Temperature (AI)

Simple Definition

Temperature is a setting that controls how random or creative an AI model’s responses are.

• Low temperature (close to 0): responses are more focused, consistent, and predictable
• High temperature (close to 1 or above): responses are more varied, creative, and sometimes surprising

Think of it like a dial between “safe and precise” and “wild and creative.”

How Temperature Works

When an AI generates text, it assigns probabilities to potential next tokens (words or word-pieces). Temperature adjusts how those probabilities are used:

• Low temperature: strongly favors the most likely next token. The output is predictable and consistent.
• High temperature: spreads probability across more options. The output is more varied but may be less coherent.

Practical Guide

Most general-purpose AI tools default to around 0.7.

When You Can Set It

Consumer tools like ChatGPT don’t expose temperature as a user setting. You typically control it when using:

• The OpenAI, Anthropic, or Google AI APIs directly
• AI coding environments like LangChain or LlamaIndex
• Developer tools and custom AI applications

Related Terms

• LLM, the models temperature applies to
• Inference, the process where temperature affects output
• Prompt Engineering, working with model settings to get better outputs
• Token, the unit of text temperature affects at each generation step