Effort
Effort is how much thinking the model does before answering.
More effort spends more output tokens working the problem, and gets better results on hard ones.
On easy work it is waste. On something genuinely difficult it is the difference between a usable answer and a confident wrong one.
Was this page helpful?