Extended Thinking

What is Extended Thinking

Extended thinking is a mode where the model spends additional, visible reasoning steps working through a problem before committing to a final answer. Instead of producing a response in one pass, it lays out intermediate reasoning first, checking its own logic, weighing alternatives, catching an inconsistency before it ever reaches the final output.

This tends to pay off specifically on harder problems: a tricky bug with a non-obvious cause, a design decision with real tradeoffs, a multi-step task where an early wrong turn would otherwise go unnoticed until much later. On a simple, well-understood question, the extra deliberation mostly adds time without changing the answer.

Extended thinking is closely related to reasoning effort and chain of thought but is its own distinct switch: reasoning effort is the general dial for how much internal work goes into an answer, chain of thought describes the shape that reasoning takes once made explicit, and extended thinking is the specific mode that surfaces a visible version of that deliberation for genuinely hard tasks.