A 10-person team inside Mustafa Suleyman’s Microsoft AI division shipped a speech-to-text model on September 3 that picks a fight with OpenAI, Google, and ElevenLabs simultaneously. MAI-Transcribe-2 ranks first on FLEURS across 60 languages with 5.2% average word error rate, and Microsoft claims it runs 10x faster than GPT-Transcribe, 7x faster than ElevenLabs Scribe v2, 5x faster than Gemini 3.5 Transcribe — at half the GPU cost.
Why the price is the real story
$0.10 per hour of audio through end of year, down 72% from the previous model’s $0.36. Transcription is a volume business — captions, meetings, call centers — and Microsoft is betting nobody follows it down. Speaker diarization, word-level timestamps, code-switching, and keyword biasing all come standard, so the cheap tier isn’t a stripped-down one.
How to use the API
It’s live on Azure AI Foundry, Speech Service, the MAI Playground, and OpenRouter. Feed it audio, get back diarized transcripts with timestamps — clean or verbatim. Typical builds: meeting notes, clinical documentation, voice agents that need to know who said what, when.
You Might Also Like
- Microsoft mai Models mai Transcribe 1 mai Voice 1 mai Image 2 are Live Redmonds ai Independence Starts now
- Openai Codex Claude md Auto Import Makes Switching From Claude Code a two Click Move
- Microsoft Mdash Scores 88 45 on Cybergym Beating Anthropic Mythos and Openai gpt 5 5
- Microsoft mai Code 1 Flash is Live in Github Copilot 60 Fewer Tokens Than Comparable Coding Models
- Google Gemini Enterprise Gives Every ai Agent a Cryptographic id its Real Fight With Openai and Anthropic
Top AI Product 
Leave a comment