News / AI costs

Microsoft cut cloud transcription to 10 cents an hour, and the catch is in the small print

Home/News/Microsoft transcription pricing

October 20, 20262 min readNumbers checked by hand

The short version

Microsoft’s batch transcription model runs at $0.10 per hour of audio until 31 December. The streaming version is $0.54 per hour, or 5.4 times the batch rate, for half a point of accuracy. Both are introductory prices with no announced successor.

$0.10Batch, per audio hour
$0.54Streaming, per audio hour
5.4xStreaming premium
Microsoft AI announcement page for MAI-Transcribe-2-Streaming, October 2026
Microsoft’s announcement page, captured 20 October 2026.

What changed

Microsoft shipped two versions of the same speech model. The batch version, MAI-Transcribe-2, costs $0.10 per hour of audio at a 2.0% word error rate. The streaming version, MAI-Transcribe-2-Streaming, costs $0.54 per hour at a claimed 2.5% error rate and finalises a transcript about 0.13 seconds after speech ends.

ModelPriceWhat you get
MAI-Transcribe-2 (batch)$0.10per audio hour2.0% WER, file in, transcript out
MAI-Transcribe-2-Streaming$0.54per audio hour2.5% WER, live partial transcripts
MAI-Voice-2.1$22per 1M charactersText to speech, 23 languages
MAI-Voice-2.1-Flash$15per 1M charactersFaster, cheaper TTS tier

The catch nobody prints

Three things, all in the small print. First, both rates are introductory and run out on 31 December 2026, with no replacement price published. Second, streaming costs 5.4 times the batch rate for slightly worse accuracy, which is the opposite of how volume pricing usually works. Third, this is a public preview with no service level agreement.

It is also not the cheapest option on the market. xAI lists streaming transcription at $0.20 an hour and Meta at $0.18, so Microsoft is competing on accuracy and language coverage rather than on price.

What it means if you already transcribe

If you were paying around $12 an hour for cloud transcription, this is a genuine 20 to 100 times cut, and it makes the paid tier competitive with running Whisper on your own machine. If you only transcribe a few hours a month, local is still cheaper: it costs nothing per hour and you already own the computer. Our full comparison is in local Whisper vs cloud transcription.

Source

Microsoft AI model pages and announcement posts (microsoft.ai), read 20 October 2026. Competitor streaming prices from the Artificial Analysis comparison published 2 October 2026.

Read next: the real cost per hour of transcription