Microsoft AI released MAI-Transcribe-2 on Thursday at $0.10 per hour of audio, a roughly 72 percent cut from the $0.36 that its predecessor, MAI-Transcribe-1.5, charged only weeks ago. The launch price, per the Azure AI Foundry blog, is a limited-time offer that runs through the end of 2026. Microsoft hasn’t disclosed what happens after that, a detail eWeek noted and buyers are already recalculating around.

The number matters more than the model. Speech-to-text has been priced, since the Whisper era, as a metered specialty service; at ten cents an hour, it starts to behave like storage.

On the benchmarks, Microsoft is claiming both ends of the frontier. The company says MAI-Transcribe-2 ranks first on the FLEURS multilingual benchmark with an average 5.2 percent word-error rate across 60 languages, and second on the Artificial Analysis word-error-rate leaderboard. Independent testing by Artificial Analysis, per Microsoft’s newsroom, puts the model roughly 10 times faster than OpenAI’s GPT-Transcribe, 7 times faster than ElevenLabs’ Scribe v2, and 5 times faster than Google’s Gemini 3.5 Transcribe. eWeek notes the leadership isn’t total: Google Gemini 3.1 Pro edges it out in Armenian and Danish, and ElevenLabs Scribe v2 leads in Afrikaans, while MAI-Transcribe-2 leads in Chinese and French.

What competitors normally sell as premium add-ons are folded into the base rate. Speaker diarization, word-level timestamps, keyword biasing for domain terminology, automatic language identification, and a configurable output style with a “verbatim” setting that preserves filler words for compliance and a “clean” setting that strips them for readable notes all ship at ten cents. The model accepts WAV, MP3, and FLAC files up to 300 MB through Microsoft Foundry and the MAI Playground. Microsoft Learn classifies it as a public preview without an SLA and doesn’t recommend it for production.

The competitive geometry is the point. A small firm running 20 hours of calls a month now pays $2 to get every conversation transcribed, diarized, and time-stamped. That’s roughly the price of a coffee for a month of pipeline evidence, and it collapses the business case for a whole tier of call-intelligence subscriptions.

The preview caveat is real, and so is the end-of-year expiration. Both are the sort of asterisk that tends to disappear once the market recalibrates around the headline number.

Sources