What Is Microsoft MAI-Transcribe-1?

MAI-Transcribe-1 is Microsoft’s new in-house speech recognition model, announced April 2, 2026. It’s the company’s answer to OpenAI’s Whisper and marks Microsoft’s shift toward developing its own foundation models.

Last verified: April 2026

Quick Facts

Detail	Info
Type	Speech-to-text (ASR)
By	Microsoft MAI Superintelligence team
Released	April 2, 2026
Available	Microsoft Foundry (public preview)
Powers	Copilot Voice, Microsoft Teams

What It Does

MAI-Transcribe-1 converts spoken audio into text with high accuracy for:

• Meeting transcription — Microsoft Teams call transcripts
• Voice commands — Copilot’s Voice mode
• Audio content — Podcasts, interviews, recordings
• Enterprise workflows — Downstream tasks like summaries, action items, search

Why It Matters

This is part of Microsoft’s broader strategy shift. The company has historically relied on OpenAI for foundation models, but the MAI Superintelligence team (formed November 2025) is building Microsoft’s own models. MAI-Transcribe-1 joins MAI-Voice-1 (text-to-speech) and MAI-Image-2 (image generation) as Microsoft’s first major independent model releases.

For developers, it means a new choice beyond Whisper, AssemblyAI, and Deepgram — backed by Microsoft’s Azure infrastructure.

How to Access

1. Microsoft Foundry — Public preview API access
2. MAI Playground — Try it in a web interface
3. Copilot Voice — Already integrated
4. Microsoft Teams — Phased rollout for transcription

# Via Azure Foundry API (example)
curl -X POST https://foundry.microsoft.com/v1/audio/transcriptions \
  -H "Authorization: Bearer YOUR_KEY" \
  -F "model=mai-transcribe-1" \
  -F "[email protected]"

Competitive Landscape

Model	By	Strengths
MAI-Transcribe-1	Microsoft	Enterprise, Teams/Copilot integration
Whisper	OpenAI	Open source, wide language support
Deepgram Nova-3	Deepgram	Real-time streaming, low latency
AssemblyAI	AssemblyAI	Developer-friendly, good accuracy

Last verified: April 2026