With the large models capable of generating up to 50k words in a single response, you are probably going to need to change the way you deal with responses. the site is regularly crashing in the middle of processing large ones, and because of the way the site saves the data, it looses it. because the data has to make it to the browser, and be processed on the computer instead of your servers. when the page memory gets too high, which it easy with a large response , the response is lost. i have had response with 30k words, that only 5k were saved.