For the first time in over five years since the launch of GPT-2, OpenAI is introducing two open-weight AI reasoning models, available for free download on Hugging Face.

Referred to as “state-of-the-art,” these new models from OpenAI were revealed Tuesday in a blog post from the company. OpenAI asserts that they “exceed similarly sized open models in reasoning, excel at tool utilization, and are tailored for efficient deployment on consumer hardware.”

There are two models available:
– **gpt-oss-120b**, the more substantial and robust model, which is capable of running on a single Nvidia GPU.
– **gpt-oss-20b**, a more lightweight version aimed at functioning on consumer laptops with just 16GB of RAM.

To explore the new OpenAI models, head over to the OpenAI gpt-oss page.

### What does open-weight signify?

This release showcases OpenAI’s initial dedication to open-sourcing its models. While not “entirely open source” — as the training data is not included — they are open weight, indicating that the code and model parameters are accessible for anyone to use, modify, and build upon.

Models like Meta’s LLaMA also aren’t genuinely open source since the open-source community demands access to training data.

Since the introduction of GPT-2, OpenAI has shifted toward a more closed-off strategy regarding LLM development — until this moment. The launch of these open-weight models signifies a pivotal change in direction.

With China’s DeepSeek AI and other labs gaining momentum, U.S. tech firms are under pressure to remain competitive. Last month, the Trump administration encouraged American AI developers to open source more technology to foster innovation that reflects “American values” and maintain a strategic advantage.

This initiative is a notable advancement for OpenAI and the broader open AI ecosystem.

*Disclosure: Ziff Davis, the parent company of Mashable, filed a lawsuit against OpenAI in April, claiming it infringed Ziff Davis copyrights in training and running its AI systems.*