Vibeleaderboard
← All Intel
Intel / article

Microsoft AI models are now available on AI Gateway

Source
vercel.com
Author
Kevin Dawkins
Date

Context

Microsoft's audio models add both speech generation and live transcription to AI Gateway. The release announcement distinguishes their jobs: MAI-Voice-2.1 and its Flash variant turn text into speech, while MAI-Transcribe-2 returns transcript updates as incoming audio is processed.

For live captions, those updates let an application display words before the recording ends. They remain provisional: the announcement instructs developers to replace displayed text when a revised partial transcript arrives. That behavior matters when designing a caption interface that follows an ongoing conversation.

Key takeaways · AI-distilled
  • Vercel's AI Gateway now serves three Microsoft AI audio models: MAI-Voice-2.1 for expressive speech in 23 languages that keeps a consistent speaker over long passages, a lower-latency Flash variant, and MAI-Transcribe-2 Streaming.
  • Vercel pitches MAI-Voice-2.1 for narration, audiobooks and lessons, and the Flash variant for voice agents and quick spoken replies. In AI SDK 7, generateSpeech picks a voice by name, such as en-US-Harper:MAI-Voice-2.1-Flash.
  • streamTranscribe takes a stream of audio chunks (Vercel's example uses 16 kHz, 16-bit PCM) and emits partial transcripts as audio arrives. Partials can change, so Vercel advises replacing the displayed text with each new partial rather than appending.
  • Vercel says AI Gateway is one of a limited number of platforms offering MAI models, and ties the partnership to MAI's focus on security and zero data retention.
Terms in this piece · Glossary
  • streaming — Sending a model's response token by token as it is generated, so the reader sees text immediately instead of waiting for the whole answer.
More from Kevin Dawkins
Recommended reads
Comments

Checking sign-in…

Loading comments…