Comment by xena

The main limit is that you can have 128k tokens of input, but only 4k tokens of output per run. gpt4-32k lets you have up to 32k tokens of output per run. Some applications need that much output. Especially for token dense things like code and JSON.