Llama 4 Scout is Meta's open-weight multimodal model in the Llama 4 family, built as a mixture-of-experts architecture with 109 billion total parameters and 17 billion active per token, distributed under the Llama 4 Community License Agreement. It is positioned as a multilingual vision-language model that can ingest several images and respond to queries about them, making it well suited to workloads that combine visual and textual information rather than text-only reasoning. Its very large context window is intended for use cases such as multi-document analysis, large codebase reasoning, and personalized tasks where a single prompt needs to span far more material than typical chat models handle.

On publicly reported benchmarks, Llama 4 Scout reaches 51.8% on GPQA Diamond, indicating solid graduate-level question answering relative to its size class, and the broader family framing positions it as a state-of-the-art open model for its active-parameter count. Practitioners evaluating it for production work should weigh the long context and open weights against the fact that it is a research-oriented release still evolving through community feedback, so its strongest fit is for teams that want a self-hostable, multimodal model with expansive context for document-heavy, image-aware, or code-reasoning applications rather than a drop-in replacement for the largest closed frontier systems.