Muse Voice Transcribe $3 per 1,000 minutes vs Sume STT rate

Meta launched Muse Voice Transcribe on Meta Model API at "$3.00 per 1000 minutes ($0.18/hour)". That is $0.003 per minute. Sume's speech-to-text is $0.01 per audio minute, so 1,000 minutes is $10.00 and an hour is $0.60.

Meta's price is from its developer blog post; Sume's from the rate card and the video inspect docs, read 2026-10-01. Sume's default agent fee is not in the Sume column, and no accuracy comparison is made.

How do the two prices convert?

Dividing Meta's $3.00 by 1,000 gives $0.003 a minute, and the post's own hourly figure ($0.18) is 60 times that. The table puts both on the same units.

What is in Meta's price and what is not?

The post states the price once, for the model on Meta Model API, and says the same model handles streaming and one-shot transcription. It does not break the price down by mode, so this post treats $3.00 per 1000 minutes as the figure for both.

How does Sume decide how much to reserve?

Sume bills per audio minute. You can send duration_seconds from 1 to 600 to improve the usage reservation; omit it and the reservation is 1 minute. The hint is capped at 10 minutes, so for a long recording check the docs for the audio length one request takes before you budget.

What do I do with the numbers?

For a large backlog, multiply by your real audio volume, not by the number of files; hours of audio drive the bill. Then check what else you need: Muse offers streaming over a WebSocket, while Sume runs request-response jobs. See cost to transcribe 1,000 hours for a worked backlog.