Is Apple Dictation better than Whisper on a Mac?

It depends on which Whisper model you run and what you dictate. On the one same-recording Mac test we found, a large Whisper model made far fewer errors than Apple's speech engine, and it was much slower. A vendor test in 2026 reported the reverse for the smaller Whisper models it ships. Neither result settles the question for your voice, your words or your language, so the sensible order is to start with Apple Dictation, then test a Whisper app on the text you actually write.

Apple Dictation and Whisper are not the same kind of thing. Apple Dictation is the built-in feature. Whisper is an open speech model from OpenAI, and you use it through an app or a command-line tool. An app that runs Whisper can be more accurate than Apple's feature on some audio, but it is a separate download and it has its own settings.

The one same-audio Mac test favored Whisper on accuracy

In a 2025 test by 9to5Mac, Apple's speech engine had a word error rate of 10.3 percent on one 7-minute-31-second episode ¹. Whisper Large V3 Turbo had a word error rate of 1.5 percent on the same episode ². A word error rate is the share of words that are wrong, added or missing, so the gap means roughly one word in ten needed fixing for Apple, and far fewer for Whisper.

Whisper paid for that accuracy in time. On the same episode, Whisper Large V3 Turbo took about 4.4 times as long to transcribe the audio as Apple's speech API ³.

Three limits matter before you rely on this result:

• It was one episode, one run. The author says the models were evaluated the same way, but that the exact numbers would differ under slightly different methods ⁴.
• It tested an API, not the Dictation key. The author reached Apple's speech API through a command-line tool, not through the feature you start with the Dictation key ⁵. We did not verify which engine the Dictation key uses in each macOS version.
• It was a recording, not live speech. Live dictation in a text field has different timing and noise. Test it that way too.

A 2026 vendor test found Apple ahead of the Whisper models it ships

Inscribe, a speech company, reported in 2026 that Apple's SpeechAnalyzer beat every Whisper model it ships, including Whisper Small, on the clean and noisy halves of LibriSpeech ⁶. LibriSpeech is a benchmark of read English speech, so it does not reflect how you talk in a meeting or dictate a long draft.

This result and the 9to5Mac result do not conflict. Whisper Small is a smaller model than the large one 9to5Mac tested, and Inscribe ships Whisper-based products, so it has a reason to compare its own models with Apple's engine. Treat the vendor result as a claim to check, not as proof. Apple's speech framework is also open to other apps: its SpeechAnalyzer class, available from macOS 26, lets third-party apps add speech-to-text with Apple's own models ⁷. A third-party app may therefore use Apple's engine rather than Whisper, so check which engine an app says it uses.

Where each one fits better

Whisper can drop some spoken words. A 2025 study on dementia speech found that Whisper tends to omit filler words such as "uh" and "um" ⁸. If you need a verbatim record of what was said, check the output for missing fillers before you use it. For clean prose, that omission may be a benefit.

Language is the other factor. Apple lists a set of Dictation languages for the Mac, and Whisper is a multilingual model. Check your language against both before you decide, and read our guide to Mac dictation languages.

What to do: test both on the same 300 words

1. Dictate 300 words of your own writing into Apple Dictation, in the app where you normally type.
2. Dictate the same 300 words into a Whisper-based app, using the same microphone and room. Our guide to the best dictation apps for Mac lists the Mac apps in each group.
3. Count the wrong, added and missing words in each result. A word error rate is the count divided by the number of words in your text.
4. Time the wait from the moment you stop speaking to the moment the text appears.
5. Choose the one with fewer errors that is fast enough for how you work. If the difference is small, choose the free Apple tool.

If you want the offline details before you choose, read our question on the best offline speech to text for Mac. For a longer test plan, see the word error rate entry in our glossary. Our question on whether Dragon NaturallySpeaking works on a Mac covers a third option that is no longer sold for Mac. For the whole subject, see our overview of dictation on Mac.