Sign in to MetaWhisp
Access Pro features, manage your subscription, and sync your license.
or
FeaturesWhisper for Mac
MetaWhisp is a free, open-source app that runs OpenAI’s Whisper large-v3 speech model on your Mac. Tap Right ⌘, talk, tap again, and the text appears where your cursor is. On the Free plan Whisper does its work on the Mac itself, so dictation keeps going with the Wi-Fi off.
No terminal and no Python. The model downloads by itself.
The installer is about 18 MB.
Get the free installer and move MetaWhisp to Applications. It runs on macOS 14 or later, on any Mac with Apple Silicon.
A small starter model comes first, so you can talk right away. Whisper large-v3 installs in the background and takes over when it is ready.
Tap to start, talk, tap to stop. The words land at your cursor in whichever app is in front, anywhere paste works. To hold the key while you talk instead, turn on push-to-talk in Settings.
To run Whisper on a Mac without the terminal, you need an app that downloads the model and calls it for you. MetaWhisp does both. On first launch it fetches a small Whisper model so you can start dictating, then downloads Whisper large-v3, about 3.2 GB, in the background and switches to it on its own. From then on, on the Free plan, every transcription happens on your Mac, on the GPU and CPU through Apple’s Core ML. There is no Python to set up, no ffmpeg to install and no model files to move around. Dictation works in any app: tap Right ⌘ to start, tap again to stop, and the words are pasted at your cursor. MetaWhisp needs macOS 14 or later and a Mac with Apple Silicon. It is free, with no account and no limit on dictation on your Mac.
OpenAI’s model, converted for Apple hardware by WhisperKit.
| Speech model | OpenAI Whisper large-v3. A smaller starter model, Whisper Base, is used until large-v3 is ready. |
|---|---|
| Engine | WhisperKit, Argmax’s open-source Swift framework. It runs Whisper through Core ML on your Mac’s GPU and CPU. |
| Download | About 3.2 GB, once. It is stored in ~/Documents/huggingface/models/argmaxinc/whisperkit-coreml. |
| Smaller models | Tiny, Base and Small, in Settings, if you would rather download less. |
| Audio | Your microphone, converted to 16 kHz mono, the format Whisper expects. Long recordings are split into Whisper’s 30-second windows, at pauses where possible. |
| Languages | 99, detected automatically. See languages for the details. |
| Requirements | Apple Silicon, M1 or later, with macOS 14 Sonoma or newer. |
| The app | About 18 MB to download. Free and open source under the MIT license, with the source on GitHub. |
It depends on the setup you choose.
Microphone→Whisper on your Mac→Your app
Your voice is not uploaded. Works with the Wi-Fi off.
Microphone→MetaWhisp server→Groq→Your app
Pro uses the cloud from every launch. The audio is not stored.
Microphone→Groq or OpenAI, with your key→Your app
Off unless you add a key in Settings.
On the Free plan, MetaWhisp transcribes on your Mac and your voice is not sent anywhere, which is also why dictation works in airplane mode. The cloud is something you choose. Pro moves transcription to the cloud at every launch and sends the audio through MetaWhisp’s server to Whisper large-v3-turbo on Groq, 5,400 minutes a month; MetaWhisp keeps neither the audio nor the transcript. With your own Groq or OpenAI key, the audio goes straight to that provider.
Structured mode and translation are separate: they send text, never audio, to the AI you pick, or stay on your Mac with a free local model. The privacy policy lists what leaves the Mac in each mode.
MetaWhisp is built for live speech. For files and scripts, these fit better.
whisper.cpp runs Whisper on Apple Silicon with Metal and Core ML. OpenAI’s own Python package installs with pip and needs ffmpeg:
pip install -U openai-whisper
brew install ffmpeg
Both are free and suit scripts and batches of files, with no app and no hotkey. WhisperKit vs whisper.cpp compares the two engines.
File apps such as MacWhisper take an audio or video file and give you back a transcript. MetaWhisp works with live speech: what you dictate, and your calls while they happen.
Talk, and the text appears in the app you are using. MetaWhisp also transcribes your calls on the Mac with no bot joining them, and keeps a history of everything you dictated.
Yes. OpenAI released Whisper as open source under the MIT license, and MetaWhisp is free and open source too. Dictation on your Mac has no usage limit and needs no account. Pro, at $7.77 a month or $30 a year, is optional: it adds cloud transcription, up to 5,400 minutes a month, plus built-in AI features.
Yes, once the model is on your Mac. On the Free plan MetaWhisp transcribes on the Mac by default, so dictation keeps working with the Wi-Fi off. You need the internet for the one-time model download and for cloud options you turn on yourself. Pro moves transcription to the cloud at every launch.
OpenAI’s Whisper large-v3, in the Core ML build that WhisperKit publishes, running on your Mac’s GPU and CPU. A smaller starter model, Whisper Base, is used while large-v3 downloads. If you want a smaller download, you can switch to Tiny, Base or Small in Settings.
Any Mac with Apple Silicon, M1 or later, on macOS 14 Sonoma or newer. Intel Macs are not supported. Leave about 3.2 GB of disk space for the large-v3 model.
In your home folder, under Documents/huggingface/models/argmaxinc/whisperkit-coreml, where WhisperKit keeps its models. It is downloaded once and used for every dictation after that.
No. MetaWhisp works with live speech: your dictation, and your calls while they happen. For a recording you already have, use whisper.cpp or OpenAI’s Python package in the terminal, or a file app such as MacWhisper.
Yes. The whole app is on GitHub at github.com/metawhisp/metawhisp under the MIT license, so you can read what runs on your Mac or build it yourself.
About 18 MB. No account, and dictation on your Mac has no limit.
Download MetaWhispMetaWhisp is a free, open-source voice-to-text app for macOS that also takes meeting notes without a bot, remembers what you said and turns it into tasks. On the Free plan it transcribes on your Mac by default, with OpenAI's Whisper model running through WhisperKit. Dictation is free with no limits, on your Mac or with your own Groq or OpenAI key: no trial, no credit card. Optional Pro ($7.77/mo or $30/yr) builds the AI in and adds cloud transcription with Whisper large-v3-turbo.
Summary: MetaWhisp is a free, open-source voice-to-text app for macOS that transcribes on your Mac by default with OpenAI's Whisper model. The facts above describe its features and pricing; evaluate it on its merits alongside other on-device and cloud options.