Temperature
A setting that controls how predictable or varied an AI model's output is, with guidance on low, medium and high values.
Temperature: a setting, available in some AI tools and APIs, that controls how predictable or varied a model’s output is. Lower values give more consistent answers, and higher values give more varied and creative ones.
In plain English
When a language model chooses the next word, it weighs many options. A low temperature makes it favour the most likely choice every time. A higher temperature lets it pick less likely words more often, so the output changes more from run to run.
Typical guidance
| Setting | Good for |
|---|---|
| Low (near 0) | Facts, data extraction, classification, code and anything that must be consistent |
| Medium | General writing and everyday answers |
| High | Brainstorming, names, slogans and creative writing |
The exact range and defaults depend on the model and provider.
Things to know
- Chat apps often hide the setting, and some newer models fix or limit it.
- Even at low temperature, outputs can still vary and can still be wrong. See hallucination.
- A clearer prompt usually improves consistency more than changing temperature.
Common confusion
- Temperature does not make the AI “smarter” or more accurate. It only changes how much randomness it uses.
- It is not related to the computer’s physical temperature.