Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF

Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF

Answers questions and reads images with much shorter thinking. GGUF builds of a 27B Qwen 3.8 tune from DavidAU that aims to cut reasoning tokens while keeping accuracy, for people who run models locally.

Made by DavidAU 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: This is a set of GGUF files for a 26.9B model tuned from Qwen3.8-27B by DavidAU, with help on benchmarking from Nightmedia and on datasets from TeichAI. It was trained with a method the author calls Cold Fusion. The aim is to shrink the model's thinking blocks and speed up generation, MTP included, without hurting its answers. It handles text and images and lists English and Chinese.

Who it suits: People who run large reasoning models locally and want less waiting on long thinking output.

Industry | Who uses it | How they use it | Example
Software development | Developers who run local coding and chat models | Swap it in for a regular 27B model to get shorter reasoning | Asking a coding question and getting a shorter thinking block
Research | People who compare local models | Test its thinking length against the base Qwen3.8-27B | Counting reasoning tokens on the same prompts
General use | Enthusiasts with a strong graphics card | Chat with it and give it images | Asking a question about a screenshot

Skip it if you have no hardware for a 27B model, or if you want the base model unchanged.

Where to use it: GGUF files load in local apps and runtimes that read that format. The files download on Windows, macOS or Linux.

• Local runtimes and apps that load GGUF files
• Windows, macOS or Linux: where you can fetch the files with Python
• Setups that use MTP for faster token generation

Use case | Who | What they get back
Reasoning questions | Developers | Answers with a shorter thinking block
Image questions | Everyday users | Text replies about a picture
Model comparison | Researchers | A tuned model to test against base Qwen3.8-27B

How to set it up: You need Python to download, a GGUF capable app, and room for the quant you pick. At 4-bit the weights come to about 13.4 GB, as an estimate.

• Python with pip
• A GGUF capable local app
• Disk space for one quant file

1. Install the Hugging Face hub library.
2. Download the files, or pick one quant file from the repository.
3. Load it in your app.

pip install -U huggingface_hub
hf download DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF

How to use it: The card gives no run command in the part shown. Load a quant in your app and prompt it as you would a regular Qwen model. The author says the changes work in all three thinking modes.

Tool | What it does
hf download | Fetches the model files to your computer

Example: You load the model and ask a multi step reasoning question. The author says it thinks for roughly half to a tenth of the tokens the regular Qwen3.8 model uses, then gives the answer.

For newcomers and experts: If you are new to this, a thinking model writes out its reasoning before it answers. This one is tuned to write less of it. Try one question you already tried on another model and compare the length.

If you know the field, the author says the Cold Fusion training changes how each sample is trained as the model learns, and that 4-bit keeps about 99 percent of BF16 performance. The author also reports a median cut of about two thirds in thinking size. These are the author's own claims.

Qwen3.8-27B base model
Cold Fusion GAIN training
Shorter thinking blocks
GGUF quants for local use

Good to know: The author says checks were run at each stage, ending with side by side human tests against the base model.

• License: Apache 2.0
• Cost to run: none named, you supply the hardware
• Data it reaches: the card names none
• Not for: the card names no exclusions, and it says a larger tune is still planned

On Hugging Face, 13 November 2026

• Task: image text to text
• Parameters: 26.9B
• Weights: about 13.4 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 372,115
• Likes: 342
• License: apache-2.0
• Languages: ar, zh, en, fr, de, hi, id, it
• Based on: DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1
• Tags: gguf, qwen3_5, unsloth, cold-fusion, finetune, qwen3.8, thinking, reasoning, coder, creative, story, writing

Links

• Model: https://huggingface.co/DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF
• Files: https://huggingface.co/DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-NM-DAU-NEO-MAX-MTP-GGUF

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