novels or an entire enterprise codebase in a single call. This is a qualitatively different capability that enables use cases previously impossible: full-repository code understanding, entire-book analysis, or processing months of conversation history. For teams building code intelligence tools, legal document analysis, or long-horizon agentic systems, Scout eliminates the need for complex chunking and retrieval pipelines. The trade-off is that inference at these context lengths requires significant compute.