MODELS

Llama 4 Maverick 17B-128E

by Meta

llama-4meta

Overview

Open-weight, natively multimodal mixture-of-experts Llama 4 model with 17B active parameters, 128 experts, and a 1M-token context window.

Details

Llama 4 Maverick (17B-128E) is Meta’s open-weight, natively multimodal mixture-of-experts model. Meta describes it as having 17B active parameters, 128 experts, and 400B total parameters. The official model card describes multilingual text-and-image input, text-and-code output, and a 1M-token context window. Hosted documentation from AWS describes multimodal inference, streaming, guardrails, and tool calling for the Maverick Instruct model.

When to Use

Use for applications that need to process text and images and return text or code. Use for workflows that benefit from a long up-to-1M-token context window. Evaluate for agentic applications requiring tool calling streaming or structured interactions through a supported hosted provider.

Getting Started

Review Meta’s Llama 4 overview and the official Maverick model card.
Select an access path
such as model weights or a supported hosted service.
For Amazon Bedrock
use the Maverick Instruct model documentation and sample API calls to configure inference.
Test multimodal inputs
context-length requirements
and tool-calling behavior against your application before deployment.

Key Features

•Mixture-of-experts architecture with 17B active parameters and 128 experts.
•400B total parameters.
•Natively multimodal text-and-image input.
•Multilingual text-and-image input with text-and-code output.
•Up to 1M-token context window.
•Hosted AWS Bedrock support for streaming
•guardrails
•and tool calling.

Capabilities

•multimodal input
•image understanding
•text generation
•code generation
•multilingual
•long context
•tool calling
•streaming

Last updated Aug 11, 2026