MAI-Transcribe-2-Streaming at $0.54 an hour: budget a weekly show
Microsoft's $0.54 per hour streaming rate runs through year end. Budget a 3-hour weekly show, then caption the cut-down clips on Sume at $0.20 each.

A weekly three-hour livestream costs about $1.62 a show in MAI-Transcribe-2-Streaming at $0.54 per hour of audio, and that rate is an introductory price Microsoft says holds through the end of the year (read 2026-10-03). From now to December 31 is about 13 shows, so roughly $21. Microsoft's pages I read do not give a price for 2027, so do not put a 2027 number in a budget until it does.
Live captions are only half the job. The clips you cut from the show for social feeds need burned-in captions of their own, and that is a different system with a different price.
The arithmetic
The last row is a thought experiment. Microsoft calls $0.54 an introductory rate through the end of the year, and the page does not state what replaces it. The same launch post also says the model produces first partial hypotheses in just over 100 ms and supports 60 languages with automatic, continuous language detection (read 2026-10-03).
| Input | Value | Result |
|---|---|---|
| Streaming rate | $0.54 per audio hour | - |
| One 3-hour show | 3 x $0.54 | $1.62 |
| Remaining weekly shows in 2026 (Oct 3 to Dec 31) | about 13 | about $21.06 |
| A full year at the same rate (not guaranteed) | 52 x $1.62 | $84.24 |
What streaming transcription gives you, and what it does not
Streaming output arrives as partial hypotheses that get revised, then settled text. That is what a live caption overlay or a voice agent needs. It is not a finished subtitle file for a clip. A clip needs final text, cut to the clip's own start and end, with line breaks a viewer can read.
Sume does not offer a streaming transcriber in the docs I read. Its speech-to-text is a job endpoint (POST /v1/stt-1.0/transcribe, named on the timeline audio page). So the clean split is: use a streaming model for the live room, and use Sume after the show for the clips.
Captioning the clips after the show
Sume's standalone video captions job takes a public HTTPS video URL and returns a captioned video. Each accepted job reserves and captures $0.20 for videos up to 60 seconds under the current fixed estimate, so ten 45-second clips from one show are $2.00. If you already hold the final text from your live transcript, you can pass it as cues with text, start and end in seconds and skip transcription; the docs say that path burns exactly that copy at those times.
A request for one clip looks like this:
curl -X POST https://api.sume.com/v1/video-captions \
-H "Authorization: Bearer $SUME_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: show-042-clip-03" \
-d '{
"video_url": "https://media.sume.com/artifacts/example/show-042-clip-03.mp4",
"style": "punch",
"cues": [
{ "text": "The rate is introductory.", "start": 0.4, "end": 2.1 },
{ "text": "Budget it through December.", "start": 2.2, "end": 4.0 }
]
}'
What to ask before you rely on the year-end rate
Three questions are worth answering before you build a plan around an introductory price. What happens to the rate on January 1: Microsoft's launch post and model page, as I read them, say only that the price runs through the end of the year. Is your audio the unit Microsoft bills: the post says per hour of audio, so a stream with long silences still counts as hours unless you stop sending audio. And what does a language switch cost: the model detects language continuously across 60 languages, which suits a bilingual show, but the price is not stated per language.
On the Sume side, check the clip length before you budget. The $0.20 figure applies to videos up to 60 seconds under the current fixed estimate, and the docs tell you to confirm live pricing in GET /v1/catalog and OpenAPI. A 90-second recap is not covered by that sentence, so quote it from the catalog rather than multiplying. If a clip speaks two languages, leave language out and the job detects it automatically, or pass your own cues and skip speech recognition altogether.
Keep the dates on the numbers
Put the read date next to every price in your budget sheet. This post's figures were read on 2026-10-03; a rate called introductory is the kind that changes.
A one-line budget
For the last quarter of 2026: live transcription about $21 for 13 three-hour shows, plus $2 per ten clips captioned on Sume, so 40 clips cost $8. Total about $29 for the quarter, before voice, music and editing. The live transcription is not the expensive part. Re-check Microsoft's page in January, because the rate it names ends on December 31.
Sources
Related posts
More in Pricing
- Meeting transcript budget: a year of weekly calls at $0.01 a minute
A year of ten 30-minute meetings a week is 15,600 audio minutes and $156.00 in transcripts on Sume. Four load levels from $62.40 to $780.00, table inside.
- Mercury Voice pricing: tokens to dollars per minute of talk
Mercury Voice lists at $0.40/$1.50 per million tokens, half off at launch, about $0.009 a minute. How to turn that into your bill, plus the speech steps.
- MAI-Voice-2.1-Flash '60% cheaper': the baseline that implies
Microsoft calls Flash about 60% cheaper than comparable models at $15 per 1M characters. That implies a $37.50 baseline. A check against a $47.50 TTS rate.
- H3 Max 1080p, 10 seconds: $1.84 before the margin change, $2.00 now
Sume bills MiniMax H3 Max at list x 1.25 like every other model, no longer x 1.15. A 10-second 1080p clip moves from $1.84 to $2.00. 480p and 768p rows too.
Written by Sume