Temperature

A sampling setting that controls how randomly a model picks its next token: lower values make output more predictable, higher values more varied. Many products choose the value for you and never expose it. Temperature shapes variety — never truth or accuracy.

Learn it in

← All terms