MacWhisper: Price, Models and Features

MacWhisper is a Mac app that turns recordings into text on your own computer. You drag in an audio or video file ¹, record a call in Zoom, Teams or Webex ², or dictate into any app ³, and a speech model running on the Mac writes the transcript. It is built by Jordi Bruin and friends at Good Snooze in the Netherlands ⁴.

The short answer on price: the Free version costs €0, free forever ⁵, and MacWhisper Pro is a single payment of €64 with lifetime updates on the maker's website ⁶. On the same day the Gumroad store, where the purchase actually happens, listed it at €65 ⁷. The App Store edition is priced differently (see below). On privacy, the maker's policy says no audio, text or other data leaves your device ⁸; the exceptions are the optional cloud services you connect yourself, covered below.

MacWhisper suits people who transcribe files and meetings more than they dictate: journalists, researchers, podcasters and anyone with interviews to type up. This guide covers what it does, what Pro adds, which model to pick (with our own accuracy measurements of the engines it runs), meetings and speaker labels, and how it compares with Superwhisper, cloud services and the dictation built into macOS. For the wider field, see our overview of dictation software.

MacWhisper transcribes audio and video on the Mac with Whisper

MacWhisper is a file transcription app first, with meeting recording and dictation added around it. Its core job is to take a recording, run it through a speech recognition model on your Mac, and give you an editable, searchable transcript you can export. That design is the main difference from dictation-first apps such as Superwhisper or Wispr Flow, which are built to type as you speak.

It runs open speech models locally, and lets you choose them

The app is named after Whisper, the open speech recognition model family OpenAI trained on 680,000 hours of audio with transcripts ¹⁵. MacWhisper Pro also accepts custom GGML model files, the format used by whisper.cpp, the open-source C++ port of Whisper ¹⁶. It now runs several engines side by side:

Because the models run on the Mac, speed depends on the chip. MacWhisper's free version promises up to about 30 times real time on the GPU ²². Apple used MacWhisper in its own launch material: it said the Mac mini with M4 Pro transcribes up to 20 times faster in MacWhisper than the Intel Core i7 model it replaced ²³. The maker warns that Intel Macs lack dedicated machine learning hardware, which makes on-device transcription significantly slower ²⁴, and that the Medium and Large models want more than 8 GB of memory ²⁵. If you are choosing hardware, an Apple silicon Mac with 16 GB is the comfortable setup.

It handles files, links, podcasts and live audio

Beyond single files, MacWhisper transcribes YouTube links, podcasts recorded with one track per host, whole folders in a batch, and files dropped into a watched folder. Most of those extras are on the Pro list (next section). Transcripts can be searched and edited, and exported as subtitles or documents ²⁶. One limit worth knowing before you buy: it cannot transcribe a file that mixes languages in one pass; the maker suggests running it once per language ²⁷.

Two editions share one engine

MacWhisper is sold in two ways, and they are not identical. The direct download from macwhisper.com (paid through Gumroad) is the full app. The App Store edition is called Whisper Transcription; only the direct version can dictate into any text field and detect meetings ⁹. The App Store edition uses weekly, monthly and yearly subscriptions and cannot take a MacWhisper license key ²⁸. Its US listing shows in-app purchases including a Pro Lifetime option at $99.99 ²⁹, and the Mac version requires macOS 15 ³⁰. On the US App Store the edition averages 3.91 stars ³¹ from 171 ratings ³², against 4.5 on the maker's Gumroad page ¹⁴.

On iPhone and iPad the same app is free to use with all local models, with an optional subscription for cloud transcription and AI summaries ¹⁰. If you are on Windows, our page on MacWhisper for Windows covers the options there.

Free and Pro: what the license unlocks

The free version is a capable Whisper transcriber; Pro is worth buying if you need speed, speaker labels or volume. Pro is a perpetual license: you pay once and keep the app with all current and future updates ⁶. Our question page Is MacWhisper free? answers the price question on its own.

What each version includes

The maker's Gumroad page splits the feature list into a free section and a Pro section. The free version is a true free tier rather than a trial: the list sets no minute cap or time limit on it.

The page does not say exactly which Whisper models are limited in the free version ³³, so check the model list inside the app before you pay if a particular model is the reason to upgrade.

What Pro costs, with discounts and teams

The 50-license pack works out to about 32 percent less per seat than a single license ⁴⁰. A personal license is meant for your own devices, with a soft limit of three ⁴¹. If Pro disappoints, the maker offers a refund within 7 days ⁴². Prices appear in euros on the store; your bank converts them if you pay in another currency.

Is Pro worth it?

Pro pays for itself if any of these apply to you:

• You transcribe a lot. Parakeet is much faster than Whisper (see the next section), and batch and watch folders remove the manual steps.
• You record conversations. Automatic speaker labels, system-audio recording and live meeting transcription are all Pro features.
• You want the app in a workflow. The command-line tool and the integrations with Notion, Zapier, Obsidian and webhooks are Pro.

If you transcribe the odd voice memo or lecture in English, the free version, or the free tools in our guide to free dictation software, may be all you need.

Which Whisper model to choose

For most people on an Apple silicon Mac, Whisper Large v3 Turbo is the best default and Parakeet v3 is the better choice when speed, background noise or accents matter. That is what our own tests of the same engines show, run on a Mac with an M2 Pro with nothing sent over the network. Our question page Which MacWhisper model is best? gives the short version, and our guide to running Whisper locally covers the models in depth.

Turbo is a slimmed-down Large v3

Large v3 Turbo is OpenAI's optimized version of Large v3 with only 4 decoder layers instead of 32 ⁴³. It has 809 million parameters ⁴⁴ and runs about 8 times faster than the large model ⁴⁵. The trade-off: Turbo is not trained for translation, so turning foreign speech into English text needs Medium or Large ⁴⁶. Parakeet TDT 0.6B v3 covers 25 European languages ⁴⁷, and Qwen3-ASR covers 52 languages and dialects ⁴⁸. Whisper's own reach, over 100 languages in MacWhisper ¹³, is the widest.

Our measurements: Whisper is most accurate on clean English, Parakeet holds up better under stress

We measure accuracy as word error rate, the share of words an engine gets wrong. On 100 clean English recordings from the FLEURS test set, Whisper Large v3 Turbo made 3.92 percent word errors ⁴⁹, Parakeet TDT v3 4.48 percent ⁵⁰ and Apple's on-device model 6.73 percent ⁵¹. Across all 25 European languages tested, the median was 11.34 percent for Whisper ⁵² and 10.11 percent for Parakeet ⁵³.

Conversation is harder. On English conversations from 13 accent groups, Parakeet's error rate was 12.57 percent ⁵⁴, against 17.66 percent for Whisper ⁵⁵ and 16.39 percent for Apple's model ⁵⁶. On 3 to 4 minute recordings handed over whole, Whisper made 2.91 percent errors ⁵⁷, Parakeet 2.70 percent ⁵⁸ and Apple 2.34 percent ⁵⁹.

Speed favors Parakeet over Whisper. In our runs Whisper Turbo transcribed English about 12 times faster than real time ⁶⁰, Parakeet about 19 times ⁶¹ and Apple's model about 34 times ⁶² (see real-time factor). A 9to5Mac test inside MacWhisper agrees: on a 7-minute-31-second episode, Whisper Large V3 Turbo needed 40 seconds ⁶³, about 11 times real time ⁶⁴, while Parakeet v2 needed 2 seconds ⁶⁵. In that one episode Whisper was far more accurate, at 1.5 percent word errors ⁶⁶ against 12.3 percent for Parakeet v2 ⁶⁷, so the older Parakeet v2 is a speed pick, not an accuracy pick.

Background noise separates the two engines

When other people talk in the background, Whisper degrades faster. With babble 5 dB quieter than the speaker, roughly a loud cafe, Whisper's errors rose from 3.78 to 13.92 percent ⁶⁸, about 3.7 times its clean rate ⁶⁹. Parakeet went from 4.44 to 7.85 percent ⁷⁰, about 1.8 times ⁷¹.

The full curve: Whisper 3.78 ⁷², 4.20 ⁷³, 5.51 ⁷⁴, 13.92 ⁶⁸ and 66.09 percent ⁷⁵; Parakeet 4.44 ⁷⁶, 4.44 ⁷⁷, 4.48 ⁷⁸, 7.85 ⁷⁰ and 44.47 percent ⁷⁹. When the background is as loud as the speaker, no engine is usable.

Long files: Whisper can drop passages

Whisper is trained on 30-second chunks of audio and cannot take a longer stretch as one input ⁸⁰, and on long recordings it occasionally skips a passage. In our long-recording test, Whisper Turbo left out a passage of 10 or more words in a row in 4 of 84 readings handed over whole ⁸¹, against 1 of 84 for Parakeet ⁸². Cut into pieces, Whisper's error rate fell to 2.35 percent ⁸³. Whisper's authors also list hallucination, text unrelated to the audio, among its known failure modes ⁸⁴. MacWhisper's maker recommends the WhisperKit engine for long files with silence or music ¹⁷; Argmax's own paper on WhisperKit reports 2.2 percent word errors at 0.46 seconds latency in a real-time test ⁸⁵.

Pick by language and situation

In the major European languages the two are close, so let the situation decide:

1. Clean English or a language outside Europe: Whisper Large v3 Turbo, on the WhisperKit engine if your Mac has an M-series chip.
2. Meetings, noisy rooms, strong accents, or hours of audio: Parakeet v3 (Pro).
3. Translating speech into English: Whisper Medium or Large, not Turbo.
4. An Intel Mac or 8 GB of memory: a smaller Whisper model such as Small, or a cloud provider.

Our guide to speech recognition accuracy explains these tests, and speech recognition and accents goes through each accent group.

Meetings, speakers and exports

MacWhisper records and transcribes meetings without a bot joining the call, labels speakers automatically in Pro, and exports to every common subtitle and document format. The details below come from the maker's help center and store page.

Meeting recording works in most call apps, but it is still marked beta

MacWhisper can notice when a meeting starts and offer to record it. Its help center lists 15 apps it watches, including Zoom, Teams, Webex, Slack, WhatsApp, Chrome, Safari and other browsers; FaceTime is excluded because of macOS privacy restrictions ¹⁰⁰. Recording needs macOS Screen Recording and Microphone permission ¹⁰¹, because that is how a Mac app captures other apps' sound. The help center calls automatic meeting recording a beta with possible occasional data loss, and advises manual recording for critical meetings ¹⁰².

The audio is captured on your Mac, so no assistant joins the call as a participant, which is how many cloud note takers work (see our guide to meeting transcription). Recording system audio is a Pro feature ³⁶, as is live transcription during the meeting with speaker labels ³⁷. Whatever app you use, check the recording-consent rules where you and the other people are. Our question Can MacWhisper record meetings? covers setup.

Speaker recognition is automatic in Pro

Speaker diarization means splitting a transcript by who spoke when. MacWhisper's website says it can distinguish and label the different speakers in your recordings ¹⁰³. In Pro it does this with local models on M-series Macs or through ElevenLabs and Deepgram, and with NVIDIA's Nemotron 3 Diarization ³⁵, a model that handles up to eight speakers ¹⁰⁴. The free version lets you assign up to two speakers by hand ³⁴.

Speakers appear as Speaker 1, Speaker 2 and so on; you can rename them, merge two that are really one person, and move a segment to another speaker with a number key. Speaker recognition does not run in dictation, which assumes one speaker, or in podcast mode, where each file is one host ¹⁰⁵. The help center says automatic speaker recognition works with local WhisperKit models or with ElevenLabs and Deepgram ¹⁰⁶; the store page now adds Nemotron 3.

How well it works depends on the audio. When MacWhisper added Parakeet in June 2025, 9to5Mac transcribed a 3-hour podcast with speaker recognition in 1 minute 22 seconds on an M2 Pro MacBook Pro ¹⁰⁷, about 132 times real time ¹⁰⁸. One App Store reviewer in April 2025 found that speaker identification often attached a word at the edge of a sentence to the wrong person ¹⁰⁹. Expect to fix some labels by hand, especially where people talk over each other. More on this in Does MacWhisper recognise different speakers? and our guide to timestamps and speaker labels.

Exports cover subtitles, documents and data

MacWhisper exports SRT and WebVTT subtitles, plus CSV, DOTE, DOCX, PDF, Markdown and HTML ²⁶. MacWhisper also has a command-line tool, mw, that talks to the running app over a local socket and uses the same models and settings ¹¹⁰. That makes MacWhisper scriptable from Terminal or an AI agent without sending audio anywhere.

Dictation is included, but it is not the main act

The direct version can type what you say into any text field after you press a shortcut you choose ¹¹¹. Optional AI clean-up of dictated text needs Pro and your own API key for a provider such as OpenAI or Anthropic. Dictation is missing from the App Store edition ⁹. If dictation is your main use, the dictation-first apps in our comparison of AI dictation apps are built around it.

Privacy: local by default, cloud only if you connect it

MacWhisper's privacy policy says the app does all its work on your device and that no audio, text or other data leaves it ⁸; the one exception it names is anonymous analytics about licensing status ¹¹². That makes it suitable for interviews and other sensitive recordings, which the maker itself stresses ¹¹³.

Two caveats. First, the policy describes local processing, but the app can now switch between local models and models hosted remotely ¹¹⁴: if you add an API key for OpenAI, Groq, ElevenLabs or Deepgram and pick that provider, your file goes to that company under its terms (OpenAI accepts at most 25 MB per file through MacWhisper) ²¹. Second, AI summaries, chat and dictation clean-up send the transcript text to whichever AI provider you connect. Organizations can switch cloud transcription off for everyone through device management ¹¹⁵. Our guide to Whisper AI privacy explains what happens to audio in each kind of Whisper app.

How it compares with Superwhisper and cloud transcription

MacWhisper is the cheapest way to own a full-featured local transcriber on a Mac; Superwhisper is the stronger dictation app; cloud services win on convenience and team sharing. The table puts them side by side with the free dictation built into macOS.

Against Superwhisper: same models, different jobs

The two apps share a lot under the hood. Superwhisper also runs Whisper through whisper.cpp ¹²⁷ and Parakeet through Argmax's WhisperKit ¹²⁸, so accuracy on the same model should be similar. The difference is the job. Superwhisper is designed around a shortcut that types at your cursor, with AI modes that rewrite the text, and it added meeting recording in version 3.0.0, dated October 15, 2026 ¹²⁹. MacWhisper is designed around the transcript: an editor, speaker labels, subtitles, batch work and exports.

Price favors MacWhisper if you plan to keep the app: €64 once ⁶ against $84.99 a year for Superwhisper ¹¹⁷. Superwhisper wins if you also need Windows, or want dictation with AI rewriting. Our Superwhisper guide and the question Superwhisper vs MacWhisper go further. For an open-source local option, see VoiceInk and our guide to open-source dictation software.

Against cloud transcription: you pay with hardware instead of minutes

Cloud services charge by time or seat. Otter's free plan stops at 300 minutes a month ¹²², and OpenAI's gpt-4o-transcribe costs about $0.36 per hour of audio through its API ¹³⁰. MacWhisper has no metered usage: once you own the Mac and the license, an extra hour of audio costs nothing but time. The cloud wins when you need shared team notes, a meeting bot that works without your laptop, or fast results on an old Intel Mac ²⁴. MacWhisper wins on privacy and on long-run cost for heavy users.

Against built-in dictation: complementary, not competing

For short dictation, the dictation built into macOS is free and processes general text on the device ¹²⁵; you may not need a paid app at all. MacWhisper earns its place when you start from a recording rather than live speech. Our guide to the best dictation apps for Mac compares the live-dictation options. Yap, which we make, is an on-device dictation app for Mac and Windows; if typing by voice is your main need rather than transcribing files, you can download it and compare.

Who MacWhisper suits

• Journalists and researchers with interviews: local processing, speaker labels, exports, and a 25 percent discount ³⁹.
• Podcasters and video makers: subtitles in SRT and VTT and one-track-per-host podcast mode.
• Teams and IT departments: volume packs ³⁸ and device management with cloud features switchable off ¹¹⁵.
• Not ideal for: Windows users, people on Intel Macs with long files, or anyone who mainly wants dictation with AI rewriting.

MacWhisper comes up often in developer circles: it appeared in 87 Hacker News stories and comments in the 12 months to September 2026 ¹³¹. For the rest of the market, start from our guide to the best dictation software or the Whisper AI explainer.