Home Glossary Temperature

Temperature - Page 9

Temperature is a generation setting that changes how strongly an AI model favors its highest-probability next tokens. Lower values concentrate probability on likely choices and often produce more consistent output, while higher values spread probability across alternatives and can increase variation. Temperature does not directly control truthfulness, creativity, or intelligence, and its effect depends on the model and other sampling settings. A very low value may still produce an error, while a high value can sometimes be appropriate for brainstorming. Applications choose and test the setting according to the task, balancing repeatability, diversity, formatting reliability, and risk. Some model interfaces fix or reinterpret temperature, so behavior should be verified rather than assumed.