OpenAI has released two open-weight AI reasoning models, marking its first major open model launch since GPT-2. The models, gpt-oss-120b and gpt-oss-20b, are now available on Hugging Face under the Apache 2.0 license, allowing for commercial use without restrictions.
The larger 120b model runs on a single Nvidia GPU, while the lighter 20b version operates on laptops with 16GB of RAM. Both were trained using high-compute reinforcement learning and mixture-of-experts techniques to enhance reasoning while maintaining efficiency.
OpenAI says these text-only models outperform leading open-source rivals on benchmarks like Codeforces and Humanity’s Last Exam, though they lag behind the company’s proprietary o-series in hallucination rates and complex accuracy.
The move comes as OpenAI faces pressure from Chinese open-model competitors and renewed calls from the Trump administration to promote open AI aligned with U.S. values. OpenAI has not released training data due to ongoing copyright litigation concerns but claims extensive safety testing was conducted before launch.
With this release, OpenAI aims to reassert leadership in the open-source AI ecosystem while aligning with national policy goals and developer demand for transparent, high-performance AI infrastructure.
