gpt-transcribe at $0.0045 a minute vs Scribe v2 and Nova-3
Of the three, Scribe v2 lists lowest at $0.22 an hour, then Nova-3 mono at $0.258 and gpt-transcribe at $0.27. Per-hour table and what Sume bills.

Among the three in the title, the cheapest listed price for completed recordings is ElevenLabs Scribe v2 at $0.22 an hour. Deepgram's pre-recorded Nova-3 mono is $0.258 an hour, and OpenAI's gpt-transcribe at $0.0045 a minute is $0.27 an hour. The spread is small, about 23% from lowest to highest of the three, so features and accuracy on your audio matter more than the rate.
Per-minute and per-hour prices
Per-hour figures are the listed per-minute price times 60, or the listed hourly price. OpenAI marks some rates with a tilde, which I copy.
| Model | Listed price | Per hour |
|---|---|---|
| ElevenLabs Scribe v2 | $0.22 per hour | $0.22 |
| Deepgram Nova-3 mono | $0.0043 per minute | $0.258 |
| Deepgram Whisper Large | $0.0048 per minute | $0.288 |
| Deepgram Nova-3 multilingual | $0.0052 per minute | $0.312 |
| OpenAI gpt-4o-mini-transcribe | about $0.003 per minute | about $0.18 |
| OpenAI gpt-transcribe | $0.0045 per minute | $0.27 |
| OpenAI whisper-1 | $0.006 per minute | $0.36 |
| OpenAI gpt-4o-transcribe | about $0.006 per minute | about $0.36 |
| OpenAI gpt-4o-transcribe-diarize | about $0.006 per minute | about $0.36 |
Which OpenAI model for which job
OpenAI's transcription guide separates the models by job, which affects the price you actually pay.
- gpt-transcribe: completed recordings.
- gpt-live-transcribe: live streams, $0.017 a minute, so about 3.8 times the price of gpt-transcribe (0.017 / 0.0045).
- gpt-4o-transcribe-diarize: required for speaker labels.
- whisper-1: timestamps, subtitle formats and translation to English.
- Prompts, keywords and language hints improve accuracy.
Cost of 10 hours of audio
Arithmetic on the per-hour column. If you need subtitles or speaker labels, the model that provides them sets your price, not the cheapest row.
| Service | 10 hours |
|---|---|
| Scribe v2 | $2.20 |
| Deepgram Nova-3 mono | $2.58 |
| OpenAI gpt-transcribe | $2.70 |
| OpenAI whisper-1 | $3.60 |
| Sume STT 1.0 | $6.00 (transcript line only) |
What Sume bills
Sume's Video inspect can transcribe the audio of a clip you already host with transcribe: true. The public rate is $0.01 per audio minute, which is $0.60 an hour. That is well above the $0.22 an hour listed for Scribe v2, and the Sume docs do not break the $0.01 down against any vendor's list price. Each inspect also bills its own compute on top of the transcript line, so the all-in clip price is higher still.
It is a clip tool, not a bulk-transcription API: it reads one Sume-hosted clip per job, reserves one minute unless you pass duration_seconds (up to 600), and accepts a language_code hint such as ko. Sume does not offer gpt-transcribe or Deepgram. For a backlog of long recordings, a dedicated STT API is cheaper on the listed numbers; for transcripts attached to video you already process in Sume, the flat $0.01 a minute is simple to budget.
Sources
Related posts
More in Pricing
- Grok Imagine image prices: $0.02, $0.04, $0.05 and Sume's one row
xAI lists three Grok image prices ($0.02, $0.04, $0.05). Sume lists one x-ai/grok-image row; read its live endpoints pricing array for the billed figure.
- Grok Imagine video cost: a 15-second clip is $1.20 at xAI's 1.5 rate
xAI's video guide says duration and resolution both set the cost, with 15 seconds the maximum. At $0.080 per second a 15 s clip is $1.20.
- H3 2K at $1.63 vs Omni 1080p at $1.88: 10-second price check
On Sume's price list, a 10-second H3 2K upscale is $1.63, Omni Flash 1.1 at 1080p is $1.88 and 4K is $3.75, H3 4K $2.00. What each row lists today.
- H3 Max lip sync: why 5.1 seconds of audio is priced as 6 seconds
Sume prices H3 Max Lip Sync at ceil(duration_seconds) x the per-second rate: 5.0 s costs $0.50 at 768p, 5.1 s costs $0.60. Table of ceil steps and a trim tip.
Written by Sume