Temperature
The setting that decides how much the model is allowed to stray from its most likely next word. Zero gives the steadiest answer; higher values give a different one each time.
At every step the model holds a ranked list of candidate tokens. Temperature flattens or sharpens that ranking before one is picked: low values keep the top candidate almost always, high values give the ones below it a real chance.
The range differs between APIs, so a value copied from one vendor's example does not mean the same thing at another. What carries across is the direction and the purpose: anything a program will parse, compare, or bill against wants the steadiest setting, and prose that should not read identically twice can afford more.
It is not a truth dial. A low temperature makes an answer repeatable, not correct — the model will state the same wrong thing with the same confidence every time.