Model reference · open weights

Llama-4-Maverick-128E

LLMs meta-llama Vision + text 2 builds Its own licence terms 78k dl/mo

Llama-4-Maverick-128E is an open-weight language model from Meta. Llama-4-Maverick-17B-128E-Instruct (BF16) weighs 803 GB; it needs more than the reference configurations below.

Llama-4-Maverick-128E is a natively multimodal model developed by Meta that uses a mixture-of-experts architecture for text and image understanding.
It is intended for commercial and research use in assistant-like chat, visual reasoning, and natural language generation.
The model has 401.6B total parameters, supports 12 languages including English, Arabic, and Spanish, and is released under the Llama 4 Community License Agreement.

Summary of the meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8 model card, 2026-10-01. The estimate below is for another build of the family.

Hugging Face ↗GitHub ↗Paper ↗

What it is

Released by	Meta
Released	2025-04-01
Parameters	401.6B
VRAM	803 GB for the weights

What it runs on

Memory and cards for Llama-4-Maverick-17B-128E-Instruct (BF16)

803 GBweights, file size

762 MBruntime overhead, at least

How much memory each request adds isn't estimated yet for this architecture. The weights need at least the cards below, plus room for the context.

Card	Weights alone
RTX 3060 12 GB … 8× B200 180 GB 16 smaller cards	does not fit

Builds

Sizes, precisions & builds

Build	Params	Precision	Weights	Smallest setup
Llama-4-Maverick-17B-128E-Instruct-FP8 ↗	401.6B	FP8	417 GB	none of the reference configurations
Llama-4-Maverick-17B-128E-Instruct (above) ↗	401.6B	BF16	803 GB	none of the reference configurations

How it works

How language models work

Details

Published under	meta-llama
Type	Language models
Task	Vision + text
Runs with	transformers
Based on	meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8
Popularity	78k downloads / month
Weights	803 GB, Llama-4-Maverick-17B-128E-Instruct (BF16, file size)

Languages

ardeenesfrhiiditptthtlvi