gpt-oss-20b-GGUF

gpt-oss-20b-GGUF

Runs OpenAI's open-weight gpt-oss-20b reasoning model on your own computer, packed as GGUF files by unsloth. Suits developers who want a local model with adjustable reasoning effort.

Made by unsloth 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 GGUF copy of OpenAI's gpt-oss-20b, the smaller open-weight model of the gpt-oss series, prepared by unsloth. It generates text, reasons step by step and can call functions. The model has about 21B parameters, of which 3.6B are active at a time. At 4-bit the weights come to about 10.5 GB.

Who it suits: Developers and hobbyists who want a capable reasoning model that runs locally instead of through a hosted service.

Industry | Who uses it | How they use it | Example
Software | Application developers | Run a local model behind an app that needs function calling | A coding helper that stays on the developer's machine
Research | ML researchers | Fine-tune the model for a specialized task | A tuned copy for one field's vocabulary
IT and operations | Teams with privacy needs | Keep prompts on their own hardware | An internal question-answering tool
Education | Teachers and students | Study how a reasoning model works with its full chain of thought | A classroom demo of step-by-step answers

If you only want a hosted chat, or need the larger 117B model, you can skip this one.

Where to use it:

• Windows, macOS or Linux with Python, for fetching the files
• Programs that read GGUF files, such as llama.cpp, which the card thanks for adding support
• Your own scripts and apps that need a local text model

Use case | Who | What they get back
Local chat and Q and A | Individuals | Answers produced on their own computer
Function calling | Developers | Structured calls to tools from the model
Fine-tuning | Researchers | A model adapted to a specific job

How to set it up:

• Python, to run the download tool
• Around 10.5 GB of free disk space for a 4-bit copy
• About 16 GB of memory, which the card gives as enough for the 20b model
• A program that can load GGUF files

1. Install the Hugging Face hub tools with pip.
2. Download the GGUF files with the hf command.
3. Open the files in your GGUF-capable program.

pip install -U huggingface_hub
hf download unsloth/gpt-oss-20b-GGUF

How to use it: The card says the model was trained on the harmony response format and does not work correctly without it, so make sure your program applies that format. You can set reasoning effort to low, medium or high depending on how much speed or depth you want.

Example: You type "Plan the steps to migrate a small database and list the risks" into a local chat window running this model. With reasoning effort set to high, the model works through the problem and returns a plan. Its full chain of thought exists too, but the card says it is not meant for end users.

For newcomers and experts: Start by downloading the files with the two commands above, then load them in a GGUF-capable app and try a short question at low reasoning effort. Raise the effort when answers feel shallow.

The weights are a quantized copy of openai/gpt-oss-20b, a mixture-of-experts model trained with native MXFP4 on its expert layers. The card says one variant, F32, is MXFP4 upcast to BF16 and is not quantized. Smaller quants trade some accuracy for lower memory use.

Download the GGUF files
Load them in a GGUF-capable program
Apply the harmony prompt format
Pick low, medium or high effort
Get a reasoned answer

Good to know:

• License: Apache 2.0
• Cost to run: none named, but you need your own hardware and about 10.5 GB of disk at 4-bit
• Data it reaches: downloads files from Hugging Face, then runs where you load it
• Not for: showing the raw chain of thought to end users, which the card advises against
• Not for: the larger gpt-oss-120b, which has its own model card

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 20.9B
• Weights: about 10.5 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 614,134
• Likes: 909
• License: apache-2.0
• Based on: openai/gpt-oss-20b
• Quantized versions: 3
• Fine-tunes: 1
• Tags: transformers, gguf, gpt_oss, text-generation, openai, unsloth, conversational

Links

• Model: https://huggingface.co/unsloth/gpt-oss-20b-GGUF
• Files: https://huggingface.co/unsloth/gpt-oss-20b-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download unsloth/gpt-oss-20b-GGUF

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