Over the past year, it has become almost a ritual: every few weeks, a new announcement surfaces claiming that a large language model (LLM) has “solved an International Mathematical Olympiad (IMO) problem”.