Temperature
A parameter controlling how random the model's next-token choice is — higher temperature produces more varied output, lower produces more deterministic output.
Temperature is a number (typically between 0 and 1, sometimes higher) that adjusts how the model samples its next token. At temperature 0, the model always picks the single most-likely next token — same input, same output, every time. At temperature 1, the model samples proportionally to the probability of each candidate, producing more varied and creative output.
Practical guidance: use lower temperature for tasks where correctness matters (code, factual answers, data extraction) and higher temperature for tasks where variety matters (brainstorming, creative writing, exploring multiple angles). The default in most APIs is around 0.7-1.0, which is a reasonable middle for general chat.
Consumer chat products (ChatGPT, Claude.ai, LADLE) usually don't expose temperature as a user setting. The defaults are tuned per-provider for general use, and letting users tweak it produces confused feedback ("my answer was wrong" is different from "my answer was creative"). For power users who need control, use the API directly.