After trying a dozen or more paid and open source, I keep going back to MacWhisper. The dictation feature is in advanced beta but works well. The only thing I want it doesn't have is to have different models chosen for different tasks at the same time: one model for each drop folder, a different one for dictation and then another general-purpose one for drag-and-drop. I have the memory for it and MacWhisper can flush a model after a certain amount of unuse time anyway.

Right now it's dictation-first vs MacWhisper's file transcription focus. Planning to add meeting recording and batch processing eventually.

Heard some feedback about reliability issues with MacWhisper as well - trying to build something more stable from the ground up.

it's been a PITA though, lots of features not working, files disappearing and the developer blaming the OS (!), and separate payment for the iOS app.