ChatGPT operates more like a search engine than a groundbreaking AI. At its core, it processes user queries through a series of models, leading to a final answer generated by GPT 5.2. The system's first step involves a small classification model called snc-pg-sw-3cls-ev3, which quickly evaluates the query to determine if it can be answered using its existing knowledge, a simple search, or a complex search requiring further information. This model returns three probability scores that inform the next steps, with thresholds set for each type. If the query is determined to require a search, it moves to a model referred to as “Thinky.” This model is tasked with generating search queries and filtering results, either through traditional keyword searches or semantic searches that better capture user intent. Thinky is designed to be fast and efficient, ensuring that responses are delivered without frustrating delays. The article emphasizes that while GPT 5.2 appears to be the star of the show, it’s largely a synthesizer of information gathered by these earlier models. The entire process is optimized to minimize costs and maximize speed, highlighting the intricacies of what goes on behind the scenes when users interact with ChatGPT.