Qwen3.5-27B

Qwen3.5-27B

Reads text and images and writes text, as a 27B Qwen model with a vision encoder and a 262,144-token native context. For developers who build assistants and agents on server-class 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: Qwen3.5-27B is a Qwen model that reads text and images and writes text. It pairs a 27B language model with a vision encoder, and this repository holds the post-trained version in the Transformers format. The card lists support for 201 languages and dialects and a native context of 262,144 tokens.

Who it suits: It suits developers and teams who want an open multimodal model for assistants, agents and coding, and who have server-class hardware.

Industry | Who uses it | How they use it | Example
Software development | Developers | Build assistants that read text and pictures | Asking questions about a screenshot
Global support | Multilingual teams | Answer people in many languages | Replying in one of the 201 listed languages
AI agents | Agent builders | Use it as the model behind tool-using agents | Running a multi-step task in an agent framework
AI research | Researchers | Study a hybrid-attention vision-language model | Comparing it with the earlier Qwen3 models

You can skip it if you have a small machine, since the BF16 weights come to about 55.6 GB.

Where to use it:

• Hugging Face Transformers, the format this repository uses
• vLLM, SGLang and KTransformers, which the card names as compatible
• Windows, macOS or Linux with Python, to download the files
• A GPU server with enough memory for the weights

Use case | Who | What they get back
Picture questions | Developers | Text answers about an image you send
Coding and reasoning | Engineers | Written code and step-by-step answers
Multilingual chat | Support teams | Replies in many languages

How to set it up: You need Python for the download tool and a runtime from the list above. The weights are about 55.6 GB in BF16, or roughly 13.9 GB at 4-bit as an estimate.

• Python on Windows, macOS or Linux
• The huggingface_hub package
• One compatible runtime, such as Transformers or vLLM
• A GPU with room for the weights, or a quantized copy that fits

1. Install huggingface_hub.
2. Download the model files.
3. Load them in your runtime, or pick a quantized copy for a smaller machine.

pip install -U huggingface_hub
hf download Qwen/Qwen3.5-27B

How to use it: The visible card gives the download step, a model overview and benchmark tables, but no run command. After your runtime loads the files, you send text or an image with a question and read the text reply.

Part | Detail
Language model | 27B parameters across 64 layers
Vision | A vision encoder built into the same model
Context | 262,144 tokens natively, extensible to 1,010,000
Languages | 201 languages and dialects

Example: You load the model in a serving tool such as vLLM, send a picture with a question, and get a written answer back.

For newcomers and experts: If you are new to this, it is a large model that can look at a picture and talk about it. A first small try is a quantized copy, since the full files need a big GPU.

For people who know the field, 16 blocks each stack three Gated DeltaNet layers and one Gated Attention layer, with a feed-forward layer after each. The card says early fusion training on multimodal tokens matches Qwen3 on text and beats the Qwen3-VL models on visual tasks.

Title: Context length in tokens
Native context: 262144 tokens
Extended context: 1010000 tokens

You download the files
Your runtime loads the language model and vision encoder
You send text or an image with a question
The model writes a text answer

Good to know: The benchmark tables compare it with GPT-5-mini, GPT-OSS-120B, Qwen3-235B-A22B and other Qwen3.5 models. The scores come from the authors.

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the prompts and images you send, plus the downloaded files
• Not for: small machines at full size, since BF16 weights are about 55.6 GB

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 27.8B
• Weights: about 55.6 GB in BF16; about 13.9 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,651,125
• Likes: 1,091
• License: apache-2.0
• Quantized versions: 243
• Fine-tunes: 347
• Tags: transformers, safetensors, qwen3_5, image-text-to-text, conversational

Links

• Model: https://huggingface.co/Qwen/Qwen3.5-27B
• Files: https://huggingface.co/Qwen/Qwen3.5-27B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Qwen/Qwen3.5-27B

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