Two new open-weight AI models. GPT-oss-120b and GPT-oss-20b. Those are models where some of the parameters around how they're trained are public and accessible So OpenAI is shifting its strategy today, making its tech more accessible than it's been in six years.

Because until now, you could only use OpenAI's models through the cloud or chat and web apps like ChatGPT. But with this release, developers can get open-weight models and build apps around them. So this is similar to what Meta Microsoft backed.

Mistral and China's DeepSeek have already done. A model's weights are the values inside the network that get set during training, so making them public means that developers can freely modify and run the AI on their own systems. It is not fully open source.

OpenAI Enters Open-Weight AI Race

OpenAI still is not sharing its training data or its entire code base. However, it is cheaper to operate and better suited to sensitive work that companies don't want running in the cloud.

Meta itself is rethinking how open its next generation will be, something Mark Zuckerberg suggested on last week's earnings call. As OpenAI moves in the exact opposite direction. So today's launch makes OpenAI pretty much the only US LLM builder that's actively leaning into a more open approach, aiming at growing its developer ecosystem while also going head-to-head with Chinese rivals like DeepSeek and K2.