Scribe v2 Realtime $0.39/h vs batch $0.22/h: when does Sume STT fit?

ElevenLabs' API pricing page lists Scribe v2 at $0.22 per hour and Scribe v2 Realtime at $0.39 per hour. Sume STT, which runs Scribe v2 behind a hosted endpoint, sells at about $0.01 per minute, or $0.60 per hour. Choose Realtime if text must appear while someone is still talking. Choose a file job if the recording already exists and 10 minutes per request is enough.

The price gap from $0.22 to $0.60 per hour is what you pay for Sume's wrapper: one API key, one job format, and the same billing as the rest of your media calls.

1,000 hours, three ways

ElevenLabs numbers are from its pricing page. Sume's is $0.008 per minute provider list times 1.25.

• I could not reconcile ElevenLabs' $0.22 per hour with the $0.008 per minute that Sume's pricing table lists as the provider rate; the Sume price is shown as sold.
• Sume STT has a 10-minute limit per request, so a 1,000-hour archive is at least 6,000 jobs.

Pick in four steps

• If a person is talking now and needs live text, use a streaming product.
• If audio is a file, cut it into pieces of 10 minutes or less.
• Submit each piece with its true duration_seconds.
• Stitch the word timings back together by adding each piece's start offset.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/stt-1.0/transcribe",
    headers={
        "Authorization": f"Bearer {os.environ['SUME_API_KEY']}",
        "Idempotency-Key": "stt-demo-001",
    },
    json={
        "audio_url": "https://media.sume.com/example/clip.mp3",
        "duration_seconds": 540,
    },
    timeout=30,
)
r.raise_for_status()
print(r.json())

Stitching long files

With a 10-minute cap, a one-hour interview is six requests. Keep a small table of piece start offsets and add them to every word time you get back. Cut at silence, not at a fixed second, if you can, so you do not split a word.

Why not simply use the cheapest line

The $0.22 rate is on ElevenLabs' own price list. Using it means a separate account, a separate key and your own job handling. Sume's rate adds its margin and gives you the same job envelope, idempotency and polling pattern as its other media endpoints, plus a Sume media URL path for the audio. If you only transcribe, the direct route is cheaper. If transcription is one step in a pipeline that also generates voice, music and video, one contract may be worth the premium.

Decide with a number: how many hours a month. At 100 hours, the gap between $0.22 and $0.60 per hour is $38 a month. At 10,000 hours it is $3,800, and a direct contract deserves a serious look.

What Sume does not do

Sume STT does not stream, does not label speakers, and does not run Scribe v2 Realtime. Sume STT does not run on your device, does not stream, does not label speakers (diarization and audio-event tagging are fixed off on the server), and takes at most 10 minutes of audio per request. If you need live captions, Sume is the wrong tool.

Related

For a smaller batch see 100 ten-minute clips across services, and for realtime pricing at another vendor gpt-realtime-whisper.