Transcription Drops to $0.10 an Audio Hour: Microsoft's MAI-Transcribe-2 Cuts Price 72% — but the Rate Is Temporary and the Model Is in Public Preview

Microsoft AI released MAI-Transcribe-2, its in-house speech-to-text model, on September 3, priced at $0.10 per audio hour — roughly 72% below the previous generation's $0.36. Microsoft states plainly that this is a limited-time rate through the end of 2026, with no standard price announced for afterward. Azure's documentation also marks the model as public preview: no SLA, not recommended for production workloads. Microsoft's own figures: 60 languages, first place on FLEURS with a 5.2% average word error rate, second on the Artificial Analysis WER leaderboard, and speed claims of 10× GPT-Transcribe, 7× ElevenLabs Scribe v2, and 5× Gemini 3.5 Transcribe. Speaker diarization, word-level timestamps, keyword biasing, verbatim and clean transcription styles, and automatic language identification are all folded into that price.

What Fell Isn't Just the Rate — It's the Add-Ons

$0.36 to $0.10 is the 72% on the surface, but the more practical change is that features usually billed separately now sit inside the base rate: speaker diarization, word-level timestamps, keyword biasing (`phraseList.phrases`, for domain terms and proper nouns), automatic language identification, mid-sentence code switching, and two output styles — `verbatim`, which keeps the "um"s and the false starts for compliance and QA work, and `clean`, which strips them and produces captions and notes you can publish. Every leaderboard number here is vendor-supplied: 5.2% average word error rate across 60 FLEURS languages, which Microsoft says is first place, and second on the Artificial Analysis WER leaderboard. The three speed multiples are likewise Microsoft citing third-party measurement, with no independent reproduction yet.

Two Lines of Fine Print Matter More Than the Discount

First, the price has an expiry. Microsoft's wording is limited-time through the end of this year, and it has not said what next year costs. Multiplying $0.10 into a 2027 budget is a bad idea: by Microsoft's own example, 100,000 hours of call-center audio a year runs $10,000 at this rate, and $36,000 at the old one. Second, that public-preview notice in the Azure docs: no service-level agreement, some capabilities may be constrained, not recommended for production workloads. Together those two lines make this a good candidate for evaluation and pilots, and a poor one for cutting a running transcription pipeline over wholesale. The engineering limits are worth reading too: audio files under 300 MB, in WAV, MP3 or FLAC; `locales` forces exactly one language, and Microsoft advises against setting it unless you are certain auto-detection is failing. The language table lists `zh` (Simplified Chinese) and `yue` (Cantonese), with no separate Traditional Chinese locale — teams shipping Traditional Chinese will need to test the output characters and wording themselves.

What This Changes for Buyers

This site covered Google's Gemini 3.5 Transcribe on August 27, and quoted prices along this line have fallen quickly in the month since. For teams already self-hosting Whisper, the arithmetic is now blunt: power, ops and queue time on your own card, against $0.10 an hour — particularly when diarization and word-level timestamps no longer have to be assembled by hand. If you build meeting notes, captions or call-center QA, the number to watch is the one for January that has not been published. What you can do now is finish the evaluation and wrap the interface, rather than building a business model on a figure explicitly labeled temporary.

via: Microsoft AI announcement, Microsoft Learn: MAI-Transcribe documentation, VentureBeat