Instructions to use ccore/Qwen3.5-4B-GLB-3D-Coder with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use ccore/Qwen3.5-4B-GLB-3D-Coder with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Use Docker

docker model run hf.co/ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use ccore/Qwen3.5-4B-GLB-3D-Coder with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ccore/Qwen3.5-4B-GLB-3D-Coder"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ccore/Qwen3.5-4B-GLB-3D-Coder",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Ollama
How to use ccore/Qwen3.5-4B-GLB-3D-Coder with Ollama:
```
ollama run hf.co/ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M
```
Unsloth Desktop

How to use ccore/Qwen3.5-4B-GLB-3D-Coder with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ccore/Qwen3.5-4B-GLB-3D-Coder with Docker Model Runner:
```
docker model run hf.co/ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M
```

Lemonade

How to use ccore/Qwen3.5-4B-GLB-3D-Coder with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.5-4B-GLB-3D-Coder-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use ccore/Qwen3.5-4B-GLB-3D-Coder with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ccore/Qwen3.5-4B-GLB-3D-Coder with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "ccore/Qwen3.5-4B-GLB-3D-Coder:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Qwen3.5-4B-GLB-3D-Coder

Early preview · Create 3D scenes and assets you can export as GLB

A small 4B coding model being trained to generate Three.js scenes and assets with GLB export, so you can bring the results into editors such as Godot.

This is an independent model by CCORE, based on Qwen/Qwen3.5-4B. It is not an official Qwen release.

Latest checkpoint update

Updated November 2, 2026 at 04:50 UTC.

Both Q4_K_M and Q8_0 have been replaced with a newer checkpoint. In my latest tests, generation quality has improved substantially over the previous preview. The gallery below now shows new results, including the exported assets imported into Godot.

If you downloaded either GGUF before the UTC timestamp above, delete the old weights from LM Studio or whichever app you use and download them again. The filenames have not changed, so make sure you are loading the updated files.

The optional vision projector is now included for image-reference inputs; see Optional vision support below.

The sampling settings and system prompt remain unchanged. Training is still running, and the full Safetensors weights and training dataset will follow once it finishes.

Preview status and release plans

This is an updated preview from the ongoing training run. Training is still in progress. Once training finishes, I plan to release the full Safetensors weights and the training dataset, just as I did with the previous 3D Coder release.

In my tests, the model already generates useful scenes and assets, and most generations work on the first attempt, especially with English prompts. Some generations still need another attempt or a correction.

I am releasing this preview so people can start experimenting while training continues. I expect quality to improve with later checkpoints, and feedback on both successful results and failures is welcome.

From a prompt to a GLB asset

Describe the scene or asset you want. The model generates HTML/JavaScript that builds it in Three.js, with a GLB export option. Open the generated scene, export the asset, and import the .glb file into your editor.

The screenshots below show generated assets in the browser and exported assets imported into Godot.

Example results — updated checkpoint

The assets below were generated by this updated GLB-3D-Coder checkpoint using prompts that were not used in its training. These example prompts are not included in the training dataset I plan to release.

The captions show the user prompts used for these examples. Short English prompts are enough; use the system prompt listed below.

Jungle siege platform	Weathered snow siege platform

Prompt: `Build a jungle siege platform`	Prompt: `Build a weathered snow siege platform`
Blue goblin brute	Red orc leader

Prompt: `Create a blue goblin brute`	Prompt: `Create a red orc Leader`
Reinforced mountain command post — result 1	Reinforced mountain command post — result 2

Prompt: `Make a reinforced montain command post`	Same prompt: `Make a reinforced montain command post`

Exported assets in Godot

The images named Godot show how the exported GLB assets look after import into the Godot editor.

Click any image to view it at full size.

Available GGUF files

Quantization	File
Q8_0	Qwen3.5-4B-GLB-3D-Coder-Q8_0.gguf
Q4_K_M	Qwen3.5-4B-GLB-3D-Coder-Q4_K_M.gguf

The full Safetensors weights and training dataset will follow when training is complete.

Optional vision support

For image-reference inputs in a runtime that supports Qwen3.5 multimodal GGUF, download the optional vision encoder/projector and load it alongside the model:

Download mmproj-Qwen3.5-4B-BF16.gguf — approximately 644 MiB.

The same projector is used with either Q4_K_M or Q8_0. It is an optional companion file for image inputs; it is not needed for text-only prompts.

This is the original, unmodified Qwen3.5-4B BF16 vision encoder/projector, also provided with CCORE's previous 3D Coder model. It comes from lmstudio-community/Qwen3.5-4B-GGUF, based on Qwen/Qwen3.5-4B, under Apache 2.0.

Generation settings

These are the settings I am currently using:

Parameter	Value
Temperature	0.8
Top K	40
Top P	0.95
Min P	0.0

This is a preview, and these settings are a starting point. Feel free to test other combinations and share your results.

System prompt

Use the following as the system message, then send your asset request as the user message.

Create game assets for the Godot editor, which uses GLB files. Create whatever the user requests as a good-quality low-poly asset using only Three.js in one complete standalone HTML file. The HTML must show the asset and include an Export GLB button that downloads a working GLB. If the requested asset needs animation, add it and include the animation in the exported GLB; otherwise make it static.

Example prompts

English prompts have been the most reliable in my tests. With the system prompt above, try:

Build a jungle siege platform

Build a weathered snow siege platform

Create a blue goblin brute

Create a red orc Leader

Make a reinforced montain command post

Current limitations

Some generations can need a retry or a small code correction.
English prompts currently work better in my tests.
This is an early checkpoint, and behavior may change as training progresses.

Please share what you create, which settings you use, and any issues you find. Community testing will help guide the next updates.

❤️ Support Future Open-Source AI Models

I'm independently developing and releasing specialized AI models and datasets for 3D generation, procedural environments, and creative coding, with the goal of making these tools accessible to everyone.

Training new models, generating high-quality datasets, validating thousands of examples, and running experiments require significant GPU resources, time, and effort.

If you enjoy this project and would like to help me train more models, expand the datasets, and explore new capabilities, any contribution is greatly appreciated!

My goal is to keep sharing models, weights, datasets, and improvements openly with the community.