Otter vs Rev vs Descript vs free local transcription: cost per hour compared
What an hour of transcription costs on Otter, Rev, Descript, the speech APIs, and a model running on your own laptop. Every price checked 15 September 2026.
An hour of automatic transcription costs between $0.31 and $2.55 on the popular subscription services once you divide the monthly fee by the minutes you actually use, $0.15 to $0.36 through the cloud speech APIs, and about a cent of electricity when the model runs on your own computer. Human transcription is a different product at $119.40 an hour. The rest of this post shows the working, with every price read off the vendor’s own pricing page on 15 September 2026.
The short table
| Option | Published price | Cost per hour of audio |
|---|---|---|
| Otter Pro, annual | $8.33/user/month for 1,200 minutes | $0.42 if you use every minute |
| Otter Pro, monthly | $16.99/user/month for 1,200 minutes | $0.85 if you use every minute |
| Rev Essentials, annual | $25.49/seat/month for 5,000 minutes | $0.31 at full use, $2.55 at 10 hours a month |
| Descript Hobbyist, annual | $16/month for 10 hours | $1.60 |
| Descript Creator, annual | $24/month for 30 hours | $0.80 |
| Rev human transcript | $1.99/minute | $119.40 |
| OpenAI Whisper or gpt-4o-transcribe | $0.006/minute | $0.36 |
| OpenAI gpt-4o-mini-transcribe | $0.003/minute | $0.18 |
| Deepgram Nova-3, monolingual | $0.0043/minute | $0.26 |
| AssemblyAI Universal-2 | $0.15/hour | $0.15 |
| Local model in your browser | $0 | roughly $0.01 in electricity |
“Cost per hour” for the subscriptions assumes you use the whole allowance. Almost nobody does, which is the first thing the marketing pages do not mention.
Otter
Otter’s free Basic plan gives 300 transcription minutes a month, caps any single conversation at 30 minutes, and allows three audio or video file imports for the lifetime of the account. Pro is $16.99 a user a month, or $8.33 a month on an annual plan, for 1,200 in-app recording minutes, meetings up to 90 minutes and ten file imports a month. Business is $30 a month, or $19.99 annually, with unlimited meetings and recordings and meetings up to four hours.
On Pro, the number to watch is file imports rather than minutes. Ten a month is the ceiling if what you bring is recordings you already have, and the 1,200 minutes then go mostly unspent. A podcaster with weekly episodes and a couple of interviews is already at the import limit. At ten one-hour files a month on the annual plan the real cost is $0.83 an hour, and on the monthly plan $1.70.
Rev
Rev sells AI transcription by the seat. The free plan includes 45 AI transcription and caption minutes a month and lets you analyse up to five files at a time. Essentials is $25.49 per seat per month on an annual plan, $29.99 month to month, with 5,000 AI minutes per user per month and up to three users. Pro is $47.99 per seat annually, $59.99 monthly, with 10,000 minutes and up to five users. Human transcription is $1.99 a minute, which Rev publishes as 99%+ accurate and delivered in 12 hours or less.
Five thousand minutes is 83 hours. Used in full, Essentials is the cheapest subscription here at $0.31 an hour. Use ten hours a month and the same seat works out at $2.55 an hour, which is above every API rate further down and above Descript’s Creator plan. How much of the allowance you actually use moves the real price by a factor of eight, which is more than any difference between the vendors.
The human tier belongs in a row of its own. At $119.40 an hour it is not priced against software. It is for court, medical and broadcast work where somebody has to be accountable for every word, and where a wrong name costs more than the transcript.
Descript
Descript counts media hours rather than transcription minutes. The free plan includes 60 media minutes a month and 100 one-time AI credits. Hobbyist is $16 a month billed annually, $24 month to month, for 10 hours. Creator is $24 annually, $35 monthly, for 30 hours. Business is $50 annually, $65 monthly, for 40 hours.
Per hour on the annual plans, that is $1.60, $0.80 and $1.25. Creator is the one to compare, because 30 hours is a realistic ceiling for a working podcaster and it is the cheapest of the three per hour. The allowance is published in media hours rather than finished minutes, so work out how your raw recordings map onto it before picking a tier. A two-hour session that becomes a 40-minute episode is still a two-hour file going in.
The APIs underneath
The same job, bought directly from a speech API, has a published price too. OpenAI lists $0.006 a minute for Whisper and for gpt-4o-transcribe, and $0.003 a minute for gpt-4o-mini-transcribe. Deepgram lists Nova-3 pre-recorded at $0.0043 a minute for monolingual audio and $0.0052 for multilingual. AssemblyAI lists asynchronous transcription at $0.15 an hour for Universal-2 and $0.21 for Universal-3.5 Pro, and notes that add-ons such as speaker diarisation stack on top of the base rate.
That is a range of $0.15 to $0.36 an hour. You need to write code to use any of them, and the audio still leaves your machine, so they are not a consumer option. What they do is set the floor. When a subscription works out at $2.55 an hour, the gap is what you are buying besides the recognition: an interface, storage, editing, search, summaries and somebody to email. Our cost per minute reference keeps the same rates in one place.
Local transcription
OpenAI’s open-source Whisper runs inside a modern browser using your graphics card. There is no per-minute price because there is no server. The costs that remain are a one-time download of about 200 MB for the model, which the browser caches, and electricity.
The electricity figure is small. On the desktop we test on, an hour of audio takes about 40 minutes of compute at roughly 60 W of extra draw, which is 0.04 kWh, or about one cent at European domestic rates. A laptop is slower and draws less, so it comes out similar. Why no cap is needed walks through what happens to a site’s economics when that per-minute cost goes to zero.
What you give up is real, and it belongs in the comparison:
- It needs desktop Chrome or Edge with a working graphics adapter. Elsewhere it falls back to a slower CPU mode, and the browser support page keeps the current list.
- It is English only at the moment.
- It runs the base model, chosen so it works on ordinary hardware rather than for top accuracy. On our clean synthetic test clip it made no word errors, on the same clip with noise it made one, and names, technical vocabulary and heavy accents are the weak spots; the examples page shows the output. Read the output before you rely on it.
- On our 2021 test laptop, using its integrated Intel graphics, a 25-minute recording took about six and a half minutes with the tab open; the benchmarks page has the measured runs. A thin laptop takes roughly as long as the recording itself.
For notes, searchable archives, first-draft subtitles and anything you would rather not upload, those trade-offs are usually fine. For a deposition they are not. Run one of your own files through it before you cancel anything.
Three people, three answers
A student recording four one-hour lectures a month. Otter’s free plan allows three file imports for the life of the account. Descript’s free hour covers one lecture. Rev’s free 45 minutes covers none. Local transcription covers all four for nothing, and the recordings never leave the laptop.
A podcaster with 30 hours of raw audio a month. Descript Creator at $24 a month billed annually is the natural fit, because the editing is the product and the transcription rides along. Local transcription saves the $24 and does not give you an editor.
A researcher with 100 hours of interviews a year and a consent form that promises no third parties. Every subscription service is a third party. The choice is a local model or a human transcriber under a data agreement, and at $1.99 a minute the difference is roughly $12,000.
Frequently asked questions
Is free local transcription really free? Yes, in the sense that there is no fee, no account and no minute cap. The site carries a few ads to cover hosting. Your cost is a model download and some electricity.
Which option is most accurate? We are not going to rank products we do not run. What we can tell you is where this one sits: the base English model, with no word errors on our clean synthetic clip, one on the noisy version, and weaker on names, technical vocabulary and accents (see the examples page). Rev publishes a 99%+ figure for its human transcription, which is the only tier here where a person is accountable for the words.
Will these prices still be right next month? Published prices and allowances move. Every number above was read off the vendor’s own pricing page on 15 September 2026, and the honest instruction is to open the page yourself before you buy.
Can I use the cloud APIs without coding? Not directly. They are built for developers. Some desktop apps wrap them, and then you are paying per minute plus the app.
Sources, checked 15 September 2026
- https://otter.ai/pricing : Otter pricing (Basic, Pro and Business prices, minutes, meeting caps and file imports)
- https://www.rev.com/pricing : Rev pricing (free plan, Essentials and Pro per seat, AI minutes, human transcription at $1.99 a minute with 99%+ accuracy in 12 hours)
- https://www.descript.com/pricing : Descript pricing (free plan and Hobbyist, Creator and Business media hours, monthly and annual prices)
- https://developers.openai.com/api/docs/pricing : OpenAI API pricing (Whisper and gpt-4o-transcribe at $0.006 a minute, gpt-4o-mini-transcribe at $0.003)
- https://deepgram.com/pricing : Deepgram pricing (Nova-3 pre-recorded, $0.0043 a minute monolingual and $0.0052 multilingual)
- https://www.assemblyai.com/pricing : AssemblyAI pricing (Universal-2 at $0.15 an hour, Universal-3.5 Pro at $0.21, add-ons stacking on the base rate)
- Our own numbers for speed, model size, power draw and accuracy come from our test of the transcriber on this site
Prices change. If you find one that has moved, tell us and we will update the table.