Transcribing 10, 100 and 1,000 hours: AssemblyAI, Scribe, Sume

Transcribing 100 hours of audio costs $15 to $24 on the cheaper dedicated vendors, $60 on Sume at $0.01 per audio minute, and $61 on Gladia Starter. At 1,000 hours the same spread is $150 to $240 against $600 and $610. Sume is the dearer choice per hour; its case rests on the job model, not on being the cheapest.

Prices below were read from each vendor's pricing page on 2026-10-07. They are published per-hour rates for batch (asynchronous) transcription, and Cartesia's is the page's own approximation for its Scale plan.

The budget table

Each total is hours x the hourly rate. Sume bills a ceil minute per job, so the hourly rate holds when your files are long; many short files round up one minute at a time.

How Sume STT bills

Sume STT is $0.01 per audio minute, which is $0.60 an hour, rounded up to the minute on each job. A job takes up to 10 minutes of audio, so a one-hour recording is six jobs of $0.10 each. A 30-second clip still bills one minute, so 1,000 clips of 30 seconds cost $10.00, not $5.00.

The cost of that split is operational, not financial: you chunk long audio, submit the pieces and reassemble the transcripts. If your inputs are meetings and podcasts, an hourly-priced vendor with long-file support is simpler. If your inputs are short clips that already live in Sume, such as reference videos or generated voiceovers, the 10-minute job is a natural unit.

Long recordings are not the only shape. Fifty 90-second clips bill 50 x 2 = 100 minutes on Sume, $1.00, though the audio is 75 minutes. Hourly vendors may bill differently for short files, so check each page's rounding wording before you compare short-clip workloads.

Which row for which workload

For archives, the Universal-2 and Scribe rows are the lowest published price and the choice for bulk transcription where only text matters. For accuracy-sensitive batches, price the higher tiers (Universal-3.5 Pro, Speechmatics Enhanced at $0.75 an hour) and compare error rates on your own audio. For pipelines that transcribe the voiceovers they just made, or captions on clips in a Sume project, the integration saving usually outweighs the extra $0.40 an hour.

Do the arithmetic on hours you actually have. A podcast network with 40 hours a month pays $6 to $9.60 on the cheapest rows and $24 on Sume: a $15 to $18 difference that no one will notice. A call-center archive of 20,000 hours is a different conversation, and belongs on a committed contract.

• Under 10 hours: the dollar difference is a few dollars; use whichever integrates fastest.
• 100 hours: roughly $36 to $45 more on Sume than the cheapest dedicated vendors.
• 1,000 hours: roughly $360 to $450 more; run a pilot on a direct vendor.

Check before you batch

Submit one typical file and read the billed amount, then multiply. The Sume API pricing page lists the current per-minute rate. For a wider vendor ranking, see the cheapest STT API per hour post.

Dated facts used here

Vendor rates were read on 2026-10-07: AssemblyAI $0.15 (Universal-2) and $0.21 (Universal-3.5 Pro) per hour, ElevenLabs Scribe v2 $0.22, Speechmatics Melia 1 batch $0.24 base, Cartesia about $0.39 at Scale, Gladia Starter async $0.61. Different vendors include different features (diarization, languages, real-time), so the rates are the base transcription price only.