The term can be confusing. What does it mean for a model to be in “reasoning mode”?
On e.g. ChatGPT, you would see that you can select whether to use “Thinking” mode. For practical purposes, that means the model will use more effort to answer your requests. This means instead of directly answering your request, the model may first eg breakdown the problem, check its assumptions, validate its claims, then provide the final answer. In general, model answer quality is higher with thinking than without.
Some AI tools offer different "thinking efforts". Same idea: higher effort = model spends longer time to answer. E.g. ChatGPT:
