PANews reported on April 1 that according to IT Home, OpenAI CEO Sam Altman officially announced today that a "powerful new open weight language model" with inference capabilities will be released in the coming months. So far, users can only use its models through OpenAI's interface or integrate its API key into their applications, which means that the model itself remains secret and proprietary. However, open weight models are different. The pre-trained parameters in these models (the "weights" that define how the model responds) are openly shared with the public. Developers and researchers can download them, run them locally, fine-tune them for specific use cases, or insert them into custom applications.

It is worth noting that just because the weights are public does not mean they are open source, the training code, datasets, and detailed logs may still remain proprietary or undisclosed, and therefore may not be fully open source.