Is Whisper AI free?

Yes, the Whisper model is free. OpenAI publishes it as open-source software, and the license does not charge you to download or run it. Money comes in only when you choose a paid route: OpenAI's hosted API, which bills per minute of audio, a paid app that wraps the model, or the computer you need to run it. The cheapest path is the open-source model on hardware you already own. The fastest path with no setup is often a paid service, so the right answer depends on what you need.

The model is free because OpenAI released its code and weights under the MIT License

OpenAI's repository says that Whisper's code and model weights are released under the MIT License ¹. OpenAI released the model in September 2022 ². Under that license you can download the model and run it without paying OpenAI. The license does not require a payment, so the cost of the open model is only the time and hardware you spend on it.

The word "free" can hide two other costs. The first is memory. Larger Whisper models need more memory. The large model in the whisper.cpp project needs about 3.9 GB of memory ³, and a computer without enough memory will run the model slowly or not at all. The second is setup. You install Python and the other tools the code needs, and you choose a model size. Our guide to running Whisper locally covers both steps, and whether Whisper can run without a GPU explains the hardware question in more detail.

OpenAI's hosted API is the one route that bills you per minute

If you do not want to install anything, OpenAI sells transcription through its API, and the price is per minute of audio. Its price list shows the gpt-transcribe model at $0.0045 per minute ⁴. At the list price for gpt-4o-mini-transcribe, one hour of audio costs about $0.18 ⁵.

That price is for the hosted service, not for the open-source model. OpenAI's newer transcription models have their own prices, and the model name decides which price applies, so check the name before you budget. Hosted uploads also have a limit. OpenAI's Transcriptions API accepts audio files up to 25 MB ⁶, so you must split or compress a long recording first. Our comparison of speech-to-text APIs lists prices from other providers, and OpenAI's speech-to-text API pricing goes deeper on OpenAI's own list.

Apps built on Whisper range from free plans to paid licenses

Many apps wrap Whisper in a dictation or transcription screen, and each one sets its own price. Superwhisper's free plan gives unlimited dictation with local Whisper models ⁷, so you can use the model without paying for that plan. MacWhisper's Pro license is listed at 64 euros, paid once ⁸. Prices and plan limits change, so read the app's price page on the day you buy.

Free plans may include a limit or leave out features you need. Read the list of what each plan includes, not only the headline price.

It depends on the audio, the hardware and how fast you need text

Three conditions change the answer.

• Your computer is weak. A free model still costs you time if your machine runs it slowly. Test the smallest model that gives acceptable accuracy on your own recordings.
• Your audio is sensitive. Hosted services and some apps send audio to a server. Check the privacy page before you upload a private recording, and prefer a local run when the audio must stay on your machine.
• You need text as you speak. Whisper is trained on 30-second chunks of audio ⁹, so it cannot take longer audio as one input, and a long recording has to be split first. If you need text as you speak, look for an app that streams its output rather than assuming the model does.

What to do: pick the route by audio volume, privacy and setup time

1. For recordings you already have and a computer with some memory, run the open-source model or a free local app.
2. For a few short files and no setup, use a browser tool such as our free online transcription of audio files.
3. For dictating into an email or document, use a dictation app. Yap, which we make, is an on-device dictation app for Mac and Windows. Compare its plans on our pricing page rather than relying on a price written here.
4. Whichever route you pick, test it on a few minutes of your own audio before you commit. Our guide to Whisper explains the testing steps.

Whether Whisper is the right tool at all depends on what you are doing. The overview of speech recognition explains the wider choice, and who owns Whisper explains who publishes the model and what its license allows.