Analysis
Whisper API Pricing in 2026: Cost per Audio Hour, Compared
MyAi Network · · 6 min read
Speech-to-text vendors quote prices per minute, per hour, per second or per request, which makes them hard to compare. Below, every list price is converted to one unit, dollars per audio hour, for batch or pre-recorded audio (not live captions). We run one of these services, so we have tried to be fair: where a competitor is cheaper than us, the table says so.
The table
List prices from each vendor's pricing page, checked October 13, 2026. Prices change; check the source before you commit.
| Service | Per audio hour | Notes |
|---|---|---|
| Groq | $0.04 | whisper-large-v3-turbo; one file per request, 10 s minimum, 25 MB (free) or 100 MB (dev tier) per file |
| MyAi | $0.10 | whisper-large-v3-turbo; batches of audio URLs, files up to 4 hours, per-second billing, 100 free hours |
| AssemblyAI | $0.15–$0.21 | Universal-2 to Universal-3.5 Pro; speaker labels extra |
| OpenAI gpt-4o-mini-transcribe | $0.18 | $0.003 per minute |
| Rev.ai Reverb | $0.20 | English; other languages $0.30 |
| Deepgram Nova-3 | $0.22–$0.26 | Pre-recorded, monolingual; Growth plan to pay-as-you-go |
| OpenAI gpt-4o-transcribe | $0.36 | $0.006 per minute |
Sources: Groq, AssemblyAI, OpenAI, Rev.ai, Deepgram, MyAi.
What the per-hour number leaves out
Minimums and billing increments
A 10-second minimum per request costs nothing on hour-long podcasts and a lot on a folder of 3-second voice notes. Rev.ai bills per second with a 15-second minimum; Groq bills a 10-second minimum; MyAi bills to the second of accepted audio. If your files are short, work out the cost on your real file lengths, not the headline rate.
File size and length limits
Per-request APIs cap the upload size (Groq: 25 MB on the free tier, 100 MB on the dev tier), so a long WAV recording has to be compressed or cut into pieces first, and you stitch the transcripts back together yourself. Batch services take a URL and do the splitting. Neither is wrong; it depends on whether you would rather write that code or pay a few cents more per hour.
Features you would pay extra for
- Speaker labels (diarization). Extra on AssemblyAI, included on Deepgram's Growth plan, not available on MyAi yet.
- Caption files. SRT and VTT output saves a conversion step for video and accessibility work.
- Word timestamps. Needed for search-inside-audio and clip editors; check whether they cost more.
- Languages. Rev.ai charges more outside English; Whisper models are multilingual, but accuracy varies by language.
Where your audio goes
Read the data terms, not just the price. MyAi runs on GPUs operated by independent operators, which is how the price stays low, and it is why we ask people not to send confidential, medical, legal or regulated recordings. For that kind of audio, use a vendor whose terms fit your obligations, or run Whisper on your own hardware.
Which one to pick
- Lowest price per hour, files under the size cap: Groq.
- A backlog of long public recordings sent as URLs, captions included: MyAi (100 free hours to try it).
- Speaker labels and extra speech features: AssemblyAI or Deepgram.
- Confidential audio and a GPU on hand: run Whisper yourself.
FAQ
What is the cheapest Whisper API per audio hour?
Of the list prices we checked in October 2026, Groq's whisper-large-v3-turbo is the lowest at $0.04 per audio hour, with a 10-second minimum per request and a per-file size cap. MyAi is $0.10 per audio hour for batches of URLs. AssemblyAI, OpenAI, Rev.ai and Deepgram list between $0.15 and $0.36.
How much does it cost to transcribe 1,000 hours of audio?
At list prices: about $40 on Groq, $100 on MyAi ($90 after the 100 free hours), $150 to $210 on AssemblyAI, $180 on gpt-4o-mini-transcribe, $200 on Rev.ai Reverb (English), $220 to $260 on Deepgram Nova-3 and $360 on gpt-4o-transcribe. Add-ons such as speaker labels cost extra where they exist.
Is per-minute or per-hour pricing better?
They are the same thing at different scales: multiply a per-minute price by 60. What matters more is the billing increment and the minimum. Per-second billing with no minimum is cheapest for many short clips; a per-request minimum adds up when you send thousands of tiny files.
Should I just run Whisper myself?
If you already own a capable GPU and have the time, running open-source Whisper locally (for example with faster-whisper or whisper.cpp) costs only electricity, and it is the right choice for confidential audio because nothing leaves your machine.