Demo GPT-OSS with Ollama

This page has simple steps to help you try GPT-OSS with Ollama.

This page is intentionally as simple as possible to get our teammates up and running.

Install

Install ollama.

If you want to use curl:

curl -fsSL https://ollama.com/install.sh | sh

If you want to use macOS and brew:

brew install ollama

Serve

Start the server.

If you want to start ollama directly:

ollama serve

If you want to use macOS and brew to start the ollama server in the background:

brew services start ollama

Pull

Pull a GPT-OSS model, which means your system will download the model.

ollama pull gpt-oss:20b

Run

Run the model, which means the Ollama server will launch the model, and show you a prompt where you can type and interact with the model.

ollama run gpt-oss:20b

Bonus

If you have a powerful system, such as a computer with lots of RAM and many GPU cores, then you may want to experiment with the larger model.

Pull:

ollama pull gpt-oss:120b

Run:

ollama run gpt-oss:120b

Speed

On a MacBook M4 Max, the larger model works very well for personal chat interactions.

A simple request takes 2-4 seconds or so: "explain agile software development".

A longer request takes 10-20 seconds, and shows the chain of thought in progress: "explain agile software development (one paragraph introduction then bullet points) (between 220 and 250 words)".

Thanks

Thanks for all the developers who are creating this open source software.

Constructive suggestions about this page are welcome: joel@joelparkerhenderson.com.