Speech-to-text per hour: Sume $0.60 batch vs MAI streaming $0.54

Sume speech to text is $0.01 per audio minute, which is $0.60 per hour of audio. Microsoft lists MAI-Transcribe-2-Streaming at $0.54 per hour through year end and 60 languages (read 2026-10-08). So the Microsoft list is $0.06 an hour lower.

These are different products. Sume STT 1.0 is a batch job: you submit an audio URL, poll or take a webhook, and read the transcript. MAI-Transcribe-2-Streaming is, by its name and its announcement, a streaming model. A price per hour is comparable; the way you call them is not.

The hourly bill

Sume reserves per audio minute, and one request is capped at 10 minutes (duration_seconds is at most 600). The table budgets whole minutes.

What the cap means in practice

A ten-minute ceiling per request means a 60-minute recording is six jobs. You can cut the ranges yourself by detaching audio once and splitting it; Audio detach accepts a range, caps output at 900 seconds per job and takes a source up to 1800 seconds.

That bookkeeping is the real cost of batch. If you need live captions while a person is still talking, a streaming product is the right shape and this post does not argue otherwise.

A fair read of the gap

At 100 hours the difference is $6.00. The Microsoft rate is listed through year end, so compare again in January rather than building a model on it.

STT on Sume is also reachable inside Video inspect with transcribe: true, at the same $0.01 per audio minute, plus the inspect's own compute.