The OCR stuff is quite usable. Their models perform good at some very basic tasks, so I use them to diversify. But their current model lineup is really terrible.