Qwen3-30B-A3B-GGUF by unsloth

Qwen3-30B-A3B-GGUF by unsloth

GGUF files of Qwen3-30B-A3B, quantized by unsloth, for running a local chat and reasoning model in llama.cpp, Ollama or Open WebUI, with a thinking mode you can switch on and off per turn.

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: Qwen3-30B-A3B-GGUF holds GGUF versions of Alibaba's Qwen3-30B-A3B language model, prepared by unsloth. Hugging Face lists it at 30.5B parameters and estimates about 15.3 GB at 4-bit.

The model can reason step by step in a thinking mode or answer directly in a non-thinking mode, and you choose the mode inside the chat. The card says unsloth's Dynamic 2.0 method keeps accuracy higher than other quantization approaches, which is the maker's own claim.

Who it suits: Developers and hobbyists who want to run a capable chat model on their own hardware and decide per question whether it should think first.

Industry | Who uses it | How they use it | Example
Software | Developers | Run a local assistant for code and logic questions | Switch thinking on for a tricky bug
Education | Students | Work through math problems with visible reasoning | Ask for the steps behind an answer
Home labs | Hardware enthusiasts | Test a 30B-class model on their own machine | Load the 4-bit file in llama.cpp

People who want a hosted chat service can skip it, since you run these files yourself.

Where to use it:

• llama.cpp
• Ollama
• Open WebUI
• Windows, macOS or Linux with Python, for downloading the files

Use case | Who | What they get back
Local chat assistant | Developers | Replies from a 30B-class model on your machine
Step-by-step reasoning | Students and analysts | A think section followed by the answer
Quick general replies | Everyday users | A direct answer with thinking switched off

How to set it up: Make sure you have room for a file of roughly 15 GB at 4-bit, plus memory to run it.

• Python and pip, for the Hugging Face download tools
• A runtime that reads GGUF, such as llama.cpp, Ollama or Open WebUI
• Memory for about 15.3 GB of 4-bit weights, as an estimate

1. Install the huggingface_hub package.
2. Download the files with the hf command.
3. Load a GGUF file in your runtime.
4. Add a thinking switch to your prompts when you want one.

pip install -U huggingface_hub
hf download unsloth/Qwen3-30B-A3B-GGUF

How to use it: Chat as usual. In llama.cpp, Ollama, Open WebUI and similar tools, you can add /think or /no_think to a user prompt or system message to change the thinking mode from turn to turn. The model follows the most recent instruction in a multi-turn chat.

Tool | What it does
/think | Turns thinking mode on for that turn
/no_think | Turns thinking mode off for that turn
hf download | Fetches the model files from Hugging Face

Example: You ask who the model is and add /no_think, and it replies at once with an empty think section. Then you ask how many times the letter r appears in strawberries and add /think. It reasons first and then gives the count. The card shows this conversation.

> Who are you /no_think

<think>

</think>

I am Qwen, a large-scale language model developed by Alibaba Cloud. [...]

> How many 'r's are in 'strawberries'? /think

<think>
Okay, let's see. The user is asking how many times the letter 'r' appears in the word "strawberries". [...]
</think>

The word strawberries contains 3 instances of the letter r. [...]

For newcomers and experts: Think of it as one assistant with two gears. Thinking mode shows its working before it answers, and non-thinking mode just answers. A good first try is to load the 4-bit file and ask the same question with each switch.

Underneath, this is a quantized copy of Qwen3-30B-A3B from a family that includes dense and mixture-of-experts models. The card says the model keeps both modes in one set of weights. The trade-off is that a 4-bit file saves memory and loses some precision.

Download the GGUF file
Load it in llama.cpp, Ollama or Open WebUI
Add /think or /no_think to a prompt
Read the reply, with or without reasoning

Good to know: The card also advertises unsloth's fine-tuning notebooks and export to Ollama, llama.cpp or Hugging Face, but those are separate from running these files.

• Size: 30.5B parameters in GGUF format, English listed as the language
• Base: Qwen/Qwen3-30B-A3B
• License: Apache 2.0
• Cost to run: no API key or paid service is named, though you need hardware that holds the weights
• Data it reaches: the prompts you type into your local runtime and the files you download from Hugging Face
• Not for: the card names no restrictions beyond its license

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 30.5B
• Weights: about 15.3 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 53,664
• Likes: 294
• License: apache-2.0
• Languages: en
• Based on: Qwen/Qwen3-30B-A3B
• Tags: transformers, gguf, qwen3_moe, text-generation, qwen3, qwen, unsloth, imatrix, conversational

Links

• Model: https://huggingface.co/unsloth/Qwen3-30B-A3B-GGUF
• Files: https://huggingface.co/unsloth/Qwen3-30B-A3B-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download unsloth/Qwen3-30B-A3B-GGUF

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