Why is the CLI-Whisper so much slower than using the library? And how does CLI return data while running?

When using the transcribe-method in the Whisper library it only returns when the whole transcription is finished. Meanwhile, if I use the Whisper command line client it returns sentence by sentence. I thought the CLI-client used the library? How does the CLI-client achieve this continuous progress?

OTOH, with the same settings (just specifying the language), the CLI-client is much slower, ≈ about 1/3 slower (not sure about the quality), in a sample file 7,3x vs 5,2x on my rig. Why is that? Isn’t it (again) using the library in the background?