Model reference · open weights

Qwen3-VL

LLMs amd Vision + text · MoE 1 build Open weights 23k dl/mo

Qwen3-VL is an open-weight language model from amd. Qwen3-VL-235B-A22B-Instruct-MXFP4 (MXFP4) weighs 135 GB in vLLM, 127 GB of files; the smallest configuration that runs it is B200 180 GB.

Qwen3-VL is a vision-language model developed by AMD for image-text-to-text tasks, featuring a 118.8B parameter MoE architecture.
It supports a native 256K context length expandable to 1M and handles OCR in 32 languages.
The model is released under the Apache-2.0 license.

Summary of the amd/Qwen3-VL-235B-A22B-Instruct-MXFP4 model card, 2026-10-01

Hugging Face ↗Paper ↗

What it is

Released by	amd
Released	2026-07-27
Parameters	118.8B
VRAM	135 GB in vLLM, 127 GB of files for the weights

What it runs on

Memory and cards for Qwen3-VL-235B-A22B-Instruct-MXFP4 (MXFP4)

135 GB in vLLM, 127 GB of filesweights, 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 … H200 141 GB 11 smaller cards	does not fit
B200 180 GB	fits
8× H100 80 GB tensor parallel	does not fit
8× A100 80 GB tensor parallel	does not fit
8× H200 141 GB tensor parallel	does not fit

How it works

How language models work

Details

Type	Language models
Task	Vision + text · MoE
Runs with	transformers
Popularity	23k downloads / month
Weights	135 GB in vLLM, 127 GB of files, Qwen3-VL-235B-A22B-Instruct-MXFP4 (MXFP4, file size)