What is Temperature?

A setting that controls randomness in AI output. Low temperature is focused and deterministic; high temperature is more creative and varied.

Temperature is a sampling setting that controls how random a model's output is. When a model generates text, it assigns probabilities to many possible next tokens. Low temperature (near 0) makes it pick the most likely token almost every time, giving consistent, repeatable answers. Higher temperature (around 0.8 to 1.0 or above) flattens the probabilities, so less likely tokens get chosen more often, producing varied and sometimes more creative output. It is usually exposed in APIs and developer playgrounds rather than consumer chat apps. Practical nuance: low temperature does not mean more accurate, it means more consistent; a model can be confidently wrong at temperature 0. Use low values for extraction, classification, and code, and higher values for brainstorming, naming, and creative writing where you want variety across runs.

Example

A developer building a tool that extracts invoice numbers from emails sets temperature to 0 so the same email always produces the same result. For a headline brainstorming tool, they set it to 0.9 to get ten different options per run.

Why it matters

If you use an AI tool via API or a playground, temperature is one of the few knobs that directly changes output behavior. Wrong settings cause flaky results in structured tasks or repetitive results in creative ones. Browse the AI tools directory or the model leaderboard to put it into practice.