Glossary
Temperature
A parameter controlling how deterministic or creative an LLM's output is.
Definition
Temperature is a decoding parameter (typically 0.0 to 1.0 or higher) that controls how much randomness the model uses when selecting the next token. Temperature 0 = deterministic, always picks the highest-probability next token. Higher temperatures = more variation and creativity.
Context
For enterprise use cases requiring consistent output (extraction, classification, structured responses), temperature 0 or near-zero is standard. For creative content generation, temperatures 0.7–1.0 are common. Temperature is not a silver bullet for consistency — for that, structured output formats and evaluation harnesses matter more.