The degree of variability in a model’s output when multiple valid responses are possible.
Randomness describes how much variation a model may introduce from one response to the next. Some tasks benefit from highly consistent outputs, while others benefit from creative variation.
In practice, randomness is often controlled through parameters such as temperature and sampling settings. Lower randomness is useful for factual, structured, or repeatable tasks; higher randomness can help with ideation and creative writing.
Managing randomness is a key part of prompt system design because it directly affects predictability, testing, and user trust.
Randomness is not the same as quality. More variation can produce novel ideas, but it can also reduce consistency and make debugging harder.