Vercel AI Gateway adds Microsoft MAI voice and transcription models

Vercel and Microsoft AI (MAI) have partnered to make MAI models available on AI Gateway, positioning it as one of a limited number of platforms offering access to these models. The collaboration emphasizes security and zero data retention (ZDR), aligning with Vercel’s goal of providing users control over their data and transparency regarding training usage.

New Voice and Transcription Capabilities

The update introduces three specific models focused on audio generation and real-time transcription:

• MAI-Voice-2.1 (microsoft/mai-voice-2.1): Generates expressive speech in 23 languages with consistent speaker identity across longer passages, suitable for narration, audiobooks, and podcasts.
• MAI-Voice-2.1-Flash (microsoft/mai-voice-2.1-flash): Offers the same multilingual speech generation with lower latency, designed for voice agents, assistants, and quick spoken replies.
• MAI-Transcribe-2 Streaming (microsoft/mai-transcribe-2-streaming): Returns partial transcripts as audio arrives, enabling live captions and real-time conversation tracking without waiting for full recordings.

Integration and Pricing

Developers can access these models using AI SDK 7. For speech generation, the generateSpeech function supports MAI-Voice-2.1-Flash for immediate responses or microsoft/mai-voice-2.1 for longer audio. Live transcription is handled via streamTranscribe, which processes streams of 16 kHz, 16-bit PCM audio and updates partial transcripts dynamically.

AI Gateway bills these models at their listed rates with no platform fee or markup on inference. The gateway also provides a unified API for tracking usage, cost, and request traces, while supporting routing, retries, and failover across providers.