Microsoft ships MAI-Transcribe-2: faster, cheaper speech-to-text at $0.10/hour
Microsoft's new MAI-Transcribe-2 claims the crown in speech recognition, beating Gemini 3.5 Transcribe, GPT-Transcribe, and Whisper V3-Large on word error rate at $0.10 per hour of audio. It adds diarization, configurable transcription styles, and word-level timestamps, and tops FLEURS across 60 languages with a 5.2% average WER. At that price, the incumbent transcription API math gets awkward fast.