OpenAI is reportedly preparing to release a new open-weights language model, with a potential launch as soon as next week. This information comes from a report by Tom Warren in The Verge, citing sources familiar with OpenAI’s plans. The release would be the company’s first open-weights model since it launched GPT-2 in 2019.
Unlike OpenAI’s typical closed models, an open-weights model makes key components (known as weights) publicly available. This allows companies, researchers, and governments to run the model on their own infrastructure, independent of OpenAI’s servers.
According to the report, the new model is described by sources as being “similar to o3 mini” and possessing the advanced reasoning capabilities characteristic of OpenAI’s latest AI systems. In recent months, OpenAI has reportedly been demoing the model to developers and researchers, actively seeking feedback from the AI community.
The model is expected to be distributed on platforms like Hugging Face and various large cloud providers.
The report notes that the exact degree of openness will depend on the license OpenAI chooses. Whether the company provides full access to the model’s code and training data will determine if it can be considered truly open-source and be fully replicated by other researchers. While an imminent release is anticipated, the final date could change depending on development progress and server capacity.