The latest speech transcription model from Microsoft AI achieves a 2.0% Word Error Rate and processes audio at 411x real time. Priced at $1.67 per 1,000 minutes via Microsoft Foundry, the model is less than half the cost of high-accuracy competitors like ElevenLabs Scribe v2 at $3.67 and Smallest AI Pulse Pro at $4. It ranks second overall on the Artificial Analysis accuracy leaderboard, trailing Alibaba's Fun-Realtime-ASR-preview.

MAI-Transcribe-2 expands language support from 43 to 60 languages and introduces speaker diarization, word-level timestamps, and both clean and verbatim transcription styles. Mustafa Suleyman said the tool is 10x faster than GPT-Transcribe and 5x faster than Gemini 3.5. The model is currently available for use through both Microsoft Foundry and OpenRouter.

Sign in to suggest edits

Key sources

  1. SOURCE@mustafasuleyman“10x faster than GPT-Transcribe. 5x faster than Gemini 3.5”x.com
  2. SUPPORT@artificialanlys“ranking 2nd overall on the Artificial Analysis Word Error Rate (AA-WER) leaderboard, behind Alibaba’s Fun-Realtime-ASR-preview (1.7% WER)”x.com
  3. SUPPORT@artificialanlys“achieves 1.8% WER on AA-AgentTalk, 1.5% on VoxPopuli-Cleaned-AA, and 3.0% on Earnings22-Cleaned-AA”x.com
  4. SUPPORT@artificialanlys“leading the accuracy-speed Pareto frontier with a speed factor of 411x”x.com
  5. SUPPORT@artificialanlys“below other high-accuracy models such as ElevenLabs Scribe v2 ($3.67), Smallest AI Pulse Pro ($4), and Gemini 3.5 Transcribe ($5)”x.com
Markdown