Llama 4 is Meta's April 2025 generation of downloadable models, which use a mixture-of-experts design and understand both text and images.
In plain words
Llama 4 launched with two models. Scout has 109 billion parameters in total but uses only 17 billion for each token, spread over 16 experts, and can take a context of up to 10 million tokens. Maverick has 400 billion in total, also 17 billion active, with 128 experts and a 1 million token context. Both handle text and images in one model, using a design Meta calls early fusion. They officially support 12 languages.
A quick example
Imagine feeding Scout a large codebase and its documentation in one go and asking where a bug might be. Its very long context window is what makes a request that size possible.
Where you'll meet it
- Self-hosted assistants that read images and long documents.
- Commercial and research projects under Meta's licence.
Related
Sources
- DocsLlama-4-Scout-17B-16E-Instruct model card, Meta (Hugging Face) · read 29 Sept 2026
- Docsllama-models, Meta (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.