MAI-Transcribe-2: the highest quality, cheapest transcription at the fastest speed! 10x faster that GPT-Transcribe. Now available on Microsoft Foundry.

Sep 3, 2026 · 2:38 PM UTC

28
58
29
604
45,946
RelevantRecentLikes
Try it in the MAI Playground: playground.microsoft.ai/

MAI Playground | Microsoft AI

Explore MAI Playground, the Microsoft AI Playground for running and experimenting with new AI models.

playground.microsoft.ai
1
2
18
4,372
View Docs links on microsoft.ai/models/mai-tran… are broken

MAI-Transcribe-2 | Microsoft AI

microsoft.ai
2
476
Is it easy to get an API key to play with this?
1
2
1,000
non-streaming? 🤨
229
10x faster is a big claim to back up
2
260
Local model please 🙏
1
448
no streaming?
389
We need streaming version as well
362
And use it on OpenRouter: xitter.cf/OpenRouter/status/2095…
OpenRouter
@OpenRouterSep 3
Microsoft MAI-Transcribe 2 is live on OpenRouter! Ranked #1 on FLEURS, it supports speech-to-text across 60 languages with automatic language ID, code switching, speaker diarization, and word-level timestamps. Use it now: openrouter.ai/microsoft/mai-…

MAI-Transcribe 2 - API Pricing & Providers

MAI-Transcribe 2 is a multilingual speech-to-text model from Microsoft AI, ranked #1 on the FLEURS multilingual benchmark. Priced at $0.10 per hour.

openrouter.ai
2
1
6
964
Congrats on the release! Will cover this on today's @thursdai_pod and will compare to Muse Voice Transcribe 🫡 Amazing work from MAI!
3
465
Regions and ZDR?
144
Let's open source the transcribing model atleast if not LLMs this would significantly helpful for everyone
8
668
Why Microsoft always Gatekeeping? Release the weights to public.
4
672
good to see humans writing posts again
1
636
It doesn't seem to be available in Foundry
208
No support of streaming? What would i do with it?
332
Pretty nice，transcribed three languages correctly with word-level timestamps，also the local Chinese accent.
3
291
Why don't you add in Bing?😡😡🤡
1
42
WHY IS COPILOT CHAT STILL USING OUTDATED TRANSCRIPTION MODELS THEN?
1
139
The combination of 5.2% FLEURS WER across 60 languages, diarization, and $0.10/hour is the more useful story than “10x faster.” Caveat: those are vendor-reported evals, so I’d want the exact test setup before calling it a clean head-to-head.
10
Hey there, do you guys want to put a voice-dictation button instead of the Microsoft Copilot app with this model? It's pretty awesome.
64
Thanks
11
Load more replies