Mac dictation, 2026
Twelve ways totype by voice.
Every comparison of Mac dictation apps is written by one of the apps. So is this one. It starts with the four cases where the answer is not us, and the table includes the free option already on your Mac.
The short answer
If you have never dictated, start with Apple Dictation. It is free, already installed, and it answers the only question that matters: do you think faster out loud than you type. If you want text that reads as written rather than spoken, you want a cleanup layer, and that is the whole paid category below, plus one free entry since June: Google's Eloquent does the cleanup on your Mac, offline, in English only. If the audio must never leave your machine, the honest answers are MacWhisper, VoiceInk, Spokenly's free local models or Superwhisper's offline modes. If your work is not only in English, that is the case we built Sonira for, and the case AI Dictation builds for too, so test both.
01 / What each one is
Twelve tools, and what they are actually for
These get lumped together as AI dictation and they are not the same kind of thing. One is built into the operating system. One is a transcription app that also dictates. Four of them transcribe on your own machine at no cost, and one of those four is Google's. The rest are cleanup layers over speech models, some bought and a few, lately, built in-house. The last column is the one that decides it.
| Tool | Runs on | Offline mode | Cost | Best for |
|---|---|---|---|---|
| Apple Dictation | macOS, iOS, built in | Yes, many languages on device | Free | Finding out whether you like dictating at all. It types what you say and rewrites nothing. |
| Google AI Edge Eloquent | macOS, iOS. No Windows or Android | Yes, the whole Mac app runs on device | Free, no subscription, no cap | A free cleanup pass, in English only: filler words and restarted sentences removed by a Gemma model running on your Mac. Google says other languages are not supported yet. |
| Wispr Flow | macOS, Windows, iOS, Android | Not offered | Free tier of 2,000 words a week on desktop, then 15 USD per user a month, 12 billed annually | The most polished mainstream option, and the one most people mean when they say AI dictation. |
| Superwhisper | macOS, Windows, iOS, Android | Yes, best on Apple Silicon | Free tier, a subscription, and a one-time lifetime licence | Offline transcription with a lot of knobs: custom modes, model choice, your own prompts. |
| Aqua Voice | macOS, Windows, iOS | Not offered | Free tier of 1,000 words, then 10 USD a month, 8 billed annually. The realtime tier is 30, or 24 annually | One of the cheaper paid subscriptions here, plus an API if you want to build on it. |
| Willow | macOS, Windows, iPhone | On Pro, an Offline Mode on Mac and iOS | Free tier, then 15 USD per user a month, 12 billed annually | Learning your writing style over time and matching tone from app to app. |
| AI Dictation | macOS, Windows, iOS, Android | Yes, on device when you have no connection | Free trial of 5,000 words, then 8.49 USD a month | The other tool here that positions on languages rather than speed: accents, mixed-language sentences, and per-language testing it publishes. |
| Spokenly | macOS, Windows, Linux, iOS | Yes, local Whisper and Parakeet models | Free forever with local models or your own API keys. Cloud tier at 9.99 USD a month, 99.99 a year | Offline dictation at no cost on four platforms, with the cloud as an optional upgrade rather than the product. |
| VoiceInk | macOS, iOS | Yes, local models by default | One time, from 25 USD. No subscription | Open source and auditable, bought once. The closest thing here to owning your dictation tool. |
| Lispr | macOS 11 and later, Windows 10 and later | Not advertised | Free, no account. Pro at 7.99 USD a month | Speaking one language and landing the text in another on the same keypress. 32 target languages, with Portuguese and Brazilian Portuguese listed separately. |
| MacWhisper | macOS, iPhone and iPad | Yes, local Whisper models | Free version, Pro at 64 EUR once with lifetime updates | Transcribing files, meetings and recordings. It dictates system-wide too, but that is the smaller half of the app. |
| Sonira | macOS, Windows next | Not offered | Free during the open beta. There is no paid plan yet | European and Brazilian Portuguese as two separate standards, sentences that switch language halfway, and reading selected text back aloud. |
Platforms and prices checked on 28 September 2026 and the day after, against each maker's own site, in a browser rather than a fetch, because three of the pricing pages render in JavaScript and answer a script with an empty shell. One price had moved since the readings earlier in the month. AI Dictation's free tier had become a 5,000-word trial that does not renew. The cells and sentences about platforms, offline modes and speech models that had gone out of date are corrected. Superwhisper's figures are described rather than quoted because their pricing block did not render reliably enough to quote exactly. Lispr publishes no offline mode, so that cell says what they advertise rather than what we assume. Prices move; the makers are the source, not this page. Google AI Edge Eloquent was added on 12 October 2026 from Google's own product page and its June announcement, both read that day: free with no cap, English only, Mac and iPhone, the Mac app on device.
02 / Pick by what you need
Five cases, and what we would actually tell you
Written the way a friend who has used all of them and is not paid either way would put it.
- 01
You have never dictated before
Use Apple Dictation for a week first. It costs nothing, it is already on the machine, and it settles the only question that matters. A good share of people discover they do not think well out loud, and those people should not buy any of this. Our guide on this site teaches the free one properly, not as a straw man.
- 02
The audio must not leave your machine
Five honest answers now, not two. Google's Eloquent runs its whole Mac app on device, MacWhisper and VoiceInk run local models by default, Spokenly's free tier is local only and can block every network request, and Superwhisper's offline modes are good on Apple Silicon. Willow's Pro plan adds a sixth, an Offline Mode on Mac and iOS. Everything else here, Sonira included, sends your audio to a server to be transcribed. If you dictate clinical notes, legal drafts or anything under an agreement that forbids third-party processing, that rules most of this category out, and it should.
- 03
You want the most polished mainstream tool
Wispr Flow. Better funded, four platforms, a meeting notetaker, and a team rather than one person. We compete with it and we still say this plainly: if you are on Windows or Android today, or you want the notetaker, it is the answer and we are not.
- 04
You want to spend as little as possible
Nothing at all, if you are willing to run local models. Spokenly's local tier is free with no account and no word limit, MacWhisper has a free version, and Google's Eloquent is free with no cap, cleanup included, if English is all you dictate. If you want a cloud tool for nothing, Lispr is free and asks for no account. Among paid subscriptions, Lispr's Pro is the cheapest hosted one here, at 7.99 USD a month or 79 a year. Aqua Voice is 8 USD a month billed annually and AI Dictation 8.49, and VoiceInk's one-time 25 USD costs less than any of them inside four months and then costs nothing.
- 05
Your work is not only in English
This is the case we built for. Almost everyone here advertises 100+ languages, which is true and is not the same as being good in yours. European and Brazilian Portuguese are one entry on those lists and two different languages in practice. If your sentences also switch language halfway, that is the specific failure Sonira exists to handle. We are also no longer the only tool saying this. AI Dictation positions on languages rather than speed, publishes its own per-language testing, and runs on four platforms to our one. Test it against us, and test the two things we are actually narrow about: whether the two Portuguese standards survive as two, and what happens to a sentence that switches language halfway.
Four of those five point somewhere other than us. That is not modesty. A dictation tool is a habit you keep for years or drop in a week, and being sold the wrong one is how people conclude the whole category does not work.
03 / What the comparisons miss
Four things the tables never tell you
Transcribing a file and dictating into an app are different products
MacWhisper and the Whisper apps are built around audio you already have: a recording, a meeting, a podcast. Dictation is the opposite shape, a key you hold while you talk and text that appears where your cursor already is. Several apps do both, and each is good at the half it was designed for. Buy the one whose main job is your job.
A language count is a model spec, not a product claim
Many tools in this table buy the same handful of managed speech models, and then the language list belongs to the model, not the app. Products differ after the transcript: whether the cleanup pass leaves your words alone or quietly rewrites them into another register, and whether anyone tested it in your language instead of in English.
No two speed numbers here are comparable
Every latency claim in this category is measured by the maker, on their hardware, on their network, on sentences they chose. We publish none for that reason. If speed decides it for you, the only measurement worth anything is you holding the key, in your own app, on your own connection.
The free tier stopped being a trial
The free option used to mean a word cap designed to run out. Several tools here now give away unlimited local transcription permanently and sell only the cloud on top of it, and since June Google gives away the cleanup pass too, in English. That moves the question. It is no longer whether dictation is worth paying for, it is whether a hosted model and a cleanup pass are worth paying for, because the raw speech to text now runs free on your own hardware.
None of that is a reason to buy ours. It is a reason not to buy from a table alone, this one included.
04 / Where we fit
Built for the sentences that switch language halfway
Sonira is a hold-to-talk dictation app for the Mac, made by one person. Hold one key, speak, and clean text lands in whatever you were already typing in. The difference is narrow on purpose: English and Portuguese both done properly, with European and Brazilian Portuguese treated as two standards rather than one entry on a list, and mixed-language sentences left the way you said them.
It also reads selected text back aloud, in a natural voice and with a player, in the same tool that takes your dictation. Few tools in this table do that, and Wispr Flow does not. Voice in one direction is the category. Voice in both directions is what we are actually building.
What we do not claim: a better speech model. We buy managed speech models, as many tools here do. A few rivals now run their own: Wispr Flow has Canto, Aqua Voice has Avalon, Superwhisper has S1-Voice. The difference is what happens to the transcript afterwards, and no speed claim appears anywhere on this site.
Availability: the current build, with dictation and read-aloud, is the download on this site.
One key. Both languages. Both directions.
05 / How to judge any of them
Ten minutes that will tell you more than this page
Every tool here has a free tier or a trial. Run the same five things through each candidate, in your own apps, and the differences stop being marketing.
- 01
Dictate a real message, not a test sentence
Open the app you write in most and dictate something you were going to send anyway. Test sentences get pronounced more carefully than real ones, which is exactly what hides the failures you would live with.
- 02
Say a name only your work uses
A client, a product, an internal acronym. Nothing in this table has a custom vocabulary that solves this well, so what you are measuring is how wrong it goes and how fast you can repair it.
- 03
Switch language in the middle of a sentence
If you work in more than one language, say a sentence that starts in one and finishes in the other. Most tools either translate the whole thing or turn the second half into nonsense. It is the fastest way to separate them.
- 04
Dictate late in the day, when you are tired
Speech gets less crisp as the day goes on, and that is exactly when you most want to stop typing. A tool that only works while you are performing is not a tool you will keep.
- 05
Count the edits, not the errors
The number that matters is how often you reached for the keyboard afterwards. A tool with more raw transcription errors but a cleanup pass that removes your filler can still leave you editing less.
Worth running against Sonira too. If we lose it for your work, we would rather you found out in ten minutes than in three weeks.
06 / FAQ
The questions people actually ask
- 01What is the best dictation app for Mac?
- There is no single winner, which is why the verdicts above are five cases instead of one name. Never dictated before: Apple Dictation, free and already installed. A polished mainstream paid tool: Wispr Flow. Fully offline: MacWhisper, VoiceInk, or Spokenly's free local models. Free with no account at all: Spokenly locally, or Lispr in the cloud. Cleanup for nothing, in English: Google's Eloquent. Bought once rather than rented: VoiceInk. Built around languages rather than speed: AI Dictation, or Sonira if your case is the two Portuguese standards and sentences that switch language halfway.
- 02Is there a free dictation app for Mac?
- Yes, and more of them than these comparisons usually admit. Apple Dictation is built into macOS with no word limit. Spokenly's local models are free forever, unlimited, with no account. MacWhisper has a free version. Google AI Edge Eloquent is free with no cap and no subscription, on device, and English only. Lispr is free in the cloud and asks for no account. On top of that, most paid tools have a capped free tier: Wispr Flow gives 2,000 words a week on desktop, Aqua Voice 1,000 words, and Superwhisper and Willow both have free tiers. AI Dictation used to belong on that list and no longer does: what it offers now is a 5,000-word trial that never renews. Sonira is free in full during its open beta because it has no paid plan yet.
- 03Which Mac dictation apps work offline?
- Google's Eloquent runs its whole Mac app on device. MacWhisper and VoiceInk run local models on your Mac by default. Spokenly's free tier is local only and can block every network request. Superwhisper offers offline models and says they run best on Apple Silicon. Apple Dictation processes many languages on device. Wispr Flow, Aqua Voice, Lispr and Sonira all send audio to a server. Willow lists an Offline Mode on Mac and iOS as a Pro feature, and AI Dictation says it falls back to on-device when you have no connection.
- 04Is a paid dictation app worth it over Apple Dictation?
- For one reason: the cleanup pass. Apple Dictation types what you say, filler words and restarted sentences included, and you edit afterwards. The paid tools run the transcript through a language model that strips the filler and punctuates it, so what lands is closer to something you would send. If you heavily rewrite everything you dictate anyway, the free one is enough.
- 05Which one handles Portuguese best?
- We would say ours and we are not a neutral source, so here is the testable version. Many tools here buy the same managed speech models, and a few now run their own. Either way, the difference that matters appears after transcription: whether the tool converts European Portuguese into Brazilian forms or formalises Brazilian speech, and whether it copes with a sentence that has English in the middle. Dictate one of each into any two of these and the difference shows up in under a minute. One other tool here is aimed at the same thing rather than at speed: AI Dictation. If Portuguese is the reason you are choosing, test both of us on the same paragraph.
- 06Do any of these work on Windows?
- Wispr Flow, Superwhisper, Aqua Voice, Willow and Lispr all ship Windows builds today. MacWhisper has no Windows build, and Google's Eloquent is Mac and iPhone only, with Google saying it is evaluating other platforms. Sonira is Mac only right now, with Windows as the next platform, so if you need Windows today we are not your answer yet. AI Dictation adds Android and iPad, and Spokenly is the only one here that also ships Linux.
Try the free one first.
And if you want the one built for two languages at once, it is free while in beta.
The head-to-head with Wispr Flow
Leaving Wispr Flow specifically? Six reasons, six answers