When should I use a reasoning model in GPT instead of a faster one?

I use GPT for a mix of everyday writing, summarizing, and questions that involve several steps. A more capable reasoning model can be helpful, but it may take longer or use more resources than a quick response model.

How do you decide which type of model to choose for a task? I’m especially curious about practical signals—such as ambiguous instructions, multi-step analysis, or the need to check assumptions—that make the extra reasoning time worthwhile.