Llama 4 Maverick (17Bx128E)
Release in the Llama family · version 4 Maverick 17B-128E
Llama 4 Maverick is a natively multimodal mixture-of-experts model from Meta with 17 billion active and 400 billion total parameters across 128 experts. It accepts multilingual text and images and produces text and code; the model card lists a 1M-token context length and an August 2024 knowledge cutoff. Pretrained and instruction-tuned versions were released.1
- Documentation: Model card (external site: github.com)
- Model hub: Hugging Face (instruction-tuned) (external site: huggingface.co)
- Model hub: Hugging Face (pretrained) (external site: huggingface.co)
- License: Llama 4 Community License Agreement (external site: github.com)
- Repository: Reference implementation (external site: github.com)
Availability and license
Overall availability
Weights are downloadable from Meta or Hugging Face after submitting an access request and accepting the Llama 4 Community License. Meta's instructions say a signed download URL is emailed once the request is approved; the Hugging Face repositories are gated. The acceptable use policy withholds the license grant for Llama 4 multimodal models from individuals domiciled in, and companies with a principal place of business in, the European Union.463
Availability is separate from permission: read the license before using or redistributing.
Custom Meta license, effective April 5, 2025, covering weights and code. It grants a royalty-free, non-exclusive license to use, modify, and redistribute. Redistributors must include the agreement, show "Built with Llama", and keep an attribution notice; models trained or fine-tuned with Llama materials or outputs that are distributed must start their names with "Llama". Use must follow the Llama 4 Acceptable Use Policy, which is incorporated by reference. Licensees whose products had more than 700 million monthly active users in the month before the release date must request a separate license from Meta. The license ends for anyone who sues Meta or any other entity alleging that the Llama materials or their outputs infringe their IP. California law governs. The acceptable use policy excludes EU-domiciled individuals and EU-based companies from the rights granted for Llama 4 multimodal models.23
Model-disclosure tier
The weights can be obtained only by request, with approval, or by some users — for example a gated download that the publisher reviews. Not counted as open-weight.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Partial | Downloadable after an access request and license acceptance; Meta sends download links once a request is approved. Not available under the license to EU-domiciled individuals or EU-based companies for multimodal models.463 |
| Inference codeIs code for running the model published? | Public | Meta's llama-models repository includes Llama 4 model, generation, and quantization code, and the Hugging Face card documents inference with Transformers 4.51.0 or later.56 |
| Training codeIs the code used to train the model published? | Unknown | The model card says Meta used custom training libraries; this review found no published training code.1 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The model card describes about 22 trillion tokens drawn from publicly available data, licensed data, and information from Meta's products and services, including public Instagram and Facebook posts and interactions with Meta AI. The data is not released.1 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | Meta's announcement describes the MoE architecture, early fusion, the MetaP hyperparameter technique, and FP8 training at a high level; no full recipe is published.7 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The model card reports benchmark results measured on bf16 models; evaluation code and prompts were not located in this review.1 |
What it is useful for
The model card lists commercial and research use in 12 supported languages, assistant-style chat and visual reasoning for the instruction-tuned model, and synthetic data generation and distillation to improve other models.1
Run and use notes
Organization context
Other releases in the Llama family
- Llama 4 Scout (17Bx16E)Model-disclosure tier (USASI rubric v0.1): Restricted weights
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.