OpenAI has unveiled GPT-6 Astra, marking a significant leap forward in tackling the bounded prime gap problem. The new achievement reduces the known upper bound from 246 to 186 and includes a comprehensive Lean formal verification. This breakthrough was announced by Su Weijie, a distinguished alumnus of the Peking University Mathematics Department, Class of 2007, and currently a researcher at OpenAI. The model attained this feat by delving into novel factorization conditions to refine the sieve method.

On the same day, Axiom reported progress in pushing the upper bound down to 212. Meanwhile, Anthropic revealed that its Claude team had accomplished a formal proof establishing an upper bound of 188. Prior to these developments, mathematician Julia Stadlmann had already made strides by lowering the upper bound from 246 to 240.

The evolution of AI from merely solving mathematical problems to actively engaging in mathematical research underscores a transformative shift. This trend is poised to become a pivotal arena of competition in the upcoming phase of large-scale model development.