Temperature (LLM sampling)
Definition
Temperature is a generation setting that controls how randomly a language model samples next tokens—lower is more deterministic, higher is more varied.
By Vinespire Editorial Team, Editorial ·
This term is part of the full AI search glossary.
Full definition
When an LLM generates text, it assigns probabilities to possible next tokens. Temperature rescales those probabilities before sampling: values near zero favor the highest-probability tokens; higher values allow lower-probability tokens more often, increasing diversity and unpredictability.
Temperature is a decoding hyperparameter, not a measure of factual accuracy. Low temperature can still produce wrong answers confidently; high temperature can produce creative phrasing that is also wrong. Grounding and retrieval address truthfulness more directly than temperature alone.
In AI visibility testing, non-determinism (including temperature and other sampling settings) means repeated runs of the same prompt can yield different brand lists. Document settings when comparing engines or versions. See Prompt, Hallucination, and LLM.