Instructions to use zenlm/zen6-flash with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use zenlm/zen6-flash 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 zenlm/zen6-flash:Q8_0
# Run inference directly in the terminal:
llama cli -hf zenlm/zen6-flash:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf zenlm/zen6-flash:Q8_0
# Run inference directly in the terminal:
llama cli -hf zenlm/zen6-flash:Q8_0

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf zenlm/zen6-flash:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf zenlm/zen6-flash:Q8_0

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf zenlm/zen6-flash:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf zenlm/zen6-flash:Q8_0

Use Docker

docker model run hf.co/zenlm/zen6-flash:Q8_0

LM Studio
Jan

vLLM

How to use zenlm/zen6-flash with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "zenlm/zen6-flash"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "zenlm/zen6-flash",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/zenlm/zen6-flash:Q8_0

Ollama
How to use zenlm/zen6-flash with Ollama:
```
ollama run hf.co/zenlm/zen6-flash:Q8_0
```
Unsloth Desktop

How to use zenlm/zen6-flash with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zenlm/zen6-flash:Q8_0

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": "zenlm/zen6-flash:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use zenlm/zen6-flash with Docker Model Runner:
```
docker model run hf.co/zenlm/zen6-flash:Q8_0
```

Lemonade

How to use zenlm/zen6-flash with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull zenlm/zen6-flash:Q8_0

Run and chat with the model

lemonade run user.zen6-flash-Q8_0

List all available models

lemonade list

Hermes Agent

How to use zenlm/zen6-flash with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zenlm/zen6-flash:Q8_0

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 zenlm/zen6-flash:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use zenlm/zen6-flash with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf zenlm/zen6-flash:Q8_0

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 "zenlm/zen6-flash:Q8_0" \
  --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"

Zen 6 Flash

The compact Zen 6 · ternary 27B · 1.77 bits per weight · reads images · fits a laptop

Zen 6 Flash is the ternary build of Zen 6, from Zen LM, the open model family of Zoo Labs Foundation, a 501(c)(3) non-profit. It is chosen for the same two jobs as Zen 6, where the machine is small:

Agentic coding that runs on your own machine, in under 8 GB with its vision projector, on an Apple Silicon laptop or one GPU.
Marketing work, reading images and documents beside the brief.

Zen 6 and Zen 6 Flash are available now; Zen 7 is in research preview: request access. Zen 6 Flash answers on api.hanzo.ai as zen6-flash.

What the files are

Every weight is ternary, the embeddings and the LM head included. The language model holds 26,895,998,464 weights (read from the GGUF tensor table), so a file's bits per weight is its size in bits over that count, and its ratio to FP16 is 53.79 GB (the count × 2 bytes) over its size.

File	Size	Bits per weight	Smaller than FP16
`Ternary-Bonsai-2-27B-PTQ1_0.gguf`, densely packed	5.95 GB	1.77	9.0x
`Ternary-Bonsai-2-27B-PQ2_0.gguf`, one trit per 2-bit slot, read by GPU and Metal kernels as is	7.21 GB	2.14	7.5x
`Ternary-Bonsai-2-27B-mmproj-Q8_0.gguf`, vision projector	629 MB
`Ternary-Bonsai-2-27B-mmproj-BF16.gguf`, vision projector, full precision	931 MB
`Bonsai-2-27B-DFlash2-Q8_0.gguf`, speculative drafter	2.06 GB
`SHA256SUMS`	66 KB


Context	262,144 tokens native; about three layers in four are linear attention
Inputs	text and images
Architecture	GGUF `general.architecture`: `qwen35`

Accuracy against full precision

Suite	FP16	A conventional 2-bit build (IQ2_XXS)	Zen 6 Flash (PQ2_0)	Retained
Average of 14 reasoning tests	86.33	72.59	84.78	98.2%
Math (MathVision, GSM8K)	97.10	79.40	96.57	99.5%
Code (HumanEval, LiveCode)	90.20	74.80	89.42	99.1%
Tool calling and schema adherence	76.80	58.10	74.92	97.6%
Footprint	53.79 GB	8.8 GB	7.21 GB	86.6% smaller

Measured

Apple Silicon M4 / M5 Max (Metal): 47.2 tok/s decode alone, 92.4 tok/s with the drafter; 7.84 GB including the vision projector and a 32K cache.
DGX Spark (CUDA 13.3): 170.04 tok/s decode with the drafter at 54.2% draft acceptance; 32,768 tokens across 4 parallel slots in under 12 GB.

Serve it

llama.cpp, with the vision projector and the drafter:

llama-server \
  -m Ternary-Bonsai-2-27B-PQ2_0.gguf \
  --mmproj Ternary-Bonsai-2-27B-mmproj-Q8_0.gguf \
  -md Bonsai-2-27B-DFlash2-Q8_0.gguf \
  --spec-draft-n-max 3 \
  -c 32768 \
  --port 8080

On a Mac, one prompt:

llama-cli \
  -m Ternary-Bonsai-2-27B-PQ2_0.gguf \
  --mmproj Ternary-Bonsai-2-27B-mmproj-Q8_0.gguf \
  -p "<image>\nDescribe the system architecture shown in this diagram in detail."

Citation

@misc{zenlm2026zen6flash,
  title  = {Zen 6 Flash},
  author = {Zen LM},
  year   = {2026},
  publisher = {Zoo Labs Foundation}
}

License & attribution

Apache-2.0; see LICENSE. Zen 6 Flash is Ternary Bonsai 2 27B by prism-ml (prism-ml/Ternary-Bonsai-2-27B-gguf, Apache-2.0), itself built from Qwen3.8-27B by the Qwen team (Apache-2.0), with the speculative drafter by ProCreations (ProCreations/Ternary-Bonsai-2-27B-DFlash2).