OpenAI is preparing to launch a brand new model called "Astra."

Three sources familiar with the related planning revealed that OpenAI is preparing to launch a brand new model series tentatively named "Astra," which enhances the model's ability to handle long-term complex tasks. The sources stated that OpenAI CEO Sam Altman showcased the Astra model to policymakers and regulators in Washington this week. OpenAI emphasized a key capability of this model: supporting multiple agents to collaborate over the long term to tackle high-difficulty problems, which can be applied to large project advancement or solving complex higher mathematics issues. Astra will become part of OpenAI's new series of models, alongside the Sol, Terra, and Luna models. The sources did not disclose when OpenAI plans to officially release the model. OpenAI has yet to finalize Astra's positioning: whether to name it GPT-6 or to classify it as a derivative version of the GPT-5 series. One source indicated that this batch of models has already entered the testing phase and is planned to first comply with the new regulatory framework proposed by the Trump administrationAI models must be submitted to the federal government for review prior to being released to the public. The U.S. government aims to finalize this regulatory framework by the weekend deadline. The sources mentioned that in the short term, OpenAI also plans to release a report disclosing details of how its top AI tackled ten previously unresolved mathematical problems, in order to showcase the performance of its strongest artificial intelligence model.