Model Overview Card for Llama 4 Maverick

Basic Information

• Model Name: Llama 4 Maverick​
• Developer/Creator: Meta​
• Model Type: Multimodal AI Model​

Description

Overview

Llama 4 Maverick employs a mixture-of-experts architecture, featuring 128 experts with a total of approximately 400 billion parameters, utilizing 17 billion active parameters per token. This design enables it to outperform models like GPT-4o and Gemini 2.0 in coding and reasoning benchmarks. ​

Key Features

• Mixture-of-Experts Architecture: Utilizes 128 experts for enhanced performance.​
• Multimodal support for text and images across 12 languages.
• Cost-effective deployment options with FP8 quantization for efficiency.

Intended Use

• Complex Problem Solving: Handles advanced reasoning tasks effectively.​
• Code Generation and Analysis: Excels in creating and understanding complex code structures.​
• Multimodal Applications: Designed for multilingual assistants, creative writing, visual storytelling, and coding applications.

Technical Details

Architecture

The model employs Meta’s mixture-of-experts (MoE) framework with active parameter counts of 400 billion. Maverick utilizes a large pool of 128 experts for task-specific activation.

Training Data

Trained on curated datasets including multilingual corpora, image datasets, and synthetic reasoning examples.

Usage

Code Samples:

‍

API Documentation:

Detailed API Documentation is available here.

Ethical Guidelines

Llama 4 Maverick has implemented safeguards against misuse such as generating harmful content or violating user privacy during tool integrations.

Licensing

Custom Llama 4 Community License