GPT-oss

GPT-oss brings OpenAI's open-source language models, the 120B and 20B variants, straight to developers at no cost. Built on a Mixture of Experts (MoE) architecture, these models are designed to run locally, so you can harness advanced AI without relying on external servers.

Say goodbye to sending sensitive data to the cloud. Because GPT-oss runs entirely on your own hardware, it takes a privacy-first approach by default, keeping your conversations and code where they belong: with you.

The models are flexible about hardware too. Whether you have access to powerful H100 GPUs or just everyday consumer hardware, you can get capable performance, and a 131k token context window lets you work with long documents and extended conversations.

From natural conversations and code generation to complex reasoning, GPT-oss aims to put advanced AI capabilities into the hands of the open-source community, making sophisticated language models accessible to anyone who wants to build with them.