Whisper (OpenAI) review: free to self-host, $0.006 a minute on the API
Whisper is an open-source model rather than a subscription product, so there is no plan to buy and no list price to quote. Our check of github.com/openai/whisper on 12 June 2026 recorded it as free to download and run locally, with the OpenAI API offering the same transcription at $0.006 per minute of audio. That check is four months old.
The short answer
Buy Whisper Use Whisper if you are comfortable at a command line and would rather spend hardware than subscription. Our 12 June 2026 check of github.com/openai/whisper recorded no price at all for local use: you download the model and transcribe on your own machine, and the only money figure on the card is the OpenAI API at $0.006 per minute of audio.
Skip it if you want to upload a file in a browser and get a tidy transcript back. There is no interface, no account and no support line here, and Happy Scribe at $17 a month on Basic, checked 13 June 2026, or Otter.ai at $16.99 a month on Pro, checked 15 June 2026, sell exactly the convenience Whisper leaves out.
Where Whisper fits
Whisper is not a product with tiers, and the pricing section of this page says so plainly rather than inventing a plan. It is an open-source speech recognition model published on GitHub, and our 12 June 2026 check recorded no subscription, no monthly fee and nothing to buy for local use. What you supply instead is a computer and the willingness to run software from a terminal.
The licence is the thing that makes the zero real. Because the model is open source and distributed from OpenAI's own repository, you can download it, run it as often as you like and transcribe without a quota, and no vendor can raise a price on you later. Equally, no vendor owes you a fix, an uptime commitment or an answer to a support ticket.
Self-hosting is where the cost reappears, in hardware rather than invoices. Running the model locally wants a capable graphics card, and on a processor alone the same recording takes considerably longer, which turns a background task into an afternoon. Our snapshot records that tradeoff in words and puts no number on it, so this page puts none on it either.
That is the honest shape of the compute cost: it is yours, it varies with your machine and the model size you choose, and we do not estimate it. If you rent a cloud instance with a graphics card to do the work, the bill comes from your cloud provider and not from Whisper, and our 12 June 2026 check records no figure for that either.
The OpenAI API is the paid route and the only money figure we hold. It was $0.006 per minute of audio on our check, which is six tenths of a cent a minute, so an hour of audio is about thirty-six cents and a hundred hours comes to about thirty-six dollars at that rate. There is no subscription attached, so a month with nothing to transcribe costs nothing.
Against the subscription tools the comparison is not mainly about money, because the API rate ends that argument on volume alone. Otter.ai Pro was $16.99 a month on 15 June 2026, which is our oldest pricing evidence and worth flagging whenever it is cited. Happy Scribe Basic was $17 a month for 120 minutes on 13 June 2026, and Rev sells the human accuracy no model offers at $1.99 a minute on 6 October 2026. What those three sell is an interface, support and a service you can complain to.
This evidence was recorded on 12 June 2026 and has not been checked since, which makes it four months old. A licence is not the sort of thing that changes quietly, but an API rate is, so confirm the per-minute price before you build anything around it. The transcription and captions category holds the hosted alternatives.
Whisper pricing
There is no subscription here and no list price for the software itself. Our check of github.com/openai/whisper on 12 June 2026 recorded Whisper as free and open source for local use, with one paid route, the OpenAI API at $0.006 per minute of audio. Because the software is free, the facts that follow are about the licence, the hardware and the compute rather than about billing terms.
| Plan | Price | What it adds |
|---|---|---|
| Local, open source | Free | Download and run on your own machine. No subscription, no quota, no account. A capable graphics card is wanted |
| OpenAI API | $0.006/min | Per minute of audio, no subscription. An hour of audio is about thirty-six cents at that rate |
No plan drives a price for local use, so what follows is what the free licence does and does not cover:
- No tiers and no quota. Our 12 June 2026 check records free local use with nothing gated, so there is no upgrade to buy and no allowance to exhaust.
- No seats. The model is open source, so running it for a team of ten costs the same as running it for one, which is nothing.
- Your own hardware is the budget line. A graphics card makes the work practical and a processor alone makes it slow, and our snapshot puts no figure on either.
- Cloud compute if you rent it. That cost comes from your provider rather than from Whisper, and because our check records no rate for it, this page quotes none.
- The API route if you would rather not host anything, at $0.006 per minute of audio on our 12 June 2026 check, billed on what you send with no monthly fee underneath.
Whisper against the paid transcription tools
Whisper is the floor this category is measured against, and the dates behind these rows run from June to October 2026. Otter.ai Pro was $16.99 a month, verified 15 June 2026, which is the oldest check on the site and should be read with that gap in mind. Happy Scribe Basic was $17 a month for 120 minutes, verified 13 June 2026, with human transcription at $2.00 a minute on top. Sonix was $10 per hour of audio pay-as-you-go, verified 14 June 2026. Rev was $25.49 a month on Essentials with human work at $1.99 a minute, verified 6 October 2026. What every one of them sells against a free model is an interface, support and somebody to answer for the result.
What the Whisper licence and hardware mean in practice
There are no billing terms for local use, so these are the licence, hardware and compute facts that take their place.
- Local price
- Free, as recorded on 12 June 2026. No subscription, no monthly fee and nothing to buy.
- Licence
- Open source, published in OpenAI's own GitHub repository and free to download.
- Quota
- None recorded for local use. Transcribe as much as your machine will process.
- Hardware
- A capable graphics card is wanted. A processor alone works and is much slower.
- Compute cost
- Yours, and not recorded in our snapshot, so this page states no figure for it.
- API route
- $0.006 per minute of audio, billed per use with no subscription.
- Check date
- 12 June 2026, four months old and not re-verified since.
Whisper is free and open-source when run locally. You download the model and transcribe on your own computer at zero cost. The OpenAI API offers Whisper transcription at $0.006/minute. There's no subscription, no monthly fee: just free local processing or per-minute API cost. The tradeoff: local usage requires technical setup and a capable GPU. The API is simpler but costs money per minute.
A page about free software still needs its date, so here it is again: 12 June 2026, from github.com/openai/whisper. The licence is unlikely to have moved, the API rate might well have, and the repository and OpenAI's own pricing page win over this page if they differ. Our creator tool cost calculators will show what a year of a hosted tool would cost instead, which is the only sum worth running here.
What Whisper gives up
There is no interface. Whisper takes a file and returns text, and everything a hosted tool wraps around that, the editor, the library, the sharing, the exports, is yours to assemble or do without.
There is no support and no service commitment, which is the real price of the zero. When a transcription job fails at midnight there is a community rather than a vendor, and for some buyers that alone justifies paying a hosted vendor instead.
Setup is a genuine barrier rather than a formality. A non-technical creator will not get from a download page to a transcript without help, which is why wrapper applications exist and why they cost more than running the model yourself.
Without a graphics card the local route is slow enough to change how you work, turning a transcript you wanted in minutes into something you queue and leave. The hardware you already own therefore decides whether free is practical, and our snapshot puts no figure on the upgrade.
Accuracy falls on less common languages, heavy accents and overlapping speakers, as it does across this category, and there is no paid human pass to escalate to. Rev sells that at $1.99 per audio minute, verified 6 October 2026, and Whisper has no equivalent at any price.
Subtitle files are not a finished output. The model produces timed text and turning it into the caption formats a platform expects takes a script or a third-party tool, which is work a hosted service would have done for you.
Our evidence is from 12 June 2026, four months old at the time of writing. The licence is stable by nature, but an API rate quoted per minute of audio is the kind of figure that moves, so verify $0.006 before you budget a pipeline around it.
How Whisper compares
The three tools below are the hosted alternatives weighed against Whisper, and the axis is cost against convenience. Prices are each vendor's own published rates with the date we checked them, and those dates run from June to October 2026, so the rows are not strictly comparable. The wider field sits in the transcription and captions category.
| Tool | Best for | Price | Where it breaks |
|---|---|---|---|
| Whisper (OpenAI) | Technical users who want transcription at near zero cost | Free locally, $0.006 per minute via the API (June 2026) | No interface, no support, and hardware decides the speed |
| Otter.ai | Live meeting and interview notes with no setup | Free, $16.99/mo Pro (June 2026) | No editing, and our check is the oldest on the site |
| Happy Scribe | Transcripts and subtitles from one subscription | $0 free, $17/mo Basic for 120 minutes (June 2026) | Monthly minutes expire unused |
| Rev | Transcripts that have to be right | Free, $25.49/mo Essentials, $1.99/min human (October 2026) | Human minutes make a budget unpredictable |
How to test Whisper before you commit to it
There is nothing to sign up for, so the test is whether you can run it and whether your machine is fast enough. Our 12 June 2026 check recorded free local use and an API rate of $0.006 per minute of audio, and that rate is the figure to confirm first.
- Confirm the current API rate on OpenAI's own pricing page. Our $0.006 per minute figure is from June 2026 and is the one number here that moves.
- Install the model locally and transcribe five minutes of your own audio. If that step defeats you, the free route is not free in any sense that matters.
- Time a full-length recording on your own hardware. The difference between minutes and hours is what decides whether local use fits your week.
- Run the same file through the API and compare the time and the result. Five minutes of audio costs three cents at $0.006 a minute, which is a cheap comparison to run.
- Add up your monthly audio minutes and price them at $0.006 each, then set the total against a subscription tool before assuming the model wins.
- Test your hardest material, with accents and crosstalk in it, because a free transcript you have to retype is not a saving.
- If the answer is that you want an interface and someone to ask, price Happy Scribe at $17 a month on Basic, verified 13 June 2026, and let the cost calculators total a year of it.
Whisper questions buyers ask
Is Whisper really free?
Yes: the open-source model is completely free. Download it, run it on your computer, and transcribe unlimited audio at zero cost. The OpenAI API charges $0.006/minute. Local use requires technical setup and a capable computer; the API requires an OpenAI account and API key.
How accurate is Whisper transcription?
For clear English audio with good microphone quality, Whisper achieves 95%+ accuracy: competitive with human transcription. The large model reduces errors 10-20% compared to smaller models. Accuracy drops for non-English languages, heavy accents, noisy audio, and overlapping speakers.
Do I need a GPU to run Whisper?
For practical use: effectively yes. Whisper runs on CPU but is 5-10x slower. A 60-minute recording can take 2-10 hours on CPU versus 5-15 minutes on GPU. An NVIDIA GPU with 4+ GB VRAM (medium model) or 10+ GB VRAM (large model) is recommended. Without a GPU, use the API ($0.006/minute) instead.
Whisper vs Rev: which is better?
Rev offers human transcription, which no model does. Our check of Rev's pricing on 6 October 2026 recorded Rev human transcription at $1.99 per audio minute, alongside a Rev free plan with 45 AI minutes a month, Rev Essentials at $25.49 a month and Rev Pro at $47.99 a month. Whisper is free to run locally and $0.006 per minute of audio through the OpenAI API, as recorded on 12 June 2026, with accuracy that holds up well on clean audio. Choose Rev for critical content where every word matters and you want a person accountable for it. Choose Whisper for routine transcription at a much lower cost, provided you are comfortable with the setup.
Whisper vs Otter.ai: which is better?
Otter.ai provides a polished interface with real-time transcription, meeting integration and AI summaries. Our check of Otter.ai's pricing on 15 June 2026 recorded Otter.ai Pro at $16.99 a month, or $8.33 a month billed annually, with an Otter.ai free plan carrying 300 minutes a month. That check is the oldest pricing evidence we hold, so read it as four months old. Whisper provides raw transcription with no interface at all, free locally or $0.006 per minute of audio through the OpenAI API as recorded on 12 June 2026. Choose Otter.ai for ease of use and meeting integration. Choose Whisper for cost if you are technically comfortable.
How many languages does Whisper support?
Whisper supports 99 languages for transcription. English accuracy is strongest. Major languages (Spanish, French, German, Japanese, Portuguese) are strong. Less common languages have lower accuracy due to less training data. Test your specific language before relying on Whisper for production content.
Can non-technical people use Whisper?
Not directly: Whisper has no graphical interface and requires command-line or API knowledge. However, several consumer apps use Whisper under the hood: MacWhisper (Mac), WhisperDesktop, and various web-based tools. These provide user-friendly interfaces powered by Whisper's model, though they cost more than running Whisper directly.
What file formats does Whisper support?
Whisper supports mp3, mp4, mpeg, mpga, m4a, wav, and webm. The API has a 25 MB file size limit, so longer recordings need to be split or compressed. Local use has no file size limit. You can transcribe files of any length on your own hardware.
Is Whisper better than paid transcription services?
For accuracy: Whisper matches or exceeds most AI transcription services and approaches human-level for clear English audio. For ease of use: paid services win decisively. For cost: Whisper wins (free locally, $0.006/min via API). The 'better' answer depends on whether you value accuracy and cost (Whisper) or convenience (paid services).
Can Whisper generate subtitles and captions?
Whisper outputs timestamped text that can be formatted as SRT or VTT subtitle files. The raw output includes word-level timing. You'll need a script or tool to convert to subtitle format. Whisper doesn't produce finished SRT files directly. Tools like WhisperX add proper subtitle formatting on top of Whisper's output.