Input / prompt tokens

The text you send in. Output / generated / completion tokens — the text the model produces. They cost different amounts of work.