ChainThink reports that on August 1, OpenAI officially unveiled its next flagship model, Astra, which achieved new results on 10 long-standing unsolved problems in mathematics and theoretical computer science.
The core issues behind these questions have seen no progress for at least a decade, with many remaining stagnant for even longer. Among them, Astra has for the first time constructed a non-sofic group, resolving a central open problem in group theory;
Simultaneously disprove the Connes rigidity conjecture and solve three Erdős problems. Other results include new upper and lower bounds and hardness proofs for problems related to sphere packing, coding theory, quantum complexity, and post-quantum cryptography.
OpenAI stated that the tokens used by the model to find these 10 results are equivalent to approximately $2,000 at Sol API pricing.
The mathematical arguments were generated by Astra, subsequently refined by humans into a paper, and then converted by the model into Lean certificates for step-by-step computer verification of the derivations.