1. Subscription, lifetime, or pay-once: the October 2026 price lists
Superwhisper’s plans page, checked October 3, 2026, gives the Free plan unlimited dictation with any local Whisper model, custom vocabulary, text replacements, and up to two modes without AI processing, and says nothing you dictate on it leaves your device. Pro adds cloud models, the Parakeet and Cohere Transcribe local models, file transcription, unlimited and AI-powered modes, context awareness, realtime transcription, system audio recording, speaker separation, and your own API keys. Pro costs $8.49 a month, $84.99 a year, or $249.99 once for Lifetime, with the same features on each; new users get 3,000 words of Pro free, and there is a 30-day refund.
VoiceInk’s pricing section lists three one-time licenses: Solo at $25 for one Mac, Personal at $39 for up to two, and Extended at $49 for up to three, each with lifetime updates and a 14-day money-back guarantee. Onboarding offers a 7-day trial. Its terms, last updated April 20, 2026, describe a personal, non-transferable license to the compiled app with current and future updates, and say you keep any version released during your license term. A Mac license does not activate the iPhone app, and cloud provider charges are paid to those providers.
- Superwhisper Pro over three years: $305.64 monthly, $254.97 annually, or $249.99 Lifetime.
- VoiceInk: $25, $39, or $49 once, plus any cloud API usage you add.
- Discounts: Superwhisper 40% for students and non-profits; VoiceInk asks students to contact the developer.
2. macOS versions, Windows support, and the phone apps
Superwhisper’s quickstart requires macOS 14 or later, or Windows 10 or 11, and its Windows page says it runs natively on x64 and ARM64 with the same license as the Mac app. On Windows, Cohere Transcribe and Whisper Large v3 Turbo use Vulkan when the GPU supports it and fall back to a much slower CPU path otherwise, and Parakeet Multilanguage is the only local model on Windows ARM. There are also an iPhone keyboard and an Android bubble, and one Pro license covers unlimited Macs and Windows PCs plus an iPhone and an Android phone. Superwhisper says Intel Macs work best with cloud models.
VoiceInk is a Mac app, plus a separate iPhone app with its own license. Its pricing section says it works only on Apple silicon Macs with macOS 14.4 or later, while its installation docs and GitHub README list macOS 15 or later, so check the current download page against your Mac before buying. The docs warn that local models do not work reliably on Intel Macs. VoiceInk Refine, its on-device cleanup model, needs Apple silicon and at least 16 GB of memory. There is no Windows or Android version.
3. Which speech models each app can run, locally or in a cloud
Superwhisper pairs one voice model with an optional language model in each mode. Every Whisper model, from Tiny up to Large v3 and Large v3 Turbo, is free and runs on your device through whisper.cpp; Pro unlocks the local Parakeet (English) and Parakeet Multilanguage (24 languages) models and Cohere Transcribe (14 languages), plus cloud models: Superwhisper’s own S1-Voice, Deepgram’s Nova series, and ElevenLabs Scribe. The app advertises 100+ languages, which in practice depends on the voice model a mode uses.
VoiceInk’s onboarding starts with Parakeet V3, and its local catalog adds Parakeet V2, Ultra, and Unified, Nemotron, Whisper from Tiny to Large v3 Turbo, Cohere Transcribe, SenseVoice, Apple’s own speech engine on recent macOS, and imported whisper.cpp files. Cloud transcription uses your own API keys with providers such as Groq, Gemini, Mistral, Deepgram, ElevenLabs, Soniox, Speechmatics, AssemblyAI, xAI, and Cartesia. Text enhancement can run in the cloud (Anthropic, OpenAI, Gemini, Groq, and others) or locally through Ollama, VoiceInk Refine, or a local command.
4. Keeping dictation on the Mac: each app’s data path
Superwhisper describes every dictation as two stages: a voice model turns audio into text, and an optional language model reshapes it. Pair a local voice model with a local language model, or use the Voice to Text mode with no language model, and Superwhisper says no audio or text leaves your machine. Its cloud models are reached through API agreements with zero-data-retention terms and no training on your data, and Pro users can bring their own keys so data runs under their own provider agreements. Recordings and history stay in a local app folder.
VoiceInk’s privacy page says to check three settings, not one. Local transcription keeps audio on the Mac; cloud transcription sends audio to the provider you picked; enhancement sends the transcript, your vocabulary, and any enabled context to its provider; and Dictionary Auto Learn sends changed text plus a little surrounding text to its own review provider unless you point it at a local Ollama model. Transcripts, enhanced results, and saved audio are stored locally, and provider API keys sit in the macOS Keychain.
- Fully local in Superwhisper: a local voice model plus a local or no language model.
- Fully local in VoiceInk: local transcription, no or local enhancement, and Auto Learn off or on Ollama.
5. Claude Code plugins, shell hooks, and auto-send
Superwhisper installs plugins into coding agents: Claude Code, OpenCode, Codex, Cursor, Pi, Grok, and DeepSeek. Once connected, it pops up when an agent asks a question or finishes a turn, and you can answer, or approve or deny a tool call, by voice. A superwhisper command-line tool searches your local transcription history and can run as an MCP server so an assistant such as Claude Code can read your dictations. All of this is macOS only for now; Superwhisper’s Windows page lists the agent integrations, the CLI, Hold Shift to Auto-Send, and Simulate Keypresses as not yet available there.
VoiceInk takes a scripting approach. A mode’s Custom Command output runs a local shell command with the final text on standard input and in the VOICEINK_TRANSCRIPT variable, so you can append to a journal or hand text to a script. Local CLI enhancement passes its prompt to an installed tool, with templates for Pi, Claude, Codex, and Copilot, and VoiceInk notes those tools may send the prompt to hosted services. Auto Send presses Return, Shift+Return, or Command+Return when you stop recording with Return, which in a terminal means the line runs, so keep it off at a shell prompt.
6. A third option, labelled: IraVoice for one key on three systems
IraVoice publishes this page, so treat this section as ours. It is shaped differently from both apps: no modes and no AI rewriting, just one key. IraVoice 1.2.2 runs on macOS 13+ on Apple silicon, Windows 10 or 11 (64-bit), and Linux (beta); Linux is something neither Superwhisper nor VoiceInk offers. Speech is recognized on the computer with NVIDIA Parakeet, from the default Parakeet Unified EN model, about 700 MB, which downloads once. Hold Right Command or Right Ctrl, speak, and release.
Its focus is developers: in terminals such as Terminal, iTerm2, Ghostty, and Windows Terminal, in editors such as VS Code, Cursor, Xcode, and JetBrains IDEs, and in coding agents, it spells names from your Git project, such as getUserById, and it learns the words you correct. IraVoice Local costs $79 once for one computer, $109 for two, or $149 for three, after a 3-day free trial. That is more than VoiceInk’s Solo license, it has no phone app, file transcription, or meeting features, and it does not run on Intel Macs; pick one of the two apps above if modes and AI cleanup are what you want.