The two Mac apps that both run on your machine

Most comparisons in this category are settled by one row: does the audio leave the Mac. This one is not. superwhisper is a local-first Mac app with the same basic instinct as this one, and against it "nothing is uploaded" is not an argument, it is a shared starting point. So the page has to be about behaviour instead — what each app does with the words after you stop speaking, what it keeps, what it costs, and what it refuses to do.

The short version: superwhisper is the larger product by a wide margin, and if you want modes, a custom vocabulary, file transcription or a Windows machine in the mix, it is the one to buy. Coii VoiceInput is $19 once, does exactly one thing, and orders that one thing so that a crash cannot take your sentence with it.

What each one is trying to be

superwhisper is a voice platform. Checked on 4 September 2026, its comparison pricing page lists Pro at $8.49 a month, $84.99 a year, or $249.99 once, with a free tier that does not expire and a 30-day refund on the paid plans. Their Pro documentation sets out what the money buys: local voice models, custom modes, AI-powered modes, custom vocabulary, speaker separation and priority support, with one licence covering all platforms and no device limit on Mac and Windows computers.

Coii VoiceInput is a key you hold. $19 once, three devices, macOS 13 Ventura or later on Apple Silicon and Intel, a 30-day trial with every feature and no card. There is no free tier, no monthly option and no mode system. The whole surface is a hotkey, a pill with a level meter, and a history.

Those are not two attempts at the same product. They are a platform and a utility, and the reason this page exists is that a lot of people looking for one end up reading about the other.

The table

Nine of those rows go to superwhisper. That is the honest shape of it and no amount of writing around it changes the count.

Where superwhisper wins

Three platforms on one licence. macOS, Windows 10 and 11, and iOS, with their own documentation stating there is no device limit for Mac and Windows computers. Coii VoiceInput is macOS and nothing else, and three devices. If you switch between a Mac at home and a PC at work, this comparison is already over.

Modes. The idea that dictating a Slack message and dictating an email should not produce the same text is a good one, and superwhisper builds a whole feature around it — predefined modes, custom modes, and AI-powered ones on Pro. Coii VoiceInput has no equivalent. It writes down what you said, in the register you said it in, and if you wanted that tidied you tidy it yourself.

A custom vocabulary. Listed as a Pro feature. Anybody who dictates the same five internal product names all day will feel this one within an hour. This app has no way to teach it a word, and the same unusual surname will come out the same wrong way in March as it did in January.

A hundred languages and dialects, and translation to English. Their home page states both. This app's language support is several, not a hundred, and that is a difference of kind rather than degree if you write in more than one.

It handles recordings. File transcription, meeting recording, speaker separation. Coii VoiceInput handles speech as it happens and nothing else — if you hand it an hour of audio there is nowhere to put it. Recording and summarising a meeting is a different job that we sell as a different app, and buying two of ours to cover one of theirs is a real cost you should count.

The free tier does not expire. You can use superwhisper indefinitely without paying. This app gives you thirty days and then asks for $19. For somebody who dictates twice a week, free is a completely reasonable answer and we do not have one.

Where Coii VoiceInput wins

The words are saved before anything can go wrong with them

Let go of the key and the transcript is written to your history before the engine is asked a single question. That ordering is the product. It is not a setting, it is not a recovery feature, and it is not something you turn on after the first time it saves you.

Every person who dictates daily has a story about a sentence that went somewhere and never arrived. The app was busy, or the window had closed, or the paste landed in a field that had lost focus. It is a small thing that feels large, because the sentence you had is gone and the one you write instead is never the same. Coii VoiceInput is built so the failure mode is the words are in your history and did not land, which is thirty seconds of annoyance, rather than the words are gone, which is not.

superwhisper does not state an ordering guarantee, so this page cannot tell you what it does when the engine stalls. What it can tell you is that ours is stated, and that you can test it in an afternoon: dictate a long sentence, quit the app before the text appears, and go look at the history.

One tier, and it runs on your machine in all of them

This is a narrower point than it first sounds, and it is worth stating precisely rather than dramatically. superwhisper's own Pro documentation lists local voice models among the things a Pro licence unlocks, and describes the free tier as including limited cloud models. Their home page separately says superwhisper works offline.

Read together, that means the local behaviour is available and is part of what the money buys. Coii VoiceInput has no version that behaves differently: the model weights are fetched once from a public model host, and after that the audio and the transcript stay on the Mac and nothing is uploaded. There is no tier where that changes, because there is only one tier.

If you are choosing on this axis, the practical question to ask superwhisper's own documentation is which model you will actually be running day to day, and whether it is one of the local ones. That is a question with an answer on their site; it is not one this page should answer for them.

Where the text lands is decided when the key goes down

The target is fixed the moment you press the hotkey, not when you release it. That sounds like a detail until you notice how often you click somewhere else while you are still talking — checking a number, reading the thing you are replying to, moving a window. Apps that resolve the target on release send your sentence to wherever your mouse happened to leave you.

The other half of it is the clipboard. Where the accessibility API cannot place text, this app falls back to a synthetic paste and then puts your clipboard back. A dictation tool that pastes and leaves its own text sitting in your clipboard has quietly thrown away whatever you had copied, which you discover two minutes later at the worst possible moment.

$19 against $249.99, or $84.99 a year

Both are fair prices for what they are. A product with a hundred languages, cloud models, modes and meeting transcription has running costs, and a subscription is the right shape for that. But the arithmetic is the arithmetic: their one-time licence is a little over thirteen times this app's price, and their annual plan is more than four times it every year.

What you are buying with the difference is scope. If you will use the modes and the vocabulary and the file transcription, $249.99 once is not expensive for them. If what you actually want is a key that types what you say, you are paying for a platform to get a hotkey.

What the history is for

The history is the other half of the durability argument. It is not an archive feature and nobody buys an app for it. It matters on the days something goes wrong: text that landed in the wrong window is a row you copy, and a sentence you said an hour ago and now want again is a search rather than an act of memory.

It is also, deliberately, boring. There is no cloud copy, no sync between your three machines, and no shared vocabulary built from it. Those would all be useful and they would all require the thing this app is built not to do.

What neither of them does

Neither app controls your Mac. They type what you said. Neither will delete a paragraph, switch applications or click a button because you asked out loud — that is Talon's territory and Talon does it properly.

Neither is a proofreader. Dictation gets punctuation approximately right and proper nouns approximately wrong, in both products, and anything going to a customer still wants reading before it is sent.

And neither of them is going to be good on day one. Which brings us to the part of this decision that no table settles.

The week that actually decides it

The first two days are worse than typing, in either app. The key is not in your fingers yet, you compose sentences in written English and they come out stilted, and you correct more than you save. Most people who abandon dictation abandon it here, and they abandon it in whichever app they happened to install first — which is a shame, because it was not the app.

By day four the key is automatic and the sentences get longer, because you stop writing them in your head before you say them. Somewhere around then most people find the real use case, and it is usually not the one they expected. It is rarely long-form writing. It is the fifty short messages a day that were never quite worth the keyboard: the Slack reply, the commit message, the note to yourself, the first two paragraphs of an email you had been avoiding.

By the second week you are noticing failures rather than successes, which is the right time to judge either product. How often does a name come out wrong? Which of your apps does the text not reach? What happens on the one application you cannot avoid using? Those answers are the thing you are buying, and they are not on either vendor's website.

Both give you room to run that experiment. superwhisper's free tier does not expire and its paid plans carry a 30-day refund; this app's trial is thirty days with every feature, no card and no account. Run them a fortnight apart rather than side by side — two dictation apps listening for hotkeys at the same time is a bad test and, if you leave them on the same key, an actively broken one.

The apps that fight back

Both products are doing the same awkward thing underneath: putting text into an application that only expects it from a keyboard. How well that goes is a property of the application, not of the dictation tool, and it is the single biggest source of "it does not work" reports for everything in this category.

Normal Mac text fields — Mail, Notes, Safari, Pages — are the easy case, and everything here handles them. The hard cases are the ones you actually live in. Electron applications draw their own text fields and expose them to the system unevenly. Terminals have their own idea of what a paste even is. Some password fields refuse programmatic input on purpose, which is correct of them and not something any dictation app should be trying to work around.

Coii VoiceInput's answer is the accessibility API where it works, a synthetic paste where it does not, and your clipboard restored afterwards. That covers most of the hard cases and not all of them, and when it does not, the text is still in the history — which is the same argument as before turning up somewhere else.

Whatever you trial, spend it dictating into the three applications you use most rather than into a text editor. That is the test that predicts whether you are still using it in a month.

What switching actually costs

Almost nothing technical, and that is worth saying plainly because migration anxiety keeps people on tools they have stopped liking.

Dictation does not accumulate a library. The output went into other people's applications as you spoke — into Slack, into Mail, into a document that is now just a document. There is no archive to move, no format to convert and no project file that only one of these apps can open. Whatever history each app keeps is a convenience, not an asset you are locked to.

What does cost you something is muscle memory. A different key in a different place, a pill that appears somewhere else, a different half-second before the words arrive. People who try a new dictation app and go back almost always go back inside three days, before the new reflex has formed, and they usually describe the reason as accuracy when what they actually mean is that it felt wrong. Give it a working week before you decide anything.

The one piece of housekeeping worth doing properly: if you are leaving a subscription, export whatever the old app holds for you while it is still active, and cancel from the vendor's own account page rather than assuming the app deleting itself ends the billing. That is generic advice and it applies in every direction, including away from us.

Which one to buy

Buy superwhisper if you use Windows or an iPhone as well as a Mac; if you have more than three machines; if you write in several languages; if modes and a custom vocabulary sound like things you would genuinely configure; if you need to transcribe files or meetings; or if a free tier that never expires is the right price for how often you would actually use it.

Buy Coii VoiceInput if you are on a Mac and only a Mac; if the thing you want is one hotkey rather than a platform; if you have lost a sentence to a timeout and it annoyed you more than it should have; if you want the target window fixed before you start talking and your clipboard handed back afterwards; and if $19 once is the shape of purchase you prefer.

If you are still weighing this against the subscription apps, the Wispr Flow comparison is the $180-a-year version of the same argument, and Aqua Voice is where the offline question comes back. If you want a Mac app that transcribes recordings too, MacWhisper is closer to that than either of these. If you have not yet ruled out the free option, Apple Dictation is already on your machine, and BetterDictation is the other one-time-purchase local Mac app.

What this app sends is the complete list of connections it can make. Download it for thirty days if you would rather just find out.