Fable 5 doesn't wait to be told what to do. It researches, makes decisions, and goes deep before reporting back — consuming dramatically more tokens than GPT-5.5 or Opus in the process. Scott called it the best model he'd used for large, complex refactoring.