Wispr Flow does more. This one loses fewer sentences.

If you dictate on more than a Mac, buy Wispr Flow. It runs on Mac, Windows, iOS and Android, handles 100+ languages, learns your names and jargon as you go, and comes with a meeting notetaker that identifies speakers and answers questions about past calls. It publishes SOC 2 Type II and ISO 27001, and offers a signed BAA for dictation. At $15 per user per month it is priced like the service it is, and there is a free tier to start on.

Buy Coii VoiceInput if two narrower things decide it: that nothing you say leaves the Mac, and that the words are saved before anything downstream can go wrong with them. $19 once, macOS only, no account. That is the argument, and this page is written so you can tell which of those descriptions is you.

The table

Rows we lose are marked as losses.

Where Wispr Flow wins

Four platforms. Dictation is a habit, and a habit that stops working on your phone is a habit with a hole in it. Coii VoiceInput builds for macOS 13 and later and nothing else. For a lot of people that single row settles this.

It learns you. Wispr Flow advertises that it picks up your names and jargon as you go. Anyone who has dictated a colleague's surname four times and watched four different spellings appear knows what that is worth. This app has no personal vocabulary and no way to teach it one — you fix those words by hand, every time.

A hundred languages. If you write in Polish on Tuesdays, that is not a close call.

It also takes your meeting notes. Their Notetaker records meetings on the Mac, identifies speakers and answers questions about past calls. Coii VoiceInput does not do meetings at all — that is Coii AudioNotes, a separate $19 app. Two products where they have one is worse for you if you want both, and the honest thing to say is that buying two of ours to match one of theirs is $38 and two apps in the menu bar.

Certifications, on paper. SOC 2 Type II, ISO 27001, and HIPAA-ready dictation with a signed BAA. This app claims none of those, because it has none of them. If your employer's procurement process needs a certificate rather than an argument, that is a real and disqualifying difference — and one no amount of "but nothing is uploaded" gets around, because the certificate is the artefact they are asking for.

Where Coii VoiceInput wins

The words are saved before anything can go wrong with them.

This is the least copyable thing on this page, and it is an ordering decision rather than a feature. Let go of the key and the row is written to your history before the engine is asked a single question.

Everyone who dictates regularly has lost a sentence: the model took too long, the app quit, the paste went into a window that had closed. It is a small disaster in the moment, because the sentence you had in your head is gone and the one you rewrite is never quite it. This app is built so that the failure mode is the text is in your history and did not land, which is a nuisance, rather than the text is gone, which is not.

Nothing you say leaves the Mac.

No audio, no transcripts. The model weights are fetched once from a public model host, which sees your IP address and which file you asked for, and after that the files are used from your disk and the host is never contacted again. Then it works with the network off — on a plane, on a locked-down laptop, in a building that does not let you talk to the internet.

Wispr Flow is a cloud service, which is what makes the hundred languages and the learned vocabulary possible. Its own feature list offers opting out of model training, which tells you both that the default is in and that there is something in their hands to train on. That is not a criticism of them; it is the shape of the two products, and it is not a setting either one can flip to become the other.

Where the text lands is decided before you speak.

The target is chosen the moment you press the key, not when you release it. If you start talking and then click into another window — which is a normal thing to do while thinking — the text still goes where you were, not where you ended up.

The clipboard detail matters more than it sounds. A dictation app that pastes by hijacking your clipboard and leaves it there has quietly thrown away whatever you had copied. This one puts it back.

$19 once against $180 a year.

Both prices are honest for what they are. A service with a hundred languages and a learned vocabulary has running costs and a subscription is the correct shape for it. It is a different trade, not a worse one: one year of Wispr Flow Pro is roughly nine times this app, and this app has no second year.

What the history is for

Every dictation is a row you can go back to, on your own disk.

This is the other half of the durability argument. Something that landed in the wrong window is not lost, it is a row you copy. Something you said an hour ago and want again is a search rather than a memory.

The truth about speed

Marketing numbers in this category are usually measured from the end of the audio to the first character, on a fast machine, on a short phrase. By that measure everything is fast and the comparison is noise.

What actually decides whether dictation sticks is different: whether you can speak at a normal pace without waiting, and whether the app is ready the instant you press the key rather than half a second later. Around half a second after you let go is what this one aims for, and on Apple Silicon that is what it does.

The honest caveat is the same as for anything running locally: a longer sentence takes longer, and an older Intel Mac takes longer still. A cloud service does not slow down on your hardware — it slows down on your connection instead, which is worse in exactly the places you would most want to dictate.

The apps that fight back

Every dictation tool on macOS is doing the same awkward thing underneath: putting text into an application that was not written to receive it from anywhere but a keyboard. How well that goes depends on the application, and it is the single biggest source of "it does not work" for any of these products.

The well-behaved case is a normal text field in a normal Mac app — Mail, Notes, Safari, Pages. The accessibility API is designed for exactly this and the text appears where the cursor is.

The awkward cases are the ones people actually spend their day in. Electron applications draw their own text fields and expose them to the system unevenly. Terminals have their own rules about what a paste even means. Some password fields deliberately refuse programmatic input, which is correct behaviour and not something any of these apps should be trying to defeat.

Coii VoiceInput's answer is to use the accessibility API where it works and a synthetic paste where it does not, and to put your clipboard back afterwards. That covers most of the awkward cases and not all of them, and when it fails the text is still in your history — which is the same argument as before, arriving in a different place.

Whichever you try, spend part of the trial dictating into the three apps you use most rather than into a text editor. That is the test that predicts whether you will still be using it in a month, and it is not a test either vendor's marketing page can run for you.

The accessibility case

A meaningful share of the people who buy dictation software are not buying a productivity trick. They are buying a way to keep working with an injury, a condition or a disability that makes typing slow, painful or impossible.

Two things are worth saying plainly to that reader, and neither is a sales pitch.

The first is that no dictation app is a treatment. This one types what you say. That is the whole of the claim, and anybody promising more than that about a repetitive strain injury is selling you something they cannot deliver.

The second is that reliability matters differently when dictation is not optional. If typing is a fallback you can drop to, an app that loses a sentence occasionally is an annoyance. If it is not, the same failure is the difference between finishing the work and not. That is the case where writing the words down before the engine runs stops being an implementation detail — and it is also the case where working with the network off matters most, because a connection dropping should not stop you from being able to write at all.

Wispr Flow serves this reader on four platforms and in a hundred languages, which for many people will be the more useful set of properties. The comparison here is genuinely close and it should be decided on a trial, not on a page.

What a week of it looks like

The tables above make this sound like a purchase decision. It is really a habit decision, and the first week tells you most of it.

Days one and two are worse than typing. The hotkey is not in your fingers yet, you speak in written English rather than spoken English and it comes out stilted, and you correct more than you save. Almost everybody who abandons dictation abandons it here.

By day four the key is automatic and the sentences get longer, because you stop composing them in advance. Around then most people find the actual use case — usually not long-form writing, which is where they expected it, but the fifty short messages a day that were never worth the keyboard: Slack replies, commit messages, a note to yourself, the first draft of an email you were putting off.

By the second week you notice the failure modes rather than the successes, which is the right time to judge either product. How often does a name come out wrong? Where does the text not land? What happens when you dictate into the one app you cannot avoid? Those answers are what you are actually buying, and they are worth more than any row in the table above.

What neither of them does

Neither app controls your Mac. They type what you said; they do not delete a paragraph, switch applications or click a button on command. That is Talon's territory and Talon does it properly.

Neither is a transcription service for existing recordings — both are for speech as it happens. And neither replaces proofreading: dictation gets punctuation approximately right and technical terms approximately wrong, and anything going to a customer still wants a read-through.

What switching costs

Nothing technical. There is no library to migrate, because dictation does not accumulate one — the output went into other people's apps as you spoke.

What you should do before cancelling anything is export whatever Wispr Flow holds for you while the subscription is live, and give the new hotkey a full week. The hardest part of switching dictation tools is muscle memory: a different key, a different pill, a different half-second. People who go back usually go back in the first three days, before the new reflex has formed.

Run both for a fortnight if you can. They do not conflict as long as they are not on the same hotkey, and a week of real work tells you more than any table on this page.

Two products, or one

Worth being explicit, because it is a real cost of choosing us. Wispr Flow sells dictation and meeting notes as one subscription. We sell them as two apps: this one, and Coii AudioNotes at $19 for recording and summarising meetings.

If you want both, that is $38 once against $180 a year — still a large gap, but it is two purchases, two menu bar items and two trials rather than one signup. Whether that is a reasonable trade depends on whether you actually want both. Plenty of people dictate constantly and never record a meeting, or the reverse.

What we will not do is blur them to make the comparison look better. Dictation and meeting notes are different jobs with different failure modes, and the two apps stay separate because they are separate — not as a pricing tactic.

Which one to buy

Buy Wispr Flow if you dictate on a phone or a PC as well as a Mac; if you write in more than a couple of languages; if names and internal jargon are most of what you dictate and having them learned is worth $15 a month; if you want meeting notes and dictation from one vendor; or if your employer needs a compliance certificate rather than an architecture.

Buy Coii VoiceInput if you are on a Mac and only a Mac; if what you dictate should not be uploaded; if you need it to work with the network off; if losing a sentence to a timeout is a thing that has happened to you and annoyed you more than it should have; and if $19 once suits you better than $180 a year for features you would not use.

There is a third answer: macOS has dictation built in, it is free, and for someone dictating a text message twice a week it is enough. These apps start earning their price when dictation becomes the way you write rather than something you occasionally try.

If you want to keep reading, the library has the other comparisons and the migration guides. The home page is the short version of what this app does, what it sends lists every connection it can make, and the FAQ covers the trial and the languages. Or just download it — thirty days, every feature, no card and no account, and the price only comes up after that.