What are max tokens?
Max tokens is the cap on how long the model's answer may be, counted in tokens. Hit the cap and the answer stops mid-sentence. It is not the same as the context window, which is how much text the model can see in total.
Also called: max output tokens
In plain English
If you ask for a full proposal and the cap is tiny, you get a heading and half a paragraph. If the cap is huge, a simple text becomes an essay and spends more of the plan. Some apps hide this behind 'shorter' and 'longer.' Others show a number. Either way, the cutoff is a setting, not the model running out of ideas. If replies keep stopping mid-sentence, the cap is too low for that job. If replies ramble and the bill climbs, the cap is a kindness. Set it a little above the length you actually want, and say the length in the prompt too: 'eight lines.' A cap is a ceiling, not a style.
How an owner would use it
Ask for the length in the prompt first ('eight lines'). Touch a max token setting only when answers are getting chopped. Leave room for the whole answer, not a novel.
Common mistakes
Treating max tokens as a quality setting is the miss. A higher cap does not make a truer answer. It makes a longer one, until it stops. Do not set the cap so low that phone numbers and times are the part that gets clipped. Do not set it to the maximum on every call and then wonder why the bill moved.
A worked example
A florist
The delivery window got cut off
This is an illustration, not a customer result. A wedding reply kept ending before the dock time. The cap was stopping the draft mid-message. The owner raised the cap enough for one complete reply and kept the prompt to the date, the venue door, and the flower list they typed. The model did not get to add a rush fee.
Related terms
Prompts and guides
Questions owners ask
What are max tokens?
Max tokens is the limit on the length of the answer. When the answer reaches it, the model stops, even if the sentence is unfinished.
How do I use max tokens on customer messages?
Say the length you want in normal words. If the draft is cut off, raise the cap slightly or ask for fewer sections. Check that the time, the price, and the ask survived.
Is the context window the same number?
No. The context window includes your prompt, attachments, history, and the answer. Max tokens usually caps only the answer. A large window can still chop the reply if the answer cap is small.
Product guides are on AI apps for business.