Microsoft debuts MAI-Transcribe-2-Streaming at No. 1 on Artificial Analysis alongside new voice models

Microsoft has launched MAI-Transcribe-2-Streaming, its first streaming transcription model, which debuts at No. 1 on Artificial Analysis for accuracy in both final and partial transcripts. The model supports 60 languages with automatic continuous language detection and delivers words in transcripts 2x faster than its closest competitor in internal evaluations. It is available at an introductory price of $0.54 per hour of audio through the end of the year.

Alongside the transcription model, Microsoft announced two new voice models: MAI-Voice-2.1 and MAI-Voice-2.1-Flash. MAI-Voice-2.1 is a multilingual text-to-speech model supporting 23 languages and 26 locales with a single speaker identity that adapts accents naturally. It is priced at $22 per 1M characters. MAI-Voice-2.1-Flash is optimized for high-volume, latency-sensitive applications, generating 45s of audio with 150ms end-to-end latency. It offers 55% faster inference and ~60% lower cost than comparable models, priced at $15 per 1M characters. Both voice models support voice cloning with consent guardrails.

These models are designed to reduce latency in conversational AI loops. MAI-Voice-2.1 and MAI-Voice-2.1-Flash are available via OpenRouter, while all three models are accessible through Microsoft Foundry, MAI Playground, and Vercel. LiveKit support is coming soon. A new demo called 'Chatter' is available in the MAI Playground to showcase the models working together.