OpenAI's new model, Astra, solves 10 long-standing math problems

According to Beating Monitor, OpenAI has officially announced its next flagship model, Astra, for the first time. The internal version of Astra has achieved new results on 10 long-standing unsolved problems in mathematics and theoretical computer science. The core conclusions of these problems had seen no progress for at least a decade, with many remaining stagnant for much longer. Some of these problems have been directly solved or refuted. Astra has constructed a non-sofic group for the first time, resolving a central open question in group theory; it has also disproven the Connes rigidity conjecture and solved three Erdős problems. Additional achievements include new upper and lower bounds and hardness proofs for problems related to sphere packing, coding theory, quantum complexity, and post-quantum cryptography. OpenAI states that the tokens used by the model to discover these 10 results amount to approximately $2,000 when converted at Sol API pricing. All mathematical proofs were generated by Astra; humans subsequently assisted in organizing them into papers, and the model converted each proof into a Lean certificate to enable automated, step-by-step verification of logical validity. OpenAI has publicly released the papers, proof certificates, and the model’s reasoning process.