The short answer
Wispr Flow is the pick for most people willing to pay: it types cleaned-up text into any app on Mac, Windows, iPhone and Android for $15 a month or $144 a year, and its free plan covers 2,000 words a week on a computer. To pay nothing, Spokenly and the open-source Handy dictate offline with no word limit. On a Mac, VoiceInk costs $25 once.
Which AI dictation app should you use?
Pick by the device you dictate on and whether your voice may leave it. Prices are list prices in US dollars as of October 2026.
- Most people who will pay for one: Wispr Flow. It works in any app on Mac, Windows, iPhone and Android, removes "um"s and mid-sentence corrections, and costs $15 a month or $144 a year. The free plan allows 2,000 words a week on a computer.
- Offline dictation with AI cleanup on Mac and Windows: Superwhisper. With Pro, its cleanup can run on a local language model, and one license covers your Macs, Windows PCs, iPhone and Android phone, for $8.49 a month, $84.99 a year or $249.99 once.
- Free with no word limit on Mac, Windows, Linux or iPhone: Spokenly. It runs Whisper and NVIDIA's Parakeet models on your device for free, with no account, and on a computer a small on-device model can also remove filler words at no cost.
- Pay once on a Mac: VoiceInk. $25 for one Mac, open source, and it transcribes on the computer.
- Free and open source on Windows, Mac or Linux: Handy. By default it does one thing: you speak, and the text lands where you are typing, with nothing uploaded. AI cleanup is an optional experimental setting.
- Only short messages: the dictation already on your phone or computer. It is free and, on iPhone and on Windows voice access, runs on the device.
On Linux, Spokenly and Handy are the two picks here that run there; Wispr Flow and Superwhisper don't. This guide is about typing by voice into other apps. To turn a recording you already have into text, see the best AI transcription tools, such as MacWhisper.
Is the built-in dictation enough?
For short messages and notes, yes. Every major system includes dictation that types into any text field for free, and the iPhone's runs without an internet connection. Mostly, it doesn't edit: it types the words you say, with automatic punctuation in supported languages, and you say "new line" or fix mistakes yourself. The exceptions are recent: fluid dictation on Copilot+ PCs removes filler words, and Pixel 8 and later phones can correct grammar and typos when you say "Fix it". The AI apps below add a cleanup pass that drops filler words, keeps only your final version when you correct yourself mid-sentence ("Tuesday, actually Thursday" becomes "Thursday"), formats spoken lists, and matches the tone to the app.
- iPhone and iPad: tap the microphone on the keyboard. Apple says dictation is processed on the iPhone in many languages with no internet connection, and stops after 30 seconds of silence (Apple Support). In iOS 27, an on-device model improves spelling, punctuation and capitalization, in English on iPhone 17 Pro, iPhone 17 Pro Max, iPhone Air or later.
- Mac: press the Microphone key or the Dictation shortcut. You can dictate text of any length without a timeout, and on Apple silicon you can type while it listens. Keyboard settings say whether your dictation is processed on the Mac or sent to Apple (Apple Support).
- Windows: press Windows key + H for voice typing. It uses Microsoft's online speech recognition, so it needs an internet connection. On Copilot+ PCs, "fluid dictation" corrects grammar, punctuation and filler words in English with on-device models (Microsoft Support). Windows 11's voice access also dictates, on the device and without the internet (Microsoft Support).
- Android: tap the microphone at the top of Gboard (Gboard Help). Pixel 6 and later phones get Google's advanced voice typing, which keeps your speech on the phone and detects which language you are speaking. On Pixel 8 and later, in US English, saying "Fix it" corrects typos, grammar and punctuation; that step sends the text to Google (Gboard Help).
Try the built-in option for a week before paying. If you spend that week deleting "um"s and retyping punctuation, an AI app will save you the time.
How much do AI dictation apps cost?
The paid cloud apps cluster at one price: Wispr Flow, Willow, Typeless and Monologue all cost $144 a year on annual billing. The apps that run on your device cost far less, and three of the five picks can be bought once.
| App and plan | Monthly | Yearly | One-time | Three years on the cheapest option |
|---|---|---|---|---|
| Wispr Flow Pro | $15 | $144 | $432 | |
| Superwhisper Pro | $8.49 | $84.99 | $249.99 lifetime | $249.99 |
| Spokenly Pro (cloud models included) | $9.99 | $99.99 | $299.97 | |
| Spokenly Local+ (extras for local use) | $29 | $29 | ||
| VoiceInk, one Mac | $25 ($39 for two Macs) | $25 | ||
| Handy | Free | $0 | ||
| Willow Pro | $15 | $144 | $432 | |
| Typeless Pro | $30 | $144 | $432 | |
| Monologue Pro | $15 | $144 | $432 | |
| Aqua Voice Pro | $10 | $96 | $288 |
Sources: Wispr Flow plans, Superwhisper plans, Spokenly pricing, VoiceInk pricing, Handy, Willow pricing, Typeless pricing, Monologue pricing, Aqua Voice pricing. Prices shown are the makers' own websites; Superwhisper says buying inside the iPhone or Android app costs more because of store fees.
Two things in the table are easy to miss. Typeless charges $30 a month on monthly billing, two and a half times its annual rate, so its "$12 a month" needs a year up front. And Superwhisper's $249.99 lifetime license costs less than three years of its own annual plan ($254.97), while covering every device you own.
What do the free plans actually allow?
The apps that run on your device have no word limit; the cloud apps cap the words or the model. Wispr Flow's free plan gives you its full model, but for 2,000 words a week on a computer and only 550 on an iPhone.
| App | Free allowance, October 2026 | The catch |
|---|---|---|
| Built-in dictation | No limit | Little cleanup beyond punctuation, except fluid dictation on Copilot+ PCs and "Fix it" on Pixel 8 and later; Windows voice typing needs the internet |
| Handy | No limit, on your computer | AI cleanup only through an experimental setting; no phone app |
| Spokenly | No limit with local models or your own API keys, including on-device cleanup on Mac, Windows and Linux | A new install uses cloud models; switch to a local one |
| Superwhisper | No limit with local Whisper models; 3,000 words of Pro to try | No AI cleanup and at most two modes on Free |
| Wispr Flow | 2,000 words a week on Mac and Windows; 550 on iPhone; 1,000 on Android | Slows down at the allowance and stops at 5,000 words a week on a computer |
| Willow | No limit, on its "weaker" Frontier Mini model | The full model needs Pro |
| Aqua Voice | 1,000 words | Then Pro at $10 a month |
| Typeless | A weekly word limit Typeless doesn't publish | Lower accuracy than Pro, by Typeless's own table |
| VoiceInk | 7-day trial | Then $25 once |
Sources: Wispr Flow plans (the iPhone and Android counts share one weekly pool), Superwhisper plans, Spokenly pricing, Willow pricing, Aqua Voice pricing, Typeless pricing, VoiceInk docs.
In practice, Wispr Flow's 2,000 words a week is 400 words a working day: about four 100-word emails. That suits replying to messages, not drafting documents. On an iPhone, 550 words a week is about a paragraph a day.
Which dictation apps keep your voice on your device?
Handy, VoiceInk with a local model, Spokenly with a local model and Superwhisper's free plan transcribe on your computer, so no audio leaves it. Wispr Flow always transcribes in the cloud, and by default it also keeps your dictations on its servers and may train on them.
| App | Where your speech is transcribed | Used to improve models? | Reads text around your cursor or on screen? |
|---|---|---|---|
| Wispr Flow | Always in Wispr's cloud (US servers); transcripts, audio and history are stored there by default | Yes by default on Free and Pro; you can switch it off; always off on Growth and Enterprise | Yes by default (text near the cursor and on screen, not screenshots); on iPhone with no off switch |
| Superwhisper | On the device with local models; cloud models are optional on Pro | No | Only in Pro's context-aware modes |
| Spokenly | Cloud models by default; on the device once you pick a local model | Not stated; Spokenly's servers don't store cloud audio but pass it to third-party transcription services, which follow their own terms | Only in modes where you turn on context |
| VoiceInk | On the Mac with local models, or with a cloud provider you connect | VoiceInk runs no cloud of its own; providers you connect set their own terms | Only when you turn on context for AI cleanup; screen text is read on the Mac, then sent to the AI provider you chose |
| Handy | Always on the computer; the optional AI cleanup sends text to a cloud provider only if you pick one | No | No |
| Apple dictation | On the iPhone in many languages; the Mac says which in Keyboard settings | Only if you opt in to share recordings with Apple | No |
| Windows voice typing | Microsoft's cloud (Azure) | Not stated on Microsoft's help page | No |
Sources: Wispr Flow data controls, Wispr Flow data settings, Wispr Flow context awareness, Wispr Flow enterprise privacy, Superwhisper sensitive data, Spokenly privacy policy and Spokenly docs, VoiceInk docs, Handy, Handy post-processing, Apple Support, Microsoft Support.
"Is Wispr Flow safe" is one of the most searched questions about it. The fair answer: Wispr Flow encrypts audio in transit and at rest, says it is SOC 2 Type II and ISO 27001 certified, and offers a HIPAA business associate agreement on any plan (Wispr Flow privacy). For dictation, it keeps zero-data-retention agreements with every third-party AI provider it uses (data controls). But three settings are on by default on Free and Pro: "Improve the model for everyone", which lets Wispr use your audio, transcripts and edits to train its models; Dictation Cloud Storage, which keeps your transcripts, audio and history on Wispr's servers; and Context Awareness (Wispr Flow data settings). If that matters, open Settings, then Data and Privacy, and turn all three off. For legal, medical or confidential work, Wispr's own help page advises turning Context Awareness off; an app that never uploads audio is the simpler answer.
Wispr Flow: best for most people who will pay
Wispr Flow is the default pick because it is the most complete product here. It works in any app on Mac, Windows, iPhone and Android (only Superwhisper among the picks also covers Android), and its cleanup is on from the first dictation: it removes filler words, repetitions and false starts, and keeps only your final version when you change your mind mid-sentence (Wispr Flow). It also adapts style to the app, so an email gets a greeting and a Slack message stays short (Wispr Flow context awareness). It learns names and jargon, supports 100+ languages, and its iPhone app has 4.8 stars from about 17,700 ratings on the US App Store as of October 2026, far more ratings than any rival here.
Pro costs $15 a month, or $144 a year ($12 a month), as of October 2026 (Wispr Flow plans). Prices differ by region: Wispr sets fixed prices in 24 currencies. The free plan gives 2,000 words a week on Mac and Windows, 550 on iPhone and 1,000 on Android. Pro is unlimited.
The only published accuracy test of Wispr Flow comes from a rival. On Superwhisper's benchmark (September 2026, eight datasets), Wispr Flow's raw transcript, with its cleanup switched off, got 8.7% of words wrong. That ranks it fifth of the six cloud models measured, behind Superwhisper's own S1 Voice (6.6%), and about level with the best local model, Cohere Transcribe (8.4%). It was also the slowest cloud model and caught the fewest custom vocabulary terms (62.5%). Superwhisper runs the benchmark and sells some of the models in it.
Skip Wispr Flow if your words can't leave your device (it has no offline mode, and data is processed in the US), you use Linux, or you dictate a lot on an iPhone without paying: 550 free words a week is about a paragraph a day.
Superwhisper: best for offline AI cleanup on Mac and Windows
Superwhisper is the pick if you want what Wispr Flow does but with the option to keep everything on your machine, and one license for every device. A Pro license covers unlimited Macs and Windows PCs plus your iPhone and Android phone (Superwhisper plans). It costs $8.49 a month, $84.99 a year or $249.99 once as of October 2026, with 40% off for students and a 30-day refund.
The free plan is real: unlimited dictation with local Whisper models, custom vocabulary and up to two modes, with nothing leaving your device. What it lacks is the AI cleanup. Pro adds the AI modes that format text for email or messages, context awareness, realtime text, cloud models, and two more local models: Parakeet, which is fast, and Cohere Transcribe, the most accurate local model on Superwhisper's benchmark. Superwhisper says it does not train on your data and reaches its cloud providers under zero-data-retention terms (Superwhisper sensitive data). On Mac and Windows (but not Windows on ARM), Pro's AI cleanup can also run on a local language model, S1-mini, so a dictation can be fully offline from microphone to text (Superwhisper language models).
The platform details matter. On iPhone, Superwhisper is a keyboard and doesn't support iPad; on Android, it is a floating bubble over other apps, on phones only (iOS, Android). On Windows, the larger local models need a graphics card with Vulkan support or fall back to a much slower processor path; Parakeet runs without it (Windows).
Skip Superwhisper if you want AI cleanup without paying (the free plan transcribes but doesn't edit; Spokenly's free on-device cleanup does), or you want your setup to follow you: modes and vocabulary don't sync to your phone or to Windows yet.
Spokenly: best free dictation app
Spokenly is the pick for unlimited dictation without paying. On Mac, Windows, Linux and iPhone, it runs Whisper and NVIDIA's Parakeet models on your device for free with no word limit and no account (Spokenly pricing). Parakeet is the faster of the two model families (see which local model to choose).
It also lets you bring your own API key from OpenAI, Deepgram, Groq and others, so you pay those providers per use instead of a subscription; Spokenly adds no fee. Local+ costs $29 once and adds extras such as settings sync. Pro costs $9.99 a month or $99.99 a year as of October 2026, for Spokenly's own cloud models and hosted AI text processing with no keys to manage.
Basic cleanup is free too. On Mac, Windows and Linux, Local Model Cleanup runs a small model, Spokenly Cleanup 1, on your computer after each dictation: it removes filler words and repeats, applies corrections you say out loud ("Tuesday, actually make that Thursday"), fixes punctuation and writes numbers as digits. It is off by default; turn it on under Settings, then Local Model Cleanup on a Mac, or Settings, then Text Handling elsewhere (Local Model Cleanup).
One setting to change first: Spokenly's own summary says it uses cloud models by default (Spokenly docs). Download a local model, select it, and turn on Local Only Mode, which blocks every network request except model downloads, if you want to be sure nothing leaves the computer (Local Only Mode).
Skip Spokenly if you use Android (there is no app yet, per Spokenly), or you mainly use Windows and want the most polished version: Spokenly says its Mac and iPhone apps are the most mature.
VoiceInk: best one-time purchase on a Mac
VoiceInk is the pick for Mac users who want private dictation and never want another subscription. A license costs $25 for one Mac, $39 for two or $49 for three, with lifetime updates and a 14-day refund (VoiceInk pricing). There is a 7-day trial. Against Wispr Flow's $144 a year, it pays for itself in about two months.
It transcribes on the Mac, starting with NVIDIA's Parakeet V3, and can also use Whisper, Cohere Transcribe or a cloud provider through your own key (VoiceInk docs). Modes switch the model and writing style by app or website, and an on-device cleanup feature, VoiceInk Refine, needs Apple silicon and 16 GB of memory. The code is public under the GPL-3.0 license, so anyone can check what it sends where (GitHub). Note that a local model alone doesn't make every step offline: if you add AI cleanup through a cloud provider, the text goes to that provider.
VoiceInk's own pages disagree on what it needs. Its pricing page says it works only on Apple silicon Macs with macOS 14.4 or later; its docs list macOS 15 or later and warn only that local models don't work reliably on Intel Macs. Plan on an Apple silicon Mac. Its iPhone app is free during a beta and will need a separate license.
If you already own MacWhisper Pro (€64 once), it includes system-wide dictation too, though MacWhisper is built mainly for transcribing files and meetings.
Skip VoiceInk if you also want dictation on Windows or Android: one Superwhisper license covers those.
Handy: free and open source
Handy is the pick when you want dictation that is free for good and that you can inspect. It runs on Windows, Mac and Linux, transcribes on your computer and pastes the text wherever you are typing; nothing goes to the cloud unless you turn on cloud AI cleanup (GitHub). It is MIT-licensed, has more than 34,000 stars on GitHub as of October 2026, and installs from its website, Homebrew, winget or a .deb package.
It uses NVIDIA's Parakeet V3, which its developer recommends for most people and describes as optimized for ordinary processors, Whisper models, which run faster with a graphics card, or Cohere Transcribe, which Handy recommends for the highest-quality English on a machine with a capable graphics card (Handy models). Hold or tap a shortcut, speak, release.
Out of the box, Handy pastes a plain transcript. An experimental post-processing setting, under Settings, then Advanced, then Experimental Features, sends the transcript to a language model with a prompt you write: OpenAI, Anthropic, Groq and others with your own key, Apple Intelligence on recent Macs, or a local model through LM Studio (Handy post-processing). Only a cloud provider takes text off the computer.
Skip Handy if you want polished cleanup and per-app styles that work out of the box: Handy's AI cleanup is experimental and you set it up yourself. It also has no phone app.
Which local model should you choose: Parakeet or Whisper?
For English on a recent computer, choose Parakeet: it is about as accurate as Whisper's best fast model and many times faster. On Superwhisper's benchmark (Apple M4, September 2026), Parakeet V2 got 10.4% of words wrong and ran 133 times faster than real time; Whisper Large V3 Turbo got 10.9% at 9.4 times. Whisper's small models were much worse (15.6% for Small, 20.3% for Tiny). If accuracy matters more than speed and you have an Apple silicon Mac or a capable graphics card, Cohere Transcribe was the most accurate local model of all (8.4%), in 14 languages; it is free in Handy and VoiceInk and part of Superwhisper Pro.
| Local model | Word error rate | Speed vs real time | Languages |
|---|---|---|---|
| Cohere Transcribe | 8.4% | 24x | 14 |
| Parakeet V2 | 10.4% | 133x | English |
| Parakeet V3 | 10.7% | 118x | 24 in Superwhisper's build |
| Whisper Large V3 Turbo | 10.9% | 9.4x | 100+ |
| Whisper Small | 15.6% | 30x | 100+ |
Sources: Superwhisper benchmarks, Superwhisper voice models.
Choose Whisper for languages Parakeet doesn't cover. The free route matters: Parakeet is free in Spokenly, Handy and VoiceInk, but in Superwhisper it is a Pro feature, and Superwhisper's free plan runs only Whisper. The best cloud models are still more accurate: Superwhisper's S1 Voice scored 6.6% and ElevenLabs Scribe v2 7.5% on the same benchmark, while Wispr Flow's raw transcript (8.7%) was about level with Cohere Transcribe. Treat all these numbers as a guide, since Superwhisper runs the benchmark and sells some of the models in it.
What's the best dictation app for iPhone and Android?
On an iPhone, start with Apple's dictation: it is free, works in every app and runs on the phone in many languages. For AI cleanup, Wispr Flow's keyboard is the best rated, at 4.8 stars from about 17,700 US App Store ratings, but its free plan allows only 550 words a week. Superwhisper and Spokenly also offer dictation keyboards, and Superwhisper's can run offline. Since iOS 26.4, a third-party keyboard that opens its own app no longer returns you to the app you were in, so you swipe back; Superwhisper says this affects every third-party keyboard (Superwhisper docs).
On Android, start with Gboard's voice typing, which is free and, on Pixel 6 and later, adds advanced voice typing. Wispr Flow's free plan gives 1,000 words a week on Android. Superwhisper works through a floating bubble on phones. Spokenly, VoiceInk and Handy have no Android app.
| App | Mac | Windows | Linux | iPhone | Android |
|---|---|---|---|---|---|
| Wispr Flow | Yes | Yes | No | Yes | Yes |
| Superwhisper | Yes | Yes | No | Yes (not iPad) | Yes (phones) |
| Spokenly | Yes | Yes | Yes | Yes | No |
| VoiceInk | Yes (Apple silicon) | No | No | Yes (beta) | No |
| Handy | Yes | Yes | Yes | No | No |
What about Willow, Aqua Voice, Typeless and Monologue?
All four do much the same job as Wispr Flow, mostly in the cloud (Monologue can also transcribe offline), and none beats it on what most readers weigh first, free allowance and platform coverage, as of October 2026.
- Willow runs on Mac, Windows and iPhone. Its free plan has no word cap but uses a model Willow itself calls "weaker"; Pro costs $15 a month or $144 a year.
- Aqua Voice is the cheapest cloud subscription here, $10 a month or $96 a year, on Mac, Windows, iPhone and Android. Its free plan is 1,000 words, too few to judge it for long.
- Typeless runs on Mac, Windows, Linux, iPhone and Android, and says it keeps no dictation data in the cloud and never trains on it. Its free weekly limit isn't published, and monthly Pro costs $30.
- Monologue runs only on Mac, iPhone, iPad and Apple Watch, and comes with a subscription to Every's bundle of apps. It can transcribe offline with local models, and costs $15 a month or $144 a year after a 1,000-word trial.
If you are working from a list published earlier in 2026, check the dates. TechCrunch's widely cited ranking from May 2026 gives Wispr Flow's iPhone free tier as 1,000 words a month and its Android app as "in the works", lists Monologue at $10 a month and Typeless as Windows and Mac only (TechCrunch). As of October 2026, those are 550 words a week, an Android app that exists, $15 a month and five platforms. Asked about dictation apps in October 2026, Perplexity mostly cited dictation makers' own blogs, and recommended a Windows app on the strength of that app's own article.
What about Dragon?
Dragon is still sold, but only for Windows, and it is built for professionals who dictate long documents. Dragon Professional v16 runs on Windows 11 and 10 and adds custom voice commands and macros for boilerplate text (Nuance); Nuance doesn't list a price on that page. Dragon for Mac was discontinued in October 2018 (Nuance), so on a Mac the choice is among the apps above. For writing a book by voice, any app here with no word limit works; the limit-free options are Handy, Spokenly, VoiceInk, Superwhisper's free plan and paid Wispr Flow.
How to choose
Start with the dictation already on your device for a week. If you keep fixing filler words and punctuation, install Wispr Flow's free plan on your computer and see whether 2,000 words a week covers you; if it doesn't and you don't mind the cloud, pay $144 a year. If your voice shouldn't leave the machine, install Spokenly or Handy and pick a Parakeet model; turn on Spokenly's Local Model Cleanup to drop the filler words for free. If you are on a Mac and want per-app writing styles without a subscription, buy VoiceInk for $25. Choose Superwhisper when you want one license that also covers an Android phone.
How this guide was made
Magpie Index considered 11 dictation apps and the dictation built into Mac, Windows, iPhone and Android, using the makers' pricing pages, help centers, privacy pages and docs, Superwhisper's published speech-recognition benchmark, and the answers Perplexity and GPT-5 with web search gave to the same questions, and the US App Store ratings. One-year and three-year costs were calculated from list prices (script and inputs in the test library); no app was installed or run for this guide. Research and writing were AI-assisted.
Built from the makers’ own documentation and pricing pages and from attributed outside sources. Magpie Index did not test these tools for this guide. Prices and plans were checked on the makers’ sites on October 21, 2026; they change often, so check before you buy. No tool paid to appear here, and links to tools carry no affiliate code. How Magpie Index works