Willow is the page in this library that had to be rewritten after reading their site. It was going to be the ordinary shape — a monthly service on one side, a local purchase on the other, and the question of where your voice goes settling it. Then their home page turned out to advertise "Dictate without an internet connection, powered entirely by your device", which removes the easiest argument this app has before the comparison has started.
So the page has to be about what is left, and what is left is genuinely interesting: a subscription service with an offline capability against a $19 purchase with no service behind it at all. They are not the same shape even when the feature lists overlap, and the difference shows up in your bank statement, in whether there is an account with your name on it, and in what happens when something fails mid-sentence.
Willow is the more featured product. Take that as read from here on.
What they charge, checked today
From their pricing page on 4 September 2026:
- Basic — free, no credit card needed. "Unlimited weaker speech-to-text model (Frontier Mini)", limited personalisation, and 20 Scribe uses weekly.
- Pro — $15 a month, or $12 a month billed annually for a stated 20% saving. "Unlimited smartest speech-to-text model (Frontier Pro)", faster transcription, priority processing, "Smart memory of your writing style", unlimited Scribe and extended dictation length.
- Business — $35 a month, or $28 annually. Everything in Pro plus "Enforced privacy mode (zero data retention)", SOC 2 Type II compliance, HIPAA enforcement, and admin and team controls.
- Enterprise — custom, adding invoice and PO billing, SSO/SAML, enterprise security, advanced dashboards and an API, and dedicated support.
Coii VoiceInput is $19 once. Three devices. Updates included, nothing renews, and there is no tier above it. macOS 13 Ventura or later, Apple Silicon or Intel. The trial is thirty days of the whole app with no card and no account.
The free Basic tier matters here and is easy to skip past: somebody who dictates lightly may never need to pay Willow anything, and a free tier that does not expire is a stronger offer than a thirty-day trial for that person. Say so plainly rather than comparing only the paid columns.
Where Willow is ahead
Six things, and none of them is close.
Three platforms. Their site says "Available on Mac, Windows, and iPhone, so your voice can turn into writing across every device you use." This app is macOS and nothing else. If you write on a phone, or your work machine is a PC, the comparison ends there.
Languages. "Dictate in over 100 languages and switch between them without missing a beat." Coii VoiceInput supports several and publishes no comparable list. Where a vendor states a number and we do not, the row is theirs.
An auto-learning dictionary. Willow advertises that it "Remembers names, product terms, and abbreviations". This app has no vocabulary layer at all — the surname it mangles in week one it will still mangle in month six unless you fix it yourself each time. For anyone whose writing is full of proper nouns this is the single largest practical difference on the page.
Formatting. "Smart formatting" that "Removes filler words, fixes punctuation", and the promise of "perfect text — no editing required". This app writes down what you said, filler included.
Certifications. "SOC 2 Type II certified and HIPAA compliant, with zero data retention and end-to-end encryption", with enforced privacy mode and HIPAA enforcement on the Business tier. If your employer's procurement process asks for a SOC 2 report, this app cannot produce one and never will.
Team management. Admin controls, SSO/SAML, dashboards, an API. There is no team story here whatsoever — three devices means three installs and that is the whole of the administration.
Offline mode and no network path are different things
This is the section the whole page exists for, and it needs to be precise rather than clever, because the temptation is to imply that their offline mode is somehow less real. There is no evidence for that and this page is not going to suggest it. They say it works without an internet connection, powered by the device. Take them at their word.
The difference is scope, and it is architectural rather than a matter of degree.
Willow's offline capability is a mode inside a service. There are accounts, tiers, models with names, weekly usage allowances on the free plan, priority processing on the paid one, a retention policy that is enforced on one tier and described on another, and certifications that exist precisely because data normally moves. All of that machinery is correct and well-run and it is also present, which means there is a configuration in which things go over the network and a configuration in which they do not, and you have to know which one you are in.
Coii VoiceInput has none of that machinery to be in a mode of. There is no account. There is no tier. Model weights are fetched once from a public model host on first run; a licence activation goes out when you enter your key; there is a clock check at launch. That is the complete list, and none of it happens during a dictation. There is no code path that sends audio or a transcript anywhere, so there is nothing to switch off and nothing to remember to have switched off.
Which of those you want is a real question with a real answer on both sides. A managed service with a certification and an enforced retention policy is exactly what a compliance department wants to see, and "we have no servers" is a much harder thing for them to file. For an individual buying software for themselves, the calculus inverts: the strongest guarantee is the one that does not depend on anybody keeping a promise.
The table
| Coii VoiceInput | Willow | |
|---|---|---|
| Price | $19 once, 3 devices | Free / $12–15 mo / $28–35 mo |
| Free tier | No — 30-day trial | Yes, no card |
| Platforms | macOS 13+ | Mac, Windows, iPhone |
| Works without a connection | Yes, always | Yes, advertised |
| Account required | No | Yes |
| Server-side processing exists | No | Yes, by default |
| Languages | Several | Over 100 |
| Learns your names and jargon | No | Yes |
| Filler-word removal and reformatting | No | Yes |
| SOC 2 Type II / HIPAA | No | Yes |
| Zero data retention | No data to retain | Yes, enforced on Business |
| Team admin, SSO, API | No | Yes |
| Words saved before the engine runs | Yes | Not stated |
| Target app fixed at key-down | Yes | Not stated |
| Clipboard restored after a paste | Yes | Not stated |
| Usage limits | None | Scribe capped on Basic |
Nine rows to Willow. That is what a funded product with three platforms and a compliance programme looks like against a single-purpose app, and any page that came out closer than this would be lying about one of them.
What is left on this side
Three things, and they are the same three on every page in this library because they are the only three that are actually ours.
The sentence is written down before anything else happens
- 1Key downthe target is chosen here
- 2You speaklevel meter, over your work
- 3Key up
- 4Written to historybefore the engine is asked
- 5Engineon your Mac
- 6Typedabout half a second
Release the key and the transcript is committed to your local history first. Then the rest runs. A slow engine, an application that quit, a connection that dropped in a design where a connection is involved, a paste an application refuses — all of it happens after the words exist. The worst case is opening a list and copying a row.
Every dictation product has users with a story about a sentence that vanished. This is the design in which that story cannot happen, and it is the least copyable claim on the page because it is not a feature that can be bolted on — it is which step comes fourth.
The destination is decided at key-down
Press the key while the cursor is in a message box, talk for twelve seconds, and it does not matter what steals focus while you are talking. The text goes where you started. This is the most common silent failure in dictation and it is the one people misdiagnose as an accuracy problem, because the words were fine — they just landed in the wrong window.
$19, once, against $144 a year
Willow Pro billed annually is $12 a month, which is $144 for a year and $432 over three, for one seat. Coii VoiceInput is $19 for three Macs, forever, with updates included.
That comparison is not a criticism of their price. A service with a hundred languages, a compliance programme and three client platforms costs real money to run, and $12 a month is a reasonable number for it. The point is that you should know which of the two things you are buying, because after three years the gap is twenty-two times and you will have paid it in instalments small enough not to notice.
The account, and what it costs
Willow requires one; this app does not. That is a row in the table and it is also a set of consequences worth writing out, because "no account" sounds like a convenience feature and is not.
No account means there is no record anywhere that you use this software, no password to rotate, nothing to be included in somebody else's breach notification, and no email address that starts receiving product news. It means the app does not stop working because a card expired while you were on holiday. It means there is no seat to reassign when you leave a job, and no admin who can see how much you dictated.
It also means genuinely no sync. Your settings do not follow you to a second Mac; you configure each of the three machines by hand. Willow's account gives you continuity across Mac, Windows and iPhone, and that is a real benefit that this app deliberately does not offer. The trade is stated rather than hidden.
- 08:523.8sConfirming I can make the ten o'clock — I will dial in from the car.Messages
- 09:3412.4sThe second paragraph is doing two jobs; split it and move the caveat into a footnote.Ulysses
- 10:115.1sInvoice went out Friday, payment terms are thirty days from receipt.Mail
- 10:472.4sAgreed, ship it.Slack
"Perfect text — no editing required", and whether you want it
Willow's homepage promises that you "speak into any app, and Willow turns it into perfect text — no editing required", with smart formatting that removes filler words and fixes punctuation. That is a real capability and for a great deal of writing it is straightforwardly the better behaviour. Nobody wants their hesitations transcribed into a message to a client.
It is also a decision made on your behalf, in the half second between speaking and reading, about which of your words were the ones you meant. Most of the time that decision will be right. The cases where it is not are specific and they are the cases where the words matter most: a quotation you are recording verbatim, a clause you phrased carefully because the phrasing is the substance, a commit message where the hedge was deliberate, a line of dialogue with a deliberate stumble in it. In those, a tidying step is a small unauthorised edit you did not see happen.
Coii VoiceInput takes the other position and it is not a superior one, just a different one: the output is exactly what you said, punctuated, and any tidying is a thing you do afterwards with your hands and your eyes. That is more work per paragraph and less surprise per year.
If you know which of those two you want, you already know which app to buy, and you can skip the rest of the comparison entirely. Most people who write for a living know within about a day.
The shape of a day at sixty presses
Every argument on this page is a volume argument, so here is what the volume actually looks like when dictation stops being an event.
It is not the long-form session people picture. It is a two-line reply in Mail at 08:52, a sentence into a Slack thread at 09:07, a commit message, a comment on somebody's draft, a note to yourself before you forget it, an address into a form, a search query, a paragraph of the thing you are actually meant to be writing. Perhaps sixty of them, and most under five seconds each.
At that rate two properties dominate and neither is on a feature list. The first is time-to-first-word: whether the tool is available in the moment you thought of the sentence, or whether there is a window, a mode or a connection between you and it. The second is what happens on the failures, because at sixty a day you will meet all of them within a month — the focus that moved, the application that handled the paste oddly, the clipboard you had loaded with something you needed.
A tool that is frictionless at press one and unreliable at press forty does not get used sixty times a day. It gets used six times, on the important documents, and at six times a day it has not changed anything about how much you type. That is the entire case for caring about the unglamorous rows.
What the three-year picture looks like
Worth laying out because subscriptions are designed to be evaluated monthly and software is used for years.
One seat of Willow Pro, billed annually at $12 a month, is $144 in year one, $288 by the end of year two and $432 by the end of year three. Billed monthly at $15 it is $180, $360 and $540. Business at $28 annually is $336 a year. None of those are unreasonable prices for what is being delivered; they are simply recurring, and recurring costs are the ones people consistently underestimate when comparing against a one-off.
Coii VoiceInput is $19, in year one, and $19 at the end of year three, across three Macs, with updates included over that period. If you are the sort of buyer who still resents a subscription for something that sits behind one keypress, that number is the entire pitch and no amount of feature comparison will move you off it.
The honest counterweight: a subscription funds a hundred languages, a compliance programme and clients on three platforms, and $19 once does not. You get what each model can pay for. The question is only which of the two you need.
What neither of them does
Neither records or summarises a meeting — that is a different product and a different search. Neither transcribes an audio file you already have. Neither controls your computer by voice; if that is what you need, Talon is the tool and it is free to download. And neither will rescue a laptop microphone in an open-plan office, which remains the cheapest accuracy improvement available to anyone reading this.
Switching, in either direction
Cheap both ways, which is worth knowing before you agonise. Neither app holds a library or a document format; the output was always plain text and it went straight into your own files at the time.
Coming here from Willow costs you the dictionary. Months of learned names and product terms do not transfer, there is nothing on this side to receive them, and the first fortnight will produce spellings you have to fix. That is the real cost of the move and it is larger than the install.
Going the other way costs nothing but the subscription. If you do cancel one, cancel it from the account page rather than by deleting the app, and check the renewal date before you do — that is generic advice and it applies to every service in this category.
Which one to buy
Willow if you dictate on more than a Mac, you write in several languages, your work is dense with names the software has to learn, your employer wants a SOC 2 report, or you want the output tidied rather than transcribed. Their free tier also means you can answer most of this without paying anything.
This one if you are on a Mac, you want one payment and no account, you want the words on your own disk before anything downstream can lose them, and you want the text to be exactly what you said. Thirty days of the whole app, no card, no account, and a fortnight of ordinary work will settle it.
The neighbouring decisions: Wispr Flow is the other subscription in this bracket and the arithmetic page, Aqua Voice is the one that edits as you speak, superwhisper is the larger local Mac app, and Apple Dictation is already installed and costs nothing.