Reasoning effort
Also called effort level, thinking level
A setting for how much a model thinks before answering: more for hard problems, less for quick ones, trading speed and allowance for depth.
Thinking modes used to be on or off. Now the usual choice is how much. Reasoning effort sets the budget a model gets for working before it replies: a little for a quick check, a lot for a hard proof or a tangled bug.
The same model can behave quite differently across settings. Low effort is fast and saves your allowance; high effort is slower, uses more of your limit and helps most when a problem has many steps that all have to come out right. Picking the setting is now part of asking well.
For example
Checking the logic of a contract clause deserves high effort. Turning 3 bullet points into a friendly email doesn't.
Put it to use
Where you'll meet reasoning effort in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.