Transcription time estimator

Expected wall-clock to transcribe on-device using WhisperKit Large v3 Turbo (or SpeechAnalyzer on macOS 26+, which is ~40% faster again). Numbers are for steady-state thermal-nominal conditions.

Chip	15 min source	30 min source	60 min source	120 min source	240 min source	Throughput
M1 / M1 Pro	2s	4s	8s	15s	30s	8×
M1 Max / Ultra	1s	3s	5s	11s	22s	11×
M2 / M2 Pro	1s	2s	5s	9s	18s	13×
M2 Max / Ultra	1s	2s	4s	8s	15s	16×
M3 / M3 Pro	1s	2s	4s	7s	14s	17×
M3 Max	1s	1s	3s	5s	11s	22×
M4 / M4 Pro	1s	1s	2s	5s	10s	25×
M4 Max	0s	1s	2s	4s	8s	32×
M5 Pro	0s	1s	2s	3s	6s	40×

Reading this table: a 60-minute podcast on an M2 Pro takes about 4:37 because the chip transcribes 13 hours of audio in one real-time hour.

SpeechAnalyzer bonus: on macOS 26+, enable the system framework and expect about 40% less. An M4 Max that runs WhisperKit at 32× effectively runs SpeechAnalyzer at ~45×.

Thermal note: fanless Macs (MacBook Air, iPad Pro) throttle after sustained transcription. A 4-hour podcast on an M1 Air will take longer than the table suggests once the chip hits 80°C.

Want the full cost comparison vs cloud transcription APIs? Read Why we picked WhisperKit over OpenAI's API.