Instructions to use Mindral/ComRAD-200B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Mindral/ComRAD-200B-GGUF 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 Mindral/ComRAD-200B-GGUF
# Run inference directly in the terminal:
llama cli -hf Mindral/ComRAD-200B-GGUF

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Mindral/ComRAD-200B-GGUF
# Run inference directly in the terminal:
llama cli -hf Mindral/ComRAD-200B-GGUF

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 Mindral/ComRAD-200B-GGUF
# Run inference directly in the terminal:
./build/bin/llama cli -hf Mindral/ComRAD-200B-GGUF

Use Docker

docker model run hf.co/Mindral/ComRAD-200B-GGUF

Llama new
LM Studio
Jan

vLLM

How to use Mindral/ComRAD-200B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Mindral/ComRAD-200B-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Mindral/ComRAD-200B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Mindral/ComRAD-200B-GGUF

Ollama
How to use Mindral/ComRAD-200B-GGUF with Ollama:
```
ollama run hf.co/Mindral/ComRAD-200B-GGUF
```
Unsloth Desktop

How to use Mindral/ComRAD-200B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Mindral/ComRAD-200B-GGUF

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": "Mindral/ComRAD-200B-GGUF"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Mindral/ComRAD-200B-GGUF with Docker Model Runner:
```
docker model run hf.co/Mindral/ComRAD-200B-GGUF
```

Lemonade

How to use Mindral/ComRAD-200B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Mindral/ComRAD-200B-GGUF

Run and chat with the model

lemonade run user.ComRAD-200B-GGUF-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use Mindral/ComRAD-200B-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Mindral/ComRAD-200B-GGUF

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 Mindral/ComRAD-200B-GGUF

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Mindral/ComRAD-200B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Mindral/ComRAD-200B-GGUF

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 "Mindral/ComRAD-200B-GGUF" \
  --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"

ComRAD 200B

Boreale · Experimental GGUF preview

ComRAD 200B is an experimental Mixture-of-Experts language model with approximately 195 billion parameters, derived from Z.ai's GLM-5.3-Flash through the proprietary ComRAD weight transformation methodology, without additional training or fine-tuning.

A project by Federico Alessandro, 17, under Mindral.

ComRAD 200B is the public model name. The exact parameter count is 194,927,503,582. In conversation, the model identifies itself as ComRAD Boreale.

Property	This checkpoint
Parameters	194.93 billion
Architecture	Mixture-of-Experts
Format	GGUF, Q4_K
File size	114,333,049,568 bytes — about 114.33 GB
Interface	Text generation and conversation
Tested context	4,096 tokens
License	Apache 2.0 for ComRAD contributions; upstream MIT notice retained

Try the checkpoint

Download ComRAD-Boreale-195B-Q4_K.gguf and import it into a runtime that supports this model's architecture. Use its embedded chat template.

This checkpoint was tested with LM Studio's llama.cpp CUDA runtime 2.66.0. The tested machine had an RTX PRO 6000 with 96 GB VRAM and approximately 188 GB system RAM; part of the model was offloaded to system RAM. This is a tested configuration, not a measured minimum hardware requirement. Runtime memory use exceeds the weight file size.

For the configuration used in the current chat:

Context length: 4,096 tokens.
Maximum generated tokens: 1,536, including reasoning.
Temperature: 0; top-p: 1; top-k: 0; min-p: 0.
Repetition penalty: 1.0; frequency and presence penalties: 0.
Use the embedded template, which defaults to High reasoning.

See inference_config.json for the exact tested settings. Offload values depend on the machine and runtime. Short generation limits may cut off reasoning before the final answer.

The GGUF is standalone for inference; the upstream GLM checkpoint is not required at runtime. No proprietary ComRAD chat service is required to load it in a compatible runtime.

Evaluation so far

These are small functional checks, not standardized benchmark scores.

Check	Evidence
Identity	Returned "ComRAD Boreale" and ended naturally in the served configuration
C++17 code example	Compiled with g++ using `-std=c++17 -O2 -Wall -Wextra -pedantic`, without diagnostics
Functional checks for that C++ example	12/12 cases passed, including empty input, negative integers, repeated values, original ordering and integer extremes
Eight informal reasoning and instruction-following questions	All passed according to the author; complete transcripts were not independently reviewed
Conversation history	The author reported successful use of prior messages within the same conversation

The C++ task retained only values appearing exactly once while preserving their original order. The 12 cases evaluate one generated program, not 12 independent programming tasks.

Limitations

This is an experimental preview. Earlier targeted tests observed scientific inaccuracies, coding errors and outputs that exceeded a short generation budget. Passing the current examples does not establish that these failures are resolved for other prompts.

No broad benchmark or quality-parity claim with the source model is made. Other runtimes, operating systems, context lengths and hardware configurations have not been validated by this test session.

Conversation history is supplied by the chat application. Persistent memory between separate conversations is not included.

Release integrity

This release freezes the checkpoint tested on 13 November 2026. The complete saved GGUF was reread and its SHA-256 verified.

SHA256SUMS and manifest.json identify the exact artifact. Future experiments should use a separate version.

License and attribution

ComRAD contributions are distributed under Apache License 2.0. The upstream GLM-5.3-Flash MIT license and copyright notice are retained in LICENSE.upstream; see NOTICE.

Apache 2.0 permits commercial use, modification and redistribution subject to its terms. No separate non-commercial or no-resale condition is imposed on this release.

Base model: zai-org/GLM-5.3-Flash, revision 46a050bec13002ad29ee795c16e61cf4fa3c4a9c.

The proprietary ComRAD methodology, compiler and internal development tooling are not included in this repository.

Downloads last month: 6,007

GGUF

Model size

195B params

Architecture

glm5-next

Hardware compatibility

Model tree for Mindral/ComRAD-200B-GGUF

Base model

zai-org/GLM-5.3-Flash

Quantized

(241)

this model