Temperature controls randomness in model output sampling.

Lower temperature makes outputs more deterministic, while higher temperature increases variety and risk of unexpected answers.