OpenAI has launched two new AI models – gpt-oss-120b and gpt-oss-20b – that are open-source and free for anyone to use. These models are designed to help developers build powerful AI tools while keeping costs low.
What is an Open-Weight Model?
An open-weight model means the trained data of the AI (called model weights) is shared publicly. This allows developers and researchers to download, run, customize, and improve the models on their own systems.
Sam Altman’s Announcement
OpenAI CEO Sam Altman shared the news on X (formerly Twitter), saying:
“gpt-oss is out! We made an open model that works like o4-mini and can run on a high-end laptop. There’s also a smaller one that works on a phone. Super proud of the team!”
He added that this release is meant to give people more control over AI, especially those who want to build or modify their own tools. He also highlighted the privacy benefits of using these models locally.
Read more: Think ChatGPT Is Safe for Personal Talk? Even Sam Altman Says Be Careful
Altman said these models will boost innovation in the AI field, helping more people to work on new ideas and build unique products.
Key Features of the New Models
| Model | gpt-oss-120b | gpt-oss-20b |
|---|---|---|
| Size | 117 billion parameters | 21 billion parameters |
| Device Support | 80GB GPU | 16GB memory devices |
| Performance | Similar to o4-mini model | Lightweight, runs on phones |
| Tasks | Chatbots, coding tools, reasoning, tool use | |
| License | Apache 2.0 (free to use and modify) |
Both models use Transformer architecture and advanced memory-saving techniques. They were trained using large datasets related to coding, science, general knowledge, and STEM subjects.
Safety and Testing
OpenAI said these models went through safety training and testing. The company also tested an enhanced version of gpt-oss-120b to make sure it follows its safety rules. They partnered with organizations like AI Sweden and Orange to test real-world uses, such as private and secure deployments.
OpenAI’s new models are powerful, free, and easy to run – even on laptops and phones. They’re made for developers, researchers, and innovators who want more freedom, better privacy, and lower costs while building with AI.
