Superwhisper review: local models, limits and Pro pricing

Superwhisper's main reason to be on a developer's shortlist is control over the stages between speech and finished text. You can choose a voice model for transcription and, when needed, a separate language model for rewriting. That flexibility is useful only if the configuration fits your machine and the corrections you need afterward.

This review evaluates official documentation and pricing checked directly on September 11, 2026. It includes an evaluation procedure, not hands-on latency, accuracy or battery measurements. For a direct buying comparison, read Wispr Flow versus Superwhisper.

Two model choices determine the workflow

The voice model catalog separates local transcription from cloud transcription. A local voice model processes the recording on your device; a cloud voice model sends audio for remote processing.

The language model documentation covers the next stage: refining or transforming the transcript. A local voice model paired with cloud cleanup still sends text outside the device. If you only need transcription, Superwhisper's security guide describes Voice Mode without a language model.

For coding prompts, start with whether the raw transcript preserves identifiers, numbers and negations. Then evaluate cleanup separately. A polished request is not an improvement if it changes “do not retry” into an instruction to retry.

Offline use depends on the platform and mode

Superwhisper's sensitive-data guide says local language models are currently supported on macOS only. It advises Windows users who need local processing to use a local voice model with no AI post-processing, rather than assuming Windows supports the same local cleanup options.

That distinction is more useful than a general “works offline” label. For an offline trial, download the chosen local model first, confirm the mode does not use a cloud language model, and test without network access in a local text editor. Record the exact app and model version. Success in that trial demonstrates the task worked offline under those conditions; it is not a network audit of every app feature.

The vendor lists Mac, Windows, iPhone and iPad for its Pro license. A shared license does not establish identical features on all platforms. This review does not infer iOS local-model support from a desktop model list. Our local versus cloud guide explains the remaining checks around storage, sync and the destination app.

Choose model size by your own task

The voice catalog lists local models with different downloads, including Whisper Large V3 Turbo at about 1.6 GB and Ultra at about 3 GB. Those are published model sizes, not total memory requirements or a prediction of speed on your laptop.

The same catalog includes speed and accuracy ratings supplied by Superwhisper. They are not independently measured scores in this review. Use them to identify candidates, then compare the actual output and correction time on your hardware.

A small local model that handles your short messages may be sufficient. A larger model that reduces term errors can be worth its download and processing cost. Neither outcome can be chosen from the catalog alone, especially for accented English or mixed Vietnamese and English.

Pro pricing and the free-tier documentation gap

The Plans & Pricing page lists USD reference prices of $8.49 monthly, $84.99 annually and $249.99 lifetime. It says the paid plans have the same features and that website and App Store purchase prices can differ. Check your actual checkout currency, tax and offer before paying.

Twelve monthly payments total $101.88, which is $16.89 more than one $84.99 annual payment. Three annual payments total $254.97, only $4.98 above the current lifetime price. These are billing illustrations at unchanged prices, not a forecast of product availability or a guarantee about future license terms. If you stop using the app after a few months, the smaller upfront commitment matters more than a theoretical long-term saving.

The free-tier sources do not agree. Plans & Pricing places local voice models in Pro, while the voice catalog labels several local models Free. The product homepage also advertises unlimited Whisper-model use in Free. The current homepage and pricing guide describe a 3,000-word Pro trial.

Because of that inconsistency, this review does not promise permanent free access to a particular local model. Check the model's availability on your account after the trial, and ask support to clarify if offline access is the reason for your purchase.

A short evaluation before committing

Use the existing 20-sentence kit as input material; its instructions about Flow settings do not apply to Superwhisper. Select the appropriate language and model controls in Superwhisper itself. Include tasks you actually do, such as a short reply, a request with an exact amount and a coding prompt with an identifier.

1. Record OS, hardware, microphone, app version, voice model and any language model.
2. Start with transcription only and save the raw output before editing.
3. Repeat comparable sentences with cleanup enabled, keeping the voice model unchanged.
4. Record meaning changes, required-term errors and time to make the text usable.
5. If offline operation matters, run it as a separate condition with all required downloads completed.

Keep failed trials and note any change in room noise or microphone. Do not mix a different model, language selection and microphone into one comparison and attribute the result to only one of them.

Who should shortlist Superwhisper?

Shortlist it if you want explicit local/cloud model choices, transcription without rewriting, or a paid license structure beyond monthly subscriptions. Allow time to set up and compare modes. These are documented capabilities, not evidence that it will be faster or more accurate for your voice.

If you only need occasional voice input and your existing tool already handles it, compare the extra setup with the benefit you actually observe. The alternatives guide organizes other options by task. Before buying lifetime, verify that the mode, platform and language you need work on your own device after the trial conditions are understood.