Skip to content
AI Model Radar

Glossary

Temperature

Temperature is the dial that sets how adventurous a model's word choices are: low values make it predictable and repetitive, high values creative and more error-prone.

At every step a model holds a ranked list of possible next tokens. Temperature reshapes that list before one is picked: near zero, the top candidate almost always wins and the same question yields the same answer; higher, the runner-ups get their turn and the text loosens up — for better in a poem, for worse in a SQL query.

It is a setting of the runtime, not a property of the model. Every local runtime exposes it, usually somewhere between 0 and 1 by default; code and factual questions want it low, brainstorming wants it high. It often travels with a sibling setting that caps how many candidates are considered at all.

The trap to avoid: turning it up does not make a model smarter, and turning it down does not make it truthful — it only decides how much the model gambles on its second choices.

Where you see it on the radar

Every runtime we list exposes temperature; the models themselves do not carry one. It is your dial — which is why the radar measures memory and never claims to measure creativity.

Pick a runtime and find the dial

Related terms

TokenInferenceLocal Runtime

Also in:EspañolPortuguês

← All terms