Back to glossary

Term

Llama 4 Scout

Llama 4 Scout (April 2025) is Meta's smaller Mixture-of-Experts variant in the Llama 4 lineup — 17B active out of 109B parameters, 10M-token context, multimodal.

Llama 4 Scout — explained in more detail

Llama 4 Scout was released on April 5, 2025 as part of the Llama 4 family — alongside the larger Llama 4 Maverick. Scout is one of Meta’s first models with a Mixture-of-Experts architecture: 109 billion total parameters, of which 17 billion are active per token, distributed across 16 experts. This combines the inference efficiency of a 17B model with knowledge depth closer to larger dense networks.

Specifications

  • Context window: 10 million tokens — the largest available context at launch, with good accuracy on Needle-in-a-Haystack benchmarks.
  • Multimodal: text and image input, early-fusion architecture.
  • License: Llama 4 Community License — commercial use with restrictions above 700M monthly active users.
  • Model name: meta-llama/Llama-4-Scout-17B-16E.

Context

Scout positions itself as the efficient choice for long contexts — RAG across large codebases, document analysis, long-form reasoning. Compared to Llama 4 Maverick (400B total / 128 experts), Scout is significantly cheaper to run but less capable on pure reasoning. Competitors: Gemini 3.1 Flash-Lite (1M context, closed), Qwen 3.5.

Entdecke mehr

Themenuebersicht