OpenAI is set to be the next open-source AI brand as CEO Sam Altman confirmed on X on Monday that the company will soon release an “open-weight’ model that users will be able to run independently.
“We are excited to release a powerful new open-weight language model with reasoning in the coming months,” Altman said on a post on X.
The company is making this move to keep in step with the Chinese company DeepSeek, which took the AI industry by storm in January with its R1 reasoning model. Similarly, Meta’s Llama models have garnered attention among developer communities, Wired noted.
Altman’s announcement also follows a Reddit AMA in February where he stated that OpenAI was “on the wrong side of history” and that the company needed to reconfigure its open-source strategy.
He further detailed in his X post that the open-weight model has been a long thought-out project for OpenAI, and “now it feels important to do.”
Additionally, during the prior AMA session, OpenAI’s chief product officer, Kevin Weil said there was potential for the company to make its older, less cutting-edge models open-source. He didn’t have any specifics about which models could be used for an open source project. There is also the potential that OpenAI created a unique model to show its ability to train AI quickly and cheaply, similar to DeepSeek, Wired noted.
Altman also shared a link for developers to sign up and gain early access to the model, noting that those who sign up will have opportunities to attend events hosted by OpenAI and to get other early prototypes of the new model.
As we learn more about many AI models, it has become more evident that they are not 100% open source. While the code may be available on repositories, various training data and company details are concealed.
This evokes the title open-weight for the AI models, instead of open-source, which is being used by DeepSeek, Meta, and now OpenAI.