In simple English
OpenAI has released two open-weight language models called gpt-oss-20b and gpt-oss-120b. You can download them and run them on your own hardware or server. They support reasoning, tool use, and python code execution. You need a powerful GPU with at least 16 GB of memory for the small model and an 80 GB GPU for the large model. They use a special format called harmony so they only work correctly with compatible chat templates.
What it is
Two open-weight language models with 21 billion and 117 billion parameters respectively, plus reference code for Python, Triton, and Metal. They come with built-in tools for web browsing and python code execution.
What it is good for
Solo builders who want to run powerful reasoning models locally or on a private server without sending data to external APIs.
- Available under the permissive Apache 2.0 licence for commercial use.
- Configurable reasoning effort lets you adjust how long the model thinks.
- Includes native tools for web browsing and python code execution.
- Requires an 80 GB GPU like an NVIDIA H100 to run the 120b model.
- Requires a specific chat format called harmony or the models will not work correctly.
- Reference implementations are for educational use and not optimised for general production.
Closed AI APIs
vllm
- mudler/LocalAI50k stars
Serves text, vision, speech, image and video models from one self-hosted service that imitates the paid interfaces.
- janhq/jan45k stars
Downloads and runs language models entirely offline as a desktop application, and opens a local server other programs can use.
- Mintplex-Labs/anything-llm67k stars
One program that reads your documents and answers questions about them, with the model running locally if you want.
pip install gpt-oss # or for torch support: pip install gpt-oss[torch]