Qwen3.5-9B

Qwen3.5-9B

A 9B open model from Qwen that reads text and images and writes text, with a long context and support for 201 languages and dialects. For developers who want a mid-size model on their own hardware.

Made by Qwen on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Qwen3.5-9B is a 9B model from Qwen that reads text and images and answers in text. This repository holds the post-trained version, built from Qwen3.5-9B-Base, with a vision encoder joined to the language model. The weights come in the Hugging Face Transformers format, and the card says other tools can load them too.

Who it suits: It suits developers and teams who want a mid-size open model for chat, coding, agents and image questions on their own hardware.

Industry | Who uses it | How they use it | Example
Software | Developers | Build chat, coding and agent features on a local model | An assistant that answers questions about code and screenshots
Customer support | Support teams | Answer questions in many languages | A reply written in the customer's own language
Research | Researchers | Fine-tune or quantize a model for their own tasks | Compare a quantized copy with the original
Business | Analysts | Ask questions about charts and documents shown as images | Send a chart picture and ask what it shows

If you need a model that draws images, you can skip it, since it reads images but writes text.

Where to use it:

• Python with the transformers library
• Serving tools that the card lists as compatible, which are vLLM, SGLang and KTransformers
• Windows, macOS or Linux machines with enough memory

Use case | Who | What they get back
Chat and coding | Developers | Text answers from a 9B model
Image questions | Analysts | Text answers about a picture
Multilingual work | Support teams | Replies in many languages

How to set it up: You need these first:

• Python, for the Hugging Face download command
• Disk space for the weights, about 19.3 GB in BF16 or about 4.8 GB as a 4-bit estimate
• Enough GPU or system memory to load the model
• Transformers, or one of the serving tools above

1. Install the Hugging Face hub tool.
2. Download the model files.
3. Load them with Transformers, or serve them with one of the compatible tools.

pip install -U huggingface_hub
hf download Qwen/Qwen3.5-9B

How to use it: Send text, or an image with text, and read the text that comes back. The part of the card given here shows no run code, so read the full card for the exact call.

Tool | What it does
Transformers | Loads the weights in their Hugging Face format
vLLM | A serving tool the card lists as compatible
SGLang | Another serving tool the card lists
KTransformers | A third tool the card lists

Example: You send a screenshot of a sales chart and ask what trend it shows. The model reads the picture and replies in text. The card does not show this exact run, so treat it as an expectation.

For newcomers and experts: Image text to text means you give the model pictures and words, and it answers in words. A good first step is to read the full card and try a short prompt before anything bigger.

The language model has 32 layers, laid out as eight repeats of three Gated DeltaNet blocks followed by one Gated Attention block. The hidden dimension is 4096, and the card says its multi-token prediction head was trained with multiple steps.

Pick the BF16 weights or a quantized copy
Load them with Transformers or a serving tool
Send text, or an image with text
Read the text reply

Good to know: Weights are modest for a model with vision, and the 4-bit figure is only Hugging Face's own estimate.

Title: Weight size in GB
BF16 weights: 19.3 GB
4-bit estimate: 4.8 GB

• Context: 262,144 tokens natively, and up to 1,010,000 tokens when extended
• Languages: the card lists support for 201 languages and dialects
• Benchmarks: the card compares it with other open models in tables, and these are the authors' own numbers
• Other versions: 695 quantized and 1,316 fine-tuned versions made by other people are listed
• License: Apache 2.0
• Cost to run: none named, you supply the hardware
• Data it reaches: the text and images you give it, and the model files you download

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 9.7B
• Weights: about 19.3 GB in BF16; about 4.8 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 6,259,809
• Likes: 2,353
• License: apache-2.0
• Based on: Qwen/Qwen3.5-9B-Base
• Quantized versions: 709
• Fine-tunes: 1,325
• Tags: transformers, safetensors, qwen3_5, image-text-to-text, conversational

Links

• Model: https://huggingface.co/Qwen/Qwen3.5-9B
• Files: https://huggingface.co/Qwen/Qwen3.5-9B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Qwen/Qwen3.5-9B

Then follow the usage in its model card. Nothing in this kit runs by itself.