GPT-oss

Running on your own machine is the main point of GPT-oss, a set of open-source language models from OpenAI that developers can download and use for free.

Because the models run locally, your code and prompts stay on your hardware.

GPT-oss is an open-source AI model family meant for coding and development work, and it supports integration and fine-tuning if you want to adapt it to your own projects.

The context window is 131k tokens, which is enough to feed in a large codebase at once.

The models use a mixture-of-experts (MoE) architecture for efficiency, and hardware needs are flexible: an H100 works, and so does consumer hardware.

Benchmark results are strong, and in some areas they surpass GPT-4.

Full deployment needs your own hardware, and setup and customization can call for technical skill.

If you just want to try it first, an interactive demo and an AI assistant are available without signing up.