it's also 50% cheaper now, and there's no need to manage three different models (not to mention that transcription is lossy and doesn't capture emotion, tone, inflection, and other parameters that a multimodel model can)