Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

MAI-Transcribe-2-Streaming

Microsoft's first live transcription model: 60 languages, partial text in about 100 ms, $0.54 per audio hour while in preview.

Reviewed by Andrew Miller · Updated

Our rating

4.1/5

MAI-Transcribe-2-Streaming looks like the most accurate live transcription model you can buy, at a mid-market price that only holds until December.

What is MAI-Transcribe-2-Streaming?

MAI-Transcribe-2-Streaming is Microsoft AI's first streaming speech-to-text model, released on 1 October 2026 alongside the MAI-Voice-2.1 voice models. You send audio as a continuous stream and get partial transcripts that update as the speaker talks, then a final result for each segment. Microsoft says the first partial text arrives just over 100 milliseconds after the audio.

It covers 60 languages and detects the language automatically and continuously, so a speaker can switch mid-conversation. Microsoft says it launched first for accuracy on both final and partial transcripts on Artificial Analysis's streaming benchmark, and that words appear twice as fast as with its closest competitor in its own tests.

It is in public preview on Azure Speech, served from Sweden Central, Central US, East US 2 and Southeast Asia, and reachable through an OpenAI Realtime-compatible WebSocket API or the Azure Speech SDK. The introductory price is $0.54 per hour of audio until the end of 2026.

Is MAI-Transcribe-2-Streaming worth it?

MAI-Transcribe-2-Streaming is a serious entry into live transcription, the job behind captions, voice agents and call-centre assist. Microsoft says it ranks first for accuracy on Artificial Analysis's streaming leaderboard for both partial and final transcripts, and sits on the accuracy-versus-latency frontier there.

Speed is the second claim. Partial text arrives in just over 100 milliseconds, and Microsoft's own tests on dictation and subtitles put words on screen twice as fast as the closest competitor, which it does not name. Microsoft publishes no word error rates in its launch post, so buyers should run their own audio before switching.

On price it sits in the middle. At $0.54 per audio hour, about $0.009 a minute, it is cheaper than OpenAI's gpt-live-transcribe at $0.017 a minute but dearer than ElevenLabs Scribe v2 Realtime at $0.39 an hour or Deepgram Nova-3 streaming at $0.0048 a minute on promotion. The $0.54 rate is introductory and ends with 2026, and Microsoft has not said what follows.

Integration is easy for teams already on OpenAI's Realtime API, since the WebSocket interface is compatible, and the Speech SDK handles reconnection and retries. Coverage of 60 languages with continuous detection suits multilingual call centres, where callers switch language mid-call.

The constraints are those of a preview. Microsoft says there is no service-level agreement and does not recommend it for production. It runs from four regions only, and the launch post does not mention speaker labels. Neither the launch post nor the documentation describes how audio is retained, and Azure's preview terms allow preview features to follow different privacy practices.

Where it earns its place

  • A contact-centre software vendor serving European banks

    Transcribe live calls in several languages for agent assist

    Stream call audio from Sweden Central over the Realtime API; continuous language detection handles callers who switch language, and 10,000 call hours a month cost about $5,400 at the introductory rate.

  • A university media team

    Add live captions to lectures streamed online

    Use the Azure Speech SDK to send lecture audio and show partial captions as they arrive, about 100 ms after speech, replacing them with final text per segment.

  • A developer of a voice assistant already built on OpenAI's Realtime API

    Swap in a more accurate transcription step without rewriting the client

    Point the existing WebSocket client at Azure with the MAI-Transcribe-2-Streaming model name and compare accuracy on recorded user audio before moving traffic, given the preview status.

Before you commit

Setup
Create an Azure Speech or Foundry resource, then connect over the Realtime-compatible WebSocket API or with Azure Speech SDK version 1.52.0 or later.
Integrations
Azure Speech SDK, an OpenAI Realtime-compatible WebSocket API, Azure Voice Live, Microsoft Foundry, MAI Playground and Vercel, with LiveKit announced as coming soon.
Data and privacy
Not described in the launch post or the model documentation; Azure's supplemental preview terms note that preview features may follow different privacy practices from generally available services.
Cost at scale
At the introductory $0.54 per audio hour, 1,000 hours a month cost $540; the price after 2026 is not published.
Limits
Public preview without an SLA; served from Sweden Central, Central US, East US 2 and Southeast Asia; 60 languages.

What’s good

  • First for accuracy on Artificial Analysis's streaming benchmark, per Microsoft
  • Partial transcripts in just over 100 ms
  • 60 languages with continuous automatic detection
  • OpenAI Realtime-compatible API plus Azure Speech SDK

What’s not

  • Public preview with no SLA; not recommended for production
  • $0.54 per hour is introductory and ends with 2026
  • No published word error rates; speaker labels not mentioned
  • Served from four Azure regions

Best for

Teams building live captions, voice agents or call-centre tools on Azure who want top streaming accuracy across many languages.

MAI-Transcribe-2-Streaming alternatives

MAI-Transcribe-2-Streaming: common questions

Is MAI-Transcribe-2-Streaming worth it?
MAI-Transcribe-2-Streaming looks like the most accurate live transcription model you can buy, at a mid-market price that only holds until December. We rate it 4.1 out of 5.
How much does MAI-Transcribe-2-Streaming cost?
Introductory $0.54 per hour of audio (about $0.009 per minute) until the end of 2026; Microsoft has not published the price after that.
What are the downsides of MAI-Transcribe-2-Streaming?
Public preview with no SLA; not recommended for production; $0.54 per hour is introductory and ends with 2026; no published word error rates; speaker labels not mentioned; served from four Azure regions.
Who is MAI-Transcribe-2-Streaming best for?
Teams building live captions, voice agents or call-centre tools on Azure who want top streaming accuracy across many languages.
How much does MAI-Transcribe-2-Streaming cost for a team or in production?
At the introductory $0.54 per audio hour, 1,000 hours a month cost $540; the price after 2026 is not published.
Is MAI-Transcribe-2-Streaming safe for company data?
Not described in the launch post or the model documentation; Azure's supplemental preview terms note that preview features may follow different privacy practices from generally available services.
What are the best MAI-Transcribe-2-Streaming alternatives?
The closest alternatives we have reviewed are Deepgram Nova-3, ElevenLabs Scribe v2, GPT Transcribe and Gemini 3.5 Transcribe.

Answers checked Nov 6, 2026.

Make MAI-Transcribe-2-Streaming? Show this rating on your siteMAI-Transcribe-2-Streaming rated 4.1 out of 5 on ReviewRamble