Temperature

At each step the model has a probability distribution over the next token. Temperature controls how closely sampling follows it. Near zero, the most likely token is chosen almost every time and the same prompt gives nearly the same answer. Higher, less likely tokens get a real chance.

Low for anything where you want the same answer twice — extraction, classification, code. Higher for anything where you want options. It is not a creativity dial so much as a willingness-to-be-wrong dial.