I moreso run other small special purpose models like Whisper, SAM, Matcha, CLIP etc. and then do contextual correction passes with models like this.

Think almost like unix pipelines, have used it for many workflows.