What is LLM temperature? Definition and settings
Temperature is a generation setting that controls how much randomness goes into a language model's choice of tokens. A low value pushes the model toward the most likely tokens and makes answers more consistent. A higher value leaves more room for less likely choices, which produces more varied answers. Depending on the vendor, the range runs for example from 0 to 1 or from 0 to 2.
Definition written by Alex Dionisio, Senior Product Manager. Updated November 13, 2026.
Why it matters for a PM
Temperature is one of the few settings a product team chooses directly, and it shapes the experience. Data extraction or classification calls for consistent answers; help with brainstorming taglines benefits from some variety. A PM should know the setting exists, that it does not fix errors, and that some recent models restrict or ignore it.
Example
An app suggests headlines for real estate listings. At a low temperature, the five suggestions look alike and users keep hitting regenerate. The team raises it for that feature only. For pulling floor area and price out of the listing text, it keeps the temperature low, since variety there only adds mistakes.
Key points
- Low temperature for tasks with one right answer: extraction, classification, format conversion.
- Higher temperature for creative work: ideas, rewrites, first drafts.
- Even at zero, two identical calls do not always return exactly the same answer.
- Test the setting on a set of cases, with the model actually used in production.
Common mistakes
- Lowering the temperature to cure hallucinations, which only makes the model wrong more consistently.
- Keeping the same value after switching models without checking results again.
- Raising the temperature for creativity on a task that requires accuracy.
Go further with Module
The courses and lessons that cover this concept:
Guides on this topic
- RAG vs fine-tuning: how to choose
What each approach solves, the decision criteria and how to evaluate, with a decision grid.
- Evaluating an AI feature (evals)
Test cases, criteria, LLM-as-a-judge, release thresholds and monitoring, with an evals plan.