Does Your Reasoning Model Implicitly Know When to Stop Thinking?