Recent models have had a worrying lean towards longer output. If you're going to serve up some AI slop, at least distil it to the smallest possible footprint possible.