MODELS

Llama 4 Maverick

by Meta

multimodalllamameta

Overview

Meta Llama 4 Maverick is an open-weight, natively multimodal Mixture-of-Experts model for text and image understanding.

Details

Llama 4 Maverick is part of Meta’s Llama 4 family. Meta describes it as a natively multimodal open-weight MoE model with 17B active parameters, 128 experts, text-and-image understanding, and a 1M-token context window. The official model card identifies Llama 4 Maverick as a 17B-active-parameter, 400B-total-parameter multimodal MoE model and documents intended use, benchmarks, safeguards, red teaming, and system protections.

When to Use

Use when you need an open-weight Llama 4 model with both text and image understanding. Use when evaluating a large-context multimodal model; Meta’s launch materials describe a 1M-token context window. Use via managed platforms when appropriate: AWS Bedrock documents Llama 4 Maverick 17B Instruct with Converse and Invoke API access and Google Cloud lists token pricing for Llama 4 Maverick.

Getting Started

Read the Llama 4 model page and the Llama 4 model-card and prompt-format documentation on llama.com.
Review the official GitHub model card for intended use
benchmarks
safeguards
red teaming
and system protections.
Check Meta’s Llama get-started page for downloads and partner access options.
If using a cloud provider
review the AWS Bedrock model card or Google Cloud pricing page for platform-specific access and costs.
Review the Llama 4 Acceptable Use Policy before deploying applications built on the model.

Key Features

•Natively multimodal model for image and text understanding.
•Mixture-of-Experts architecture with 17B active parameters and 128 experts
•according to Meta’s launch materials.
•Official model card describes 400B total parameters for Llama 4 Maverick.
•Meta’s launch materials describe a 1M-token context window.
•Official documentation and GitHub resources include model card
•prompt-format information
•safeguards
•red teaming
•and system protections.

Capabilities

•text generation
•image understanding
•multimodal understanding
•long-context processing
•managed API access via supported cloud platforms

Last updated Jul 18, 2026