A hidden feature in ChatGPT called "Reasoning Effort" allows users to control how "deep-thinking" the model is before giving a final answer. In advanced models like o1 and o3-mini, OpenAI introduced "reasoning chains," where models take extra seconds to consider, review steps, and correct errors. The "Reasoning Effort" setting lets users manually choose between multiple levels: Low, Medium, or High. Lowering the effort provides extremely fast answers similar to the GPT-4 rapid model, while raising it to maximum gives the full power of the model for complex programming and mathematical problems. To access the feature, users open a new conversation in ChatGPT on the iPhone app or web, select a model that supports advanced reasoning like o1, and find a new option next to the model name called "Reasoning." Clicking it brings up a drop-down menu with a slider to select the desired effort level. By default, the setting is "Auto," where the application guesses if a question requires in-depth thought, but manual control provides a much better experience.