Temperature is a setting in language model configuration that determines how predictably or creatively the model generates text. A low temperature (close to 0) makes outputs more deterministic and focused — the model reliably chooses the most probable next token. A high temperature (closer to 1 or above) introduces more randomness, producing more varied, creative, or unexpected responses. For factual tasks like data extraction or classification, low temperatures are preferred; for creative writing or brainstorming, higher temperatures are appropriate.