Part 3/12:
Behind the scenes, these issues stemmed from a systemic failure in the auto-routing infrastructure—a crucial component responsible for directing queries to the appropriate models. Sam Altman, OpenAI’s CEO, admitted that the system broke on launch day; complex questions were routed to the basic models, while simpler queries sometimes fed into the more advanced, but sluggish and overly formal, heavyweight systems. This resulted in a perception that GPT-5 was underperforming, primarily because many users experienced interactions with less capable or misrouted models.