What $100 of transcription buys: Scribe v2, MAI, Ink and Sume STT

Hours per $100

At list prices $100 buys about 455 hours on ElevenLabs Scribe v2 ($0.22 an hour), about 256 hours on Cartesia Ink ($0.39 an hour on its Scale plan), about 185 hours on MAI-Transcribe-2-Streaming ($0.54 an hour introductory) and about 167 hours on Sume STT ($0.60 an hour).

Divide 100 by the hourly price: 100 / 0.22 = 454.5, 100 / 0.39 = 256.4, 100 / 0.54 = 185.2 and 100 / 0.60 = 166.7.

The table

Each hourly price is on the vendor's own page. They are not the same product: some are streaming, some are batch, and some prices need a plan.

What sits behind each price

ElevenLabs' pricing page lists Scribe v2 at $0.22 an hour and Scribe v2 Realtime at $0.39. Microsoft's announcement gives $0.54 an hour for MAI-Transcribe-2-Streaming as an introductory price through year end. Cartesia's pricing page shows Ink at $0.39 an hour on the Scale plan, and Scale is a $299 monthly plan.

Sume STT is $0.01 per audio minute with no plan. One job takes up to 600 seconds, returns words[] with timings, and is billed per minute you submit, which makes budgeting a matter of counting minutes.

When the cheapest hour is not the cheapest project

A low hourly rate does not include the work to run the service. A plan with a monthly fee costs the same whether you transcribe one hour or a hundred. A streaming model is built for live audio, where you hold a socket open; a batch job fits files you already have. And a promotional rate can change, so a budget based on it needs a date.

• Fewer than 20 hours a month: the difference between these services is a few dollars.
• Archives of thousands of hours: negotiate directly and test accuracy on your own audio.
• Live captions: use a streaming service; Sume STT is for files.
• Always test your hardest ten minutes before you commit.

A monthly check

Hours per $100 is a useful scale, but your own volume is what you pay. At 20 hours a month the bill is $4.40 on Scribe v2, $7.80 on Ink, $10.80 on MAI-Transcribe-2-Streaming and $12.00 on Sume STT. The gap between the cheapest and the dearest is $7.60 a month, which is smaller than the cost of one failed batch you have to redo.

So test with a sample before you move a whole archive. Pick the noisiest ten minutes you have, run it through each candidate, and count the words you would have to fix by hand. The service with the fewest corrections often wins even when its hourly rate is higher.