Llama 4 Scout: 10M Token Context in an Open-Weight Model

Announced by Meta on 5 April 2025, Llama 4 Scout is a natively multimodal open-weight model with a 10-million-token context window - the largest of any widely released model. It uses a mixture-of-experts architecture with 17 billion active parameters, 109 billion total, across 16 experts, and is downloadable from llama.com and Hugging Face.

Correction (2026-08-12): an earlier version of this item was dated July 2026 and described the release as new. Meta's announcement is dated 5 April 2025. Meta's announcement does not name a licence, so no licence claim is made here.