ModelCards

MAI-Transcribe-2

Audio modelCard summarized

Rapidly turn any audio into highly accurate text transcripts

Source: MAI-Transcribe-2 Model Card (8 pp)

Use it via API

$0.10 / hour
Code exampleCompare

Where to call it

  • Microsoft Foundry / AzureMAI-Transcribe-2
  • OpenRoutermicrosoft/mai-transcribe-2

Supports

  • Streaming
Full feature support

Good for

  • Contact center transcription with diarization and keyword biasing
  • Clinical note taking and legal/financial records
  • Meeting transcription with speaker labels and word timestamps

Watch out for

Overview

Microsoft · 01Transcribe, front

Transcribe 2

Rapidly turn any audio into highly accurate text transcripts

Typical price
$0.10 / hour
MAI-Transcribe-2 (batch)
Max length
2 h per file
Languages
60 languages
Modes
speech-to-text, speaker diarization
Released
3 Sep 2026

MAI-Transcribe-2 is the second generation of Microsoft AI's in-house speech-to-text model. It covers 60 languages with automatic language detection, and adds speaker diarization, word-level timestamps and keyword biasing, plus verbatim or clean transcription styles. Microsoft reports the lowest average FLEURS WER across 60 languages among the systems it compared (5.2%), #2 on the Artificial Analysis speech-to-text leaderboard, and long-form speed of about 10 seconds per hour of audio. pp.1–3

MAI-Transcribe-2
Typical price
$0.10 / hour
Max length
2 h per file (300 MB per model card; REST reference says under 250 MB)
Languages
60 languages
Weights
Closed
Released
3 Sep 2026
  • speech-to-text
  • speaker diarization
  • word-level timestamps
  • keyword biasing
  • verbatim or clean transcription style
  • automatic language ID and code switching