10,000 eight-second clips: STT billed per second vs per hour

Ten thousand clips of 8 seconds are 80,000 seconds, or 22.2 hours of audio. At list that is $12.00 on MAI-Transcribe-2-Streaming and $13.33 on Sume STT, a gap of $1.33. The unit of billing is the part that deserves a closer look at this scale.

The bill

Sume bills each job by the second, at 1 cent a minute, rounding a job up to a whole second. A price quoted per hour, as Microsoft's is, can be billed per second or in larger steps, and the page we read gives the hourly rate only.

Why short clips round up

Take a clip of 8.2 seconds. Sume bills each job by the second, so a clip of 8.2 seconds rounds up to 9, and its cost is $0.0015 rather than $0.00137, a 9.8 percent increase for that file. Across 10,000 clips of 8.2 seconds the audio is 82,000 seconds but the billed seconds are 90,000, so the bill is $15.00 instead of $13.67.

So the question for a batch of short clips is how long the clips really are. Measure the distribution, round each clip up to whole seconds, sum those, and price that. The mean alone understates the bill.

The same rounding applies to the 8-second clips of this post if they are really 8.0 seconds, which is rare. Treat the $13.33 figure as the floor for 80,000 seconds of audio and expect a real batch to come in above it.

Fewer, longer jobs

Short files also raise the job count. Ten thousand clips is ten thousand jobs, each with its own submit, status poll and result read. A bounded pool of workers is the way to run it, and every submit needs an Idempotency-Key, so a crashed run resumes without paying twice.

If your clips are short pieces cut from longer files, consider transcribing the long file once and slicing the text by word times. A 10-minute file is one job, and its words[] array gives start and end times you can cut the clips from. That avoids the round-up on every piece, as long as the file is within the 600-second cap.

What the table leaves out

Microsoft's figure is the launch price from 2026-10-01 (Microsoft AI, read 2026-10-05): $0.54 per hour for MAI-Transcribe-2-Streaming through year end. This post treats that hourly rate as pro rata per second of audio, which is an assumption, so check Microsoft's billing terms before you rely on the last cent.

The two are different kinds of service. MAI-Transcribe-2-Streaming is a streaming model, built to transcribe audio as it arrives. Sume STT is an async job: you send a public HTTPS audio_url, poll GET /v1/jobs/:id/status, and read text, word times and optional sentence segments from /result. A recorded file suits the job form well, and a live call does not.

Each Sume job takes at most 600 seconds of audio and each submit needs an Idempotency-Key. Sume bills each job by the second at $0.01 a minute, so the table above is the whole cost of the transcription itself.

Decide with a sample

Keep the 10,000 clips as clips only if each needs its own transcript, such as voice samples or call fragments. For slices of a single recording, a single transcript with sentence segments is cheaper and easier to keep consistent.

The answer depends on a number you can measure in an afternoon: take 200 representative clips, probe each duration, and price the rounded sum. Scale up from there.

Remember the other side of the comparison. The Microsoft figure is an hourly rate for a streaming model, so it too may round in steps that the launch page does not spell out. Until you read its billing terms, treat the $12.00 as an approximation, not a quote.