Instructions to use SciTools/gpt-oss with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use SciTools/gpt-oss 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 SciTools/gpt-oss:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SciTools/gpt-oss:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf SciTools/gpt-oss:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf SciTools/gpt-oss:Q4_K_M

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 SciTools/gpt-oss:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf SciTools/gpt-oss:Q4_K_M

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 SciTools/gpt-oss:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf SciTools/gpt-oss:Q4_K_M

Use Docker

docker model run hf.co/SciTools/gpt-oss:Q4_K_M

LM Studio
Jan
Ollama
How to use SciTools/gpt-oss with Ollama:
```
ollama run hf.co/SciTools/gpt-oss:Q4_K_M
```
Unsloth Desktop

How to use SciTools/gpt-oss with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SciTools/gpt-oss: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": "SciTools/gpt-oss:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use SciTools/gpt-oss with Docker Model Runner:
```
docker model run hf.co/SciTools/gpt-oss:Q4_K_M
```

Lemonade

How to use SciTools/gpt-oss with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull SciTools/gpt-oss:Q4_K_M

Run and chat with the model

lemonade run user.gpt-oss-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use SciTools/gpt-oss with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SciTools/gpt-oss: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 SciTools/gpt-oss:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use SciTools/gpt-oss with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf SciTools/gpt-oss: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 "SciTools/gpt-oss: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"

OpenAI gpt-oss weights in GGUF format, for Understand's AI features, served locally by ullama (llama.cpp).

gpt-oss-20b-Q4_K_M.gguf — gpt-oss-20b as quantized by Unsloth, copied unmodified from unsloth/gpt-oss-20b-GGUF (SHA-256 c27536640e410032865dc68781d80a08b98f8db5e93575919af8ccc0568aeb4f). A mixture-of-experts model: 21B parameters, about 3.6B active per token, in an 11.6 GB download. Tested on an Apple M5 MacBook Pro, it qualified for Project Chat (answers match the code 0.663, reads before answering 0.807, follows instructions 0.850) and answered fastest of the qualified models of 4B or more (19 s median). It wrote all 40 test code summaries, with accuracy 0.703 and fact recall 0.500.

These results need low reasoning effort. ullama's ullama-models.conf sets reasoning_effort to low for gpt-oss-20b*. At the model's default (medium), 22 of the 40 summaries spent the whole 4,096-token budget reasoning and returned nothing. If you serve this file with another runtime, set low reasoning effort there.

The larger gpt-oss-120b (63 GB, in two parts) is in SciTools/OnBoard. It was tested at the default medium reasoning, and ullama leaves it there.

Ship and use exactly this file: verdicts do not carry across quantizations.

gpt-oss is released by OpenAI under the Apache 2.0 license and subject to the gpt-oss usage policy. This repository redistributes the weights unmodified apart from quantization; it is not endorsed by OpenAI.

Downloads last month: 29

GGUF

Model size

21B params

Architecture

gpt-oss

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for SciTools/gpt-oss

Base model

openai/gpt-oss-20b

Quantized

(245)

this model