MAI-Transcribe-2; fastest, most accurate, and cheapest speech recognition model | Dark Hacker News