Microsoft releases MAI-Transcribe-2-Streaming, a real-time speech model, plus two new MAI voice models

Microsoft/MAI Superintelligence Models

Microsoft launched MAI-Transcribe-2-Streaming, a speech-to-text model that transcribes as people talk, alongside MAI-Voice-2.1 and MAI-Voice-2.1-Flash for generating speech. Microsoft says the transcription model ranks first on accuracy at Artificial Analysis for both final and partial text. The voice models add cloning from a few seconds of reference audio, with consent guardrails built in.

• MAI-Transcribe-2-Streaming covers 60 languages and detects the spoken language automatically and continuously.
• It shows its first partial words after just over 100 ms of audio; Microsoft says words appear about twice as fast as with its closest competitor in its internal tests.
• Introductory price: $0.54 per hour of audio through the end of 2026.
• MAI-Voice-2.1 speaks 23 languages and 26 locales in one voice with a native accent, priced at $22 per million characters.
• MAI-Voice-2.1-Flash targets high-volume use: about 150 ms end-to-end latency, roughly 60% lower cost than comparable models per Microsoft, priced at $15 per million characters.