What models does VoiceInk use?
VoiceInk runs open models that you choose: Whisper or NVIDIA's Parakeet. We measured Avalon against Whisper Large v3 on AISpeak-10, our benchmark of AI and technical terms. We have not tested Parakeet.
| Feature | Aqua Voice | VoiceInk |
|---|---|---|
| Speech model | Avalon (own, purpose-trained) | Open models (Whisper, Parakeet), local |
| Cost | $10/mo ($120/yr), free tier | $25-49 one-time, or free (self-compile) |
| Public benchmark | #1 on Voice Arena (Nov 2026) | Not published (model-dependent) |
| Technical-term accuracy | 97.4% (AISpeak-10, our benchmark) | Not published (model + hardware) |
| Setup | Install app; no local model download | Install or compile, pick a local model |
| Processing & security | Cloud; SOC 2 Type II audit report, ISO 27001, GDPR, HIPAA | On-device |
| Works offline | cloud | on-device |
| Open-source | No | Yes |
| Mac | Yes | Yes |
| Windows | Yes | No |
| iPhone | Yes | Yes |
| Android | Yes | No |
| Real-time + screen context | Deep Context; Realtime Mode on Max | Hardware-dependent |
| Account required | Yes | no account |
| Meeting notes | Mac, Windows, iPhone, Android | No |
| Point at your screen while you talk | Aqua Pen | No |
| Desktop pet | Aqua Pet | No |
| Access for AI agents | Aqua MCP and CLI | No |
| Best for | Accuracy, zero setup, technical/AI users, Windows + iPhone + Android | Open-source, on-device, Apple-only, pay once or free |
Accuracy without the setup
Avalon scores 97.4% on AISpeak-10, where Whisper Large v3 scores 65.1%. VoiceInk's accuracy depends on the model you pick and your Mac.
Zero setup, always current
There is no model to download, nothing to compile, and no hardware to match. The model improves on our servers, not only when you update.
Windows, screen context, one subscription
Aqua Voice runs on Mac, Windows, iPhone, and Android on one subscription and reads the active window for context. VoiceInk's desktop app is Mac-only, and it has no Android app.
Open-source and cheap
VoiceInk is open-source. It costs $25 to $49 once, or nothing if you compile it yourself.
On-device and offline
VoiceInk runs the model on your Mac. Your audio never leaves it, and it works with no internet connection.
Full control
You can read the code, swap models, and customize the app. In return, you manage the model and setup yourself.
Perfect drafts, ready to send without a single change, on Voice Arena’s independent dictation benchmark, run on a Mac in US English.Read the resultsVoice Arena
VoiceInk runs open models that you choose: Whisper or NVIDIA's Parakeet. We measured Avalon against Whisper Large v3 on AISpeak-10, our benchmark of AI and technical terms. We have not tested Parakeet.
VoiceInk is an open-source dictation app for the Mac, with an iPhone and iPad app. It runs open speech models on your machine, so your audio stays on the device.
Aqua Voice runs Avalon on its own servers, so there is nothing to download or configure, and the model keeps improving. The trade-off is a paid cloud service instead of an open-source, on-device one.
It depends on what you value. VoiceInk is open-source, runs on your Mac, and is cheap one-time (or free if you compile it yourself). Aqua Voice is a managed service on its own model, Avalon, with published technical-term results (97.4% on AISpeak-10), zero setup, Windows and iPhone and Android support, and ongoing improvements. If you want an open-source, on-device Mac app, VoiceInk is great; if you want managed dictation with published model results with nothing to configure and Windows support too, Aqua Voice is the stronger choice.
They're built on opposite philosophies. VoiceInk is an open-source Mac app running open speech models like OpenAI's Whisper and NVIDIA's Parakeet locally on your machine. Aqua Voice is a managed cloud app running Avalon, its own purpose-trained model, with screen-context awareness on Mac and Windows and Realtime Mode on Max, plus iPhone and Android support. Both dictate into any app with a hotkey.
We have not established an app-to-app accuracy advantage. VoiceInk runs general-purpose open models (Whisper, Parakeet), and its accuracy depends on which model and Mac hardware you choose. Aqua Voice runs Avalon, trained specifically for technical vocabulary: it scores 97.4% on AISpeak-10 (our benchmark of AI and technical terms) where Whisper Large v3 scores 65.1%, and Aqua Voice ranked #1 on Voice Arena's independent dictation benchmark in November 2026.
No, VoiceInk is cheaper. VoiceInk is a one-time purchase ($25 Solo, $39 Personal, $49 Extended), and it's free if you compile the open-source build yourself. Aqua Voice has a Free plan (2,000 lifetime words), then Pro is $10/month billed annually ($120/year). You're paying for a purpose-built, benchmarked model, zero setup, cloud processing on Mac and Windows, plus iPhone and Android support, and ongoing improvements, rather than a one-time on-device app.
No. VoiceInk runs on-device, so it works offline and your audio never leaves your Mac. Aqua Voice processes audio in the cloud to run the Avalon model, so it needs an internet connection. If fully offline, on-device dictation is a requirement, VoiceInk is the better fit.
Aqua Voice runs on Mac, Windows, iPhone, and Android on one subscription. VoiceInk's desktop app is Mac-only with no Windows version, though it now has an iPhone and iPad app. So if you need Windows dictation, Aqua Voice is the option; if you stay on Apple devices, VoiceInk now covers the Mac and iPhone too.
For accuracy, convenience, and reach. Avalon is purpose-trained and benchmarked publicly: on AISpeak-10 it scores 97.4% to Whisper Large v3's 65.1% (Whisper is one of the open models VoiceInk runs). Aqua Voice also requires zero setup (no model downloads, tuning, or compiling), runs Avalon on managed servers, adds screen-context awareness, Windows, and iPhone and Android support, and keeps improving with ongoing updates and support. VoiceInk is an excellent open-source, on-device Mac option if you'd rather self-manage an open model and pay once (or nothing).
Yes, it's what Aqua Voice is built for. Avalon was trained on the speech developers and AI users produce, reaching 97.4% accuracy on technical terms like programming keywords, framework names, and CLI commands. It also reads the active app for context and offers the Avalon API for developers.
Used by people at
Students get 70% off Pro and Max.