A model stops mid-sentence well before it has finished answering. Which setting is the first thing to check?

The principle — an answer cut off mid-sentence is usually a budget being hit, not the model losing its way.

Why the key is correct — Google describes the maximum output tokens setting as specifying the maximum number of tokens that can be generated in the response, and gives the rough conversion: a token is about four characters, and 100 tokens is roughly 60 to 80 words. That makes the ceiling easy to sanity-check against the length you actually need.

Why the others are wrong — temperature changes which words are chosen rather than how many are allowed. Prompt length does not shorten the reply one for one. And examples consume prompt space rather than the response allowance.

Remember this — estimate the words you need, convert roughly, and check the ceiling before rewriting the prompt.

Sources — Google's Prompt design strategies.