At a secret meeting in 2025, some of the world’s leading mathematicians gathered to test OpenAI’s newest large language model, o4-mini.

Experts at the meeting were amazed by how much the model’s responses sounded like a real mathematician when delivering a complex proof.

Share.