Four speech engines, all on your Mac
Whisperer ships with Whisper, Parakeet, Qwen3 ASR and WhisperKit. The default writes 97 languages on the GPU, and none of them needs the internet.
$2.99 once on the Mac App Store. Apple Silicon, macOS 26 Tahoe.
Default. About 547 MB. All engines run on this Mac. Nothing is sent to a server.
What each one is for
The default is picked for you. The rest are there when you want to try something else.
| Engine | Model | Languages | Runs on | Good to know |
|---|---|---|---|---|
| WhisperDefault | Whisperer V3 (Large V3 Turbo) | 97 languages | Metal GPU | Default. About 547 MB. |
| Parakeet | NVIDIA Parakeet | 25 languages | Neural Engine | Runs on the Neural Engine, leaving the GPU free. |
| Qwen3 ASR | Qwen3 ASR | 31 languages | On device | A second open model to try when Whisper stumbles. |
| WhisperKit | Whisper on Core ML | Whisper languages | Neural Engine | Whisper models through Core ML. |
Nothing to sign into, nothing metered
- Works with Wi-Fi off
- On a plane, on hotel Wi-Fi, in a building with no signal.
- Models stay on your Mac
- Downloaded once into your Library folder, then used from disk.
- Metal GPU
- Whisper runs through whisper.cpp on the GPU of your Apple Silicon chip.
- Neural Engine
- Parakeet and WhisperKit run on the Neural Engine.
- No per-minute fees
- Dictate for an hour or a year. Nothing is metered.
- No account
- No sign-up and no analytics. The App Store lists Data Not Collected.
What offline costs you
A model on your Mac takes disk space and a short download the first time. The default is about 547 MB. Cloud dictation apps skip that step, and in return every word you say goes to their servers.
Speed depends on your chip. An M1 MacBook Air handles everyday dictation fine; a newer Pro chip finishes long recordings and file transcriptions faster.
Engine questions
Dictate with the Wi-Fi off
Whisperer is $2.99 once. Pro adds meetings, your own AI and MCP for $14.99 once. No subscription.