Temperature

Temperature is a generation setting that changes how strongly token sampling favors higher-probability options.

Plan for this page

Understand one concept in plain language and connect it to nearby ideas.

You will leave with: Lower values usually concentrate choices around more likely continuations, while higher values usually spread probability across more alternatives. The exact effect depends on implementation and interaction with other sampling settings.
Time: 3–8 min
Before you begin: No prerequisite. Keep one real example in mind.
Do this now: Read the explanation, then compare the example with the common misunderstanding.

Prepared next stepExperiment with generation settings

TemperatureTemperature is a generation setting that changes how strongly token sampling favors higher-probability options.

Prediction
Generative AI
Large language model

Plain explanation

Lower values usually concentrate choices around more likely continuations, while higher values usually spread probability across more alternatives. The exact effect depends on implementation and interaction with other sampling settings.

Example

A drafting tool may use a higher temperature for varied slogans and a lower one for a more repeatable structured extraction.

Why it matters

Temperature changes variation, not evidence or correctness. A low value can repeatedly produce the same unsupported claim, and a high value does not create genuine creativity or understanding.

Common misunderstanding

Misconception: temperature measures factual confidence. Correction: it mainly changes the sampling distribution used to select output tokens.

Related concepts

Follow relationships, not only the alphabet

PredictionA prediction is a model’s estimated output for a given input, including a class, value, probability, sequence, or next element.Generative AIGenerative AI produces new text, images, audio, video, code, or other content in response to context and controls.Large language modelA large language model, or LLM, is a parameter-rich model trained on language sequences to predict and generate tokens in context.

Experiment with generation settings

Sources and scope

What supports this object

Machine Learning GlossaryGoogle for Developers

Defines temperature in language-model sampling; scale, supported ranges and determinism vary by model and API.

Open source

Loading page content...

…