Source: Llama 4 Community License and model page (Meta)
Meta's Llama 4 licence (700M MAU clause, Built with Llama, naming rules) and the Llama 4 Scout/Maverick model page.
Contents
- What: licence text (GitHub) and official model page
- Publisher: Meta
- Date: licence effective 2025-04-05; accessed 2026-11-14
- Original: Llama 4 License
Summary
The custom licence governing Llama 4 weights, plus Meta’s Llama 4 page describing Scout and Maverick.
Key claims
- Licensees whose products had more than 700M monthly active users before the release must request a licence that Meta may refuse.
- Redistribution requires shipping the licence and prominently displaying “Built with Llama”.
- Models trained or fine-tuned from Llama materials or outputs and then distributed must have names beginning with “Llama”.
- Use must follow the Llama 4 Acceptable Use Policy, incorporated by reference.
- Llama 4 Scout is natively multimodal, supports up to 10M tokens of context and runs on a single H100; Maverick is natively multimodal for image and text.
- Meta’s developer page still labels Llama 4 as the latest Llama and lists the newer Muse models separately; the
meta-llamaHugging Face org’s newest model repositories date from April 2025. - Model IDs encode the MoE shape: Scout 17B active with 16 experts, Maverick 17B active with 128 experts (~400B total per Hugging Face metadata).