Free Llama 4 Model Downloads for Self-Hosting
Meta currently makes Llama 4 Scout and Llama 4 Maverick weights available for download. The Llama 4 license grants a royalty-free limited license for permitted use, modification, reproduction, and distribution, so there is no model-license fee for ordinary eligible users. This is an open-weight/self-hosting offer, not a promise of free hosted inference: hardware, electricity, or third-party cloud/API charges are separate. Meta's official repository notes that full-precision Llama 4 inference is hardware-intensive; Scout can be reduced to a single 80GB GPU using Int4 quantization.
How to Claim this Deal
- 1
- 2Read and accept the applicable Llama license and submit the access request.
- 3After approval, Meta sends a signed download URL by email. Meta's official repository says these signed links expire after 24 hours and are also subject to a download limit.
- 4Install the official CLI with
pip install llama-models, runllama-model list, then usellama-model download --source meta --model-id CHOSEN_MODEL_IDand enter the signed URL when prompted. - 5You can also request access through Meta's official Hugging Face repositories instead of downloading directly from Meta.
Key Benefits
- Download Llama 4 Scout and Maverick model weights for self-hosted deployment.
- Royalty-free limited license for permitted use, modification, reproduction, and distribution.
- Llama 4 Scout supports up to a 10M-token context window; Maverick supports up to 1M, with both supporting text and image input.
- No mandatory Meta-hosted API is required to run downloaded weights; users can operate them on their own compatible infrastructure.
Eligibility & Requirements
| Requirements | • No credit card required |
|---|---|
| Criteria |
|
Meta Platforms develops Llama, a family of downloadable language and multimodal AI models intended for developers, researchers, and businesses. Current official Llama documentation lists Llama 4 Scout and Maverick as downloadable models and provides resources for local deployment, customization, and integrations. Llama uses a custom community license with usage and commercial restrictions, so "open-weight" is more precise than treating the weights as an unrestricted free API.
