Microsoft has introduced three new AI models: MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash. The streaming transcription model processes speech via WebSocket, offering live updates and confirming final transcripts. It supports over 60 languages and averages a 320-millisecond response time for initial hypotheses.
Pricing for the new models is set at $0.54 per audio hour for the streaming transcription tool. The text-to-speech models, MAI-Voice-2.1 and MAI-Voice-2.1-Flash, are priced at $22 and $15 per million characters, respectively. Both voice models support 23 languages, with the Flash version optimized for speed and cost efficiency.
These releases align with Microsoft’s strategy to reduce reliance on external providers like OpenAI and Anthropic. CEO Mustafa Suleyman aims to lower operational costs by integrating the proprietary MAI model family into Copilot agents for Excel and Outlook. The company now utilizes a three-model architecture, including the Mai-Thinking-1 reasoning model, to manage voice agent workflows.
