Qwen3.6-35B-A3B-GGUF

Qwen3.6-35B-A3B-GGUF

Runs Qwen3.6-35B-A3B, a mixture-of-experts model that reads images and text and uses about 3B parameters per token, from Unsloth GGUF files. For developers who run big models locally.

Made by unsloth 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: Qwen3.6-35B-A3B-GGUF holds Unsloth GGUF versions of Qwen3.6-35B-A3B, an open-weight model from Qwen that takes in images and text and writes text. It has about 35 billion parameters in total, and about 3 billion are active for each token.

The notes from Unsloth add support for a developer message role, so the model can work inside coding tools such as Codex and OpenCode. They also describe better tool calling when the arguments are nested objects.

Who it suits: Developers and enthusiasts who want to run a large mixture-of-experts model on their own computer, for coding agents or for questions about pictures.

Industry | Who uses it | How they use it | Example
Software | Developers using coding agents | Run a local model behind a coding tool | Connect it to Codex or OpenCode
Research | People comparing open models | Test a quantized model of a known base | Compare it with other local models
Hobby | Enthusiasts with a strong PC | Chat with a model that also reads images | Ask a question about a screenshot

Skip it if you have little memory, since the 4-bit weights alone are about 17.3 GB by Hugging Face estimate.

Where to use it:

• Local runners that open GGUF files
• Unsloth Studio, where the card shows a 4-bit version using tools
• Coding tools such as Codex and OpenCode, through the developer role
• Windows, macOS or Linux with Python, to download the files

Use case | Who | What they get back
Agentic coding | Developers | Help with front-end work and reasoning over a whole repository
Tool calling | Agent builders | Tool calls that handle nested arguments more often
Image questions | Everyday users | Answers about an image you provide

How to set it up:

• Python and pip to run the download commands
• Enough memory for about 17.3 GB of files at 4-bit
• A GGUF-capable runner, or Unsloth Studio, with the Unsloth guide for running Qwen3.6

1. Install the Hugging Face tools.
2. Download the GGUF files from this repository.
3. Open them in your runner and follow the Unsloth guide.

pip install -U huggingface_hub
hf download unsloth/Qwen3.6-35B-A3B-GGUF

How to use it: Load the files in your runner, then chat with the model or connect it to a coding tool. You can send text, or an image together with a question.

Property | Qwen3.6-35B-A3B
Total parameters | 35B, with about 3B active per token
Layers | 40, in a repeating pattern of three linear-attention layers and then one full-attention layer
Experts | 256 in all, with 8 routed plus 1 shared used for each token
Context | 262,144 tokens at the start, extendable to 1,010,000

Title: Model size in billions of parameters
Total parameters: 35 billion
Active per token: 3 billion

Example: You open the 4-bit files in Unsloth Studio and turn on tool calling. Then you ask the model to help with a front-end task in a project and watch it call tools as it works.

For newcomers and experts: If you are new to local models, start with a short chat and an image question before you wire it into a coding tool. Large files take time to download.

If you know the field, the language model mixes gated DeltaNet layers with gated attention layers, and each layer ends in a mixture-of-experts block. The card also says it was trained with multi-step prediction.

Download the GGUF files
Open them in a local runner or Unsloth Studio
Send text or an image
Receive text, or a tool call for your coding tool

Good to know:

• A new option keeps the reasoning from earlier messages, which the authors say helps iterative work
• The base model files work with Transformers, vLLM, SGLang and KTransformers, while this repository holds the GGUF versions
• Hugging Face lists 6 quantized and 1 fine-tuned version made by other people
• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the files you download and the text and images you give it
• Not for: small machines, since even the 4-bit files are large

On Hugging Face, 13 November 2026

• Task: image text to text
• Library: transformers
• Parameters: 34.7B
• Weights: about 17.3 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 1,649,871
• Likes: 1,789
• License: apache-2.0
• Based on: Qwen/Qwen3.6-35B-A3B
• Quantized versions: 6
• Fine-tunes: 1
• Tags: transformers, gguf, unsloth, qwen, qwen3_5_moe, image-text-to-text, imatrix, conversational

Links

• Model: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF
• Files: https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download unsloth/Qwen3.6-35B-A3B-GGUF

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