Temperature is a dial that controls how adventurous a language model is when it picks each next word. A low temperature (near 0) makes the model pick the most likely, safe option every time, giving focused, predictable and repetitive output. A higher temperature lets it choose less-likely words more often, producing more varied, creative — and sometimes less accurate — results.
You will not always see a temperature slider, but it explains why the same prompt can give different answers, and why creative tasks (brainstorming, storytelling) benefit from higher settings while factual tasks (summarizing, extracting data) do better at low ones. Note that raising temperature can also make hallucinations more likely.