AssemblyAI Universal-3.5 Pro at $0.21 an hour vs Whisper-class STT

AssemblyAI lists Universal-3.5 Pro for pre-recorded audio at $0.21 an hour. Deepgram lists Whisper Large at $0.0048 a minute, which is $0.288 an hour, so AssemblyAI's rate is about 27% lower ((0.288 - 0.21) / 0.288). The cheaper Universal-2 is $0.15 an hour. Add-ons change the total, so read the line items before you compare.

Listed prices

Per-hour figures for Deepgram are the per-minute price times 60.

Add-ons on AssemblyAI's page

AssemblyAI's page lists Medical Mode as an add-on of +$0.15 and speaker diarization as +$0.02. The page text shows these next to the hourly rates but does not make the unit clear, so check the unit on the page before adding them to an hourly total. The page also offers $50 in free credit for new accounts.

If diarization is +$0.02 an hour, Universal-3.5 Pro with speaker labels would be $0.23 an hour, still under Whisper Large's $0.288 an hour before any Deepgram add-ons (Deepgram's add-on prices were not checked).

Cost of 100 hours

Arithmetic on the per-hour column, with no add-ons.

What Sume offers

Sume does not ship AssemblyAI, Deepgram or a raw Whisper endpoint. Its transcription is transcribe: true on Video inspect, which reads one Sume-hosted clip and bills $0.01 per audio minute, with a language_code hint and optional sentence segments. That is nearly three times the $0.21 an hour rate (0.60 / 0.21 is 2.86), and the tool is shaped for clips you are already working with in Sume, not for archives.

Choose on accuracy for your audio, not price alone. None of the sources read for this post gives an accuracy benchmark, so transcribe a sample with two of the services and compare the errors that matter to you, such as names and numbers.