TL;DR: SuperWhisper is a good Mac app with a limited Windows port. If you’re on Windows and want the same thing done natively, that’s Invoke: local Whisper on your NVIDIA GPU, push-to-talk, auto-paste, $49 once. The modes concept maps over too, with reformatting that reads your actual project instead of a generic prompt.
SuperWhisper is a good Mac app
I want to be upfront about that. SuperWhisper runs Whisper locally, has solid transcription quality, and the lifetime pricing option means you’re not locked into a subscription. For years it was the obvious pick on a Mac, and it’s still a good app.
But I’m on Windows. And so are a lot of developers.
SuperWhisper recently shipped a Windows version, but it’s behind the Mac app on features. If you’re on Windows and want the full experience, SuperWhisper isn’t there yet.
Where SuperWhisper and Invoke overlap
Both run Whisper locally. Both do push-to-talk. Both keep your audio on your machine. The core idea is the same: speak, get text, move on.
| Invoke | SuperWhisper | |
|---|---|---|
| Price | $49 one-time | $8.49/mo or ~$250 lifetime |
| Platform | Windows / macOS (Apple Silicon) | Mac-first, Windows (limited) |
| GPU acceleration | NVIDIA CUDA on Windows, MLX on Apple Silicon | Apple Silicon (Metal) |
| Push-to-talk | Yes | Yes |
| AI reformatter | Yes, with project context | Yes, custom modes + external LLMs |
| Auto-paste | Yes, both platforms | Yes |
| Offline | Yes | Yes |
Where they differ
SuperWhisper added custom modes and external LLM support, so it does have AI reformatting now. You can bring your own API key for GPT-4, Claude, etc. and write custom prompts for how your transcription gets processed. Credit where it’s due. That’s a solid feature.
Where Invoke’s reformatter goes further is project context. It doesn’t just clean up your speech. It reads your CLAUDE.md, package.json, git history, and understands your stack. So when you say “add a post endpoint for user off,” Invoke knows you meant “authentication” because it can see your codebase. SuperWhisper’s modes are general-purpose text processing. Invoke’s reformatter is developer-specific.
The other big difference is platform, and it has moved. SuperWhisper is Mac-first, and the Windows version they shipped is still behind on features. I built Invoke on Windows because that’s what I use: CUDA acceleration on NVIDIA cards, auto-paste via pynput, the whole stack native. If you have an NVIDIA GPU, it just uses it, and if you don’t, it falls back to CPU.
Invoke also ships on macOS now, Apple Silicon only, M1 or newer. The .dmg is signed and notarized by Apple, transcription runs on the M-series GPU through MLX, and on first launch macOS asks for Accessibility, Input Monitoring, and Microphone permissions because the global hotkey and auto-paste need them. That matters for this comparison specifically, because SuperWhisper is a Mac-first tool. The two apps now meet on Mac hardware as well, not just on Windows where SuperWhisper is playing away from home.
Invoke is $49 once. Free 7-day trial, all features, no credit card.
The pricing math
SuperWhisper’s monthly plan is $8.49. That’s $102/year. Their lifetime option is around $250.
Invoke is $49 once with a year of updates, or $79 for lifetime updates. Even the lifetime tier costs less than SuperWhisper’s lifetime option.
I’m biased here, obviously. But the math is the math. If pricing is your main criterion, I broke down every no-subscription dictation option separately.
Switching from SuperWhisper: what maps to what
If you’re coming off SuperWhisper, whether you moved to (or back to) a Windows machine or you’re just switching apps on the same Mac, here’s the translation table:
- Push-to-talk hotkey → same concept. Bind Mouse5 or a key combo, hold to record, release to transcribe.
- Custom modes → prompt modes. Invoke ships modes for Claude Code, Cursor, Windsurf, Copilot, and Codex, plus commit-message and PR-description modes. The difference is the reformatter also reads your
CLAUDE.md,package.json, and git history, so the output references your real codebase instead of applying a generic text-processing prompt. - Model picker → same idea, Whisper model sizes trade speed for accuracy. On an NVIDIA card the big models are still sub-second.
- Local-only guarantee → same. Audio never leaves the machine. The optional reformatter sends text (never audio) to whichever LLM you configure, or point it at Ollama and keep the whole pipeline local.
Settings live in a plain config, history is on disk, and the 7-day trial is full-featured, so you can verify the whole workflow transfers before paying.
When SuperWhisper is the better choice
If you want a Mac-native UI and you don’t need project-aware reformatting, SuperWhisper is fine. I have no complaints about their transcription quality, and their custom modes cover a lot of ground.
That recommendation used to be automatic on a Mac, because Invoke wasn’t there to argue with. Now it’s a real comparison on both platforms, and it comes down to two things: whether a reformatter that reads your codebase changes how you work, and whether you’d rather pay $49 once than $8.49 a month. On Windows there’s a third thing, which is that Invoke is a native app and SuperWhisper’s port is not the whole product.
And if you’re comparing the broader field rather than these two specifically, I keep a full roundup of every way to run Whisper on Windows, including the free file-transcription tools. I also keep a maintained map of the landscape, because a lot of the tools named in older comparison articles have been abandoned since they were written.
Try it
7-day free trial, all features, no credit card. Download here.