The best superwhisper model for you: its own picks for cloud, local and free

superwhisper's docs name four models to start with. S1-Voice is its cloud model and the one it calls most accurate. Cohere Transcribe is its most accurate local model. Parakeet Multilanguage is the fast local choice for many languages. Whisper Large v3 Turbo is the strongest free one, because every Whisper model in the app is free and the other three need Pro.

I make YapToText, a different dictation app for the Mac, so read what I say about superwhisper with that in mind. Every fact about superwhisper here is from its own docs, read on October 10, 2026. The pages are Choose a model, Voice models and How We Benchmark. The accuracy figures below are superwhisper's own benchmark results, not my tests, and I have not measured any of these models.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

superwhisper's short list

From superwhisper's docs, read October 10, 2026. Word error rates are from its own benchmark suite.

ModelRunsSizeLanguagesPlanWhy its docs pick it
S1-VoiceCloudNo download100+ProThe lowest word error rate on its benchmarks, 6.6% across eight datasets
Cohere TranscribeOn device1.3 GB14ProThe most accurate local model, 8.4% on an Apple M4. No realtime, no translation
Parakeet MultilanguageOn device494 MB24ProFast on Apple silicon, and the only local voice model for Windows ARM
Whisper Large v3 TurboOn device1.6 GB100+FreeA balance of speed and quality, and free

If you want it free

Every Whisper model in superwhisper is free, and they all run on your device. The docs list twelve, from Whisper Tiny at 75 MB to Whisper Large v3 at 3 GB. Whisper Large v3 Turbo is the one the docs point free users to. The Large, Medium, Small, Base and Tiny multilingual models can also translate speech into English, which Turbo cannot.

The superwhisper pricing page covers what else the free plan includes.

If you want it local

  • Most accurate. Cohere Transcribe. It covers 14 languages: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese and Arabic.
  • Fastest. Parakeet for English, Parakeet Multilanguage for 24 languages. The docs say Parakeet models can struggle with punctuation and occasionally hallucinate on single-word recordings.
  • Widest language list. The multilingual Whisper models, at 100+ languages.

The docs note that local speed depends heavily on hardware, and that their figures come from M4 and M4 Max Macs. The superwhisper offline page covers running with no connection.

If you want realtime text or speaker labels

  • Realtime. Works with the Nova models, Scribe and the Parakeet models. Cohere Transcribe and the Whisper models do not support it.
  • Speaker separation. Every voice model supports it, and the docs say the Nova cloud models give the most reliable results.
  • Long recordings. The docs recommend Nova 3.
  • Medical terms. Nova Medical, a cloud model, English only.

Old names: Ultra, Nano, Fast, Standard

From the docs page Voice models, which says these models were renamed in the app.

Old nameName now
Superwhisper UltraWhisper Large
Ultra V3 TurboWhisper Large v3 Turbo
Standard, Nano and FastThe small Whisper models

How to change the model

  1. Open superwhisper's settings and click Modes.
  2. Open the mode you dictate with.
  3. Pick a model in the Voice Model dropdown.
  4. To download or remove local models, open Settings, Models library.

The model is set per mode, so two modes can use two different models. The superwhisper modes page covers the rest of a mode's settings.

What YapToText uses

YapToText has no model to choose on day one. Whisper Large v3 Turbo ships inside the App Store download and runs on the Mac, and that is the same model superwhisper's docs point its free users to. YapToText offers Whisper models only, with no Parakeet, Cohere or cloud option, and supports 29 languages plus auto-detect. Its cleanup model is included too, where superwhisper's AI modes need Pro.

The Whisper Large v3 Turbo page covers that model, Parakeet vs Whisper compares the two families, and the local models guide goes wider.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

The short version

  • Cloud. S1-Voice.
  • Local, accuracy first. Cohere Transcribe.
  • Local, speed first. Parakeet or Parakeet Multilanguage.
  • Free. Whisper Large v3 Turbo.
  • Speakers and long files. Nova 3.

If superwhisper changes its models and this page falls behind, open an issue on GitHub. Every one gets read.

Download YapToText on the Mac App StoreThis app is completely free. You don't have to make an account. There are no subscriptions, no difficult downloads, and nothing is locked. Made for Apple silicon Macs. Source on GitHub.

About the author. Ryleigh Newman is a developer whose hands make typing difficult and who dictates every day with YapToText, the app this site describes. Facts about other apps come from their own sites, with the date they were read. More about the project.

Sources

Questions

What is the best superwhisper model?

Its docs pick S1-Voice in the cloud and Cohere Transcribe on device for accuracy, Parakeet Multilanguage for speed, and Whisper Large v3 Turbo on the free plan.

What is the best free superwhisper model?

Whisper Large v3 Turbo. Every Whisper model in superwhisper is free and runs on your device.

What happened to superwhisper Ultra?

It was renamed. The docs say Whisper Large was formerly Superwhisper Ultra, and Whisper Large v3 Turbo was Ultra V3 Turbo.

Is Parakeet free in superwhisper?

No. The docs say Cohere Transcribe, the Parakeet models and all cloud models need Pro.

Which superwhisper model supports realtime transcription?

The Nova models, Scribe and the Parakeet models. Cohere Transcribe and the Whisper models do not.