Ask an AI the same question twice and you may get two different answers. Temperature is the dial that controls how much they vary.
A pianist can play a piece exactly as written, note for note. Or they can improvise, taking the tune somewhere new. Same piano, different freedom.
An AI has a dial for that freedom. Temperature controls how often it picks less likely words. Low temperature: it nearly always picks the safest next word. High temperature: it takes more chances.
What changes
- low: steady, repeatable, a bit plain. Good for facts, summaries and code
- medium: natural and varied. Most chat apps sit here
- high: surprising and creative, sometimes nonsense. Good for brainstorming
Why the odd name
It’s borrowed from physics. In hot things, particles move more wildly. A “hotter” AI picks words more wildly.
What it isn’t
It isn’t a truth dial. Low temperature makes answers more consistent, not more correct. A confident wrong answer can come out the same way every time.
Where the comparison breaks down
A jazz pianist improvises with skill and taste. A high-temperature AI is just more random, which can produce gibberish.
How the words fit together
An LLM writes one token at a time. Temperature decides how adventurous each choice is. Too high can make hallucinations more likely.
Most chat apps choose the temperature for you. If you can set it, go low for facts and summaries and higher for brainstorming or stories.

When you'll hear it
“Turn the temperature down.” Make the AI's answers more predictable and consistent.
“It's at temperature zero.” The AI picks the most likely words every time, so answers barely vary.
“Raise it for more variety.” Let the AI take more chances, so ideas are more varied.
“For this task, do we want the AI to be consistent every time, or to come up with varied ideas?”


