Temperature is a setting that controls how much randomness an AI model uses when choosing each next word. A low temperature makes it pick the most likely option almost every time, giving steady, repeatable answers. A higher temperature allows more variety and surprise.
At each step the model scores many possible next tokens. Temperature reshapes those scores before one is chosen. Near zero, the top choice nearly always wins. Higher values give less likely choices a better chance, which suits brainstorming but increases the chance of drifting off course.
Operators rarely set it themselves, but teams building assistants do. A customer-service or ordering assistant that quotes policies and prices should run at a low temperature for consistent, accurate replies. A tool that drafts social posts or names for a new menu item can run higher to give fresher ideas. Some newer models fix or limit this setting.
Low temperature does not eliminate errors, and it does not make answers identical every time on all systems. Higher temperature raises the risk of hallucination, so avoid it for anything factual. Test settings with your real questions rather than guessing.
Knowing this one dial helps explain why the same AI can be dependable in one job and wildly inventive in another.