Technical explainer · Verified 2026-09-05

WhisperKit
Whisper on Apple platforms

WhisperKit is a Swift speech-to-text framework that runs Whisper models on Apple devices through Core ML and the Apple Neural Engine. It downloads compatible models during first use and offers compressed variants. Package deployment targets and development prerequisites are different, and its model naming distinguishes a Core ML optimization variant from the underlying Whisper turbo checkpoint.

Package targets versus development tools

The package manifest lists iOS 16, macOS 13, watchOS 10 and visionOS 1, with Swift tools version 5.10. The README's development prerequisites separately call for macOS 14.0 or later and Xcode 16.0 or later. These are separate requirements: the package defines its deployment targets, and each desktop application that embeds it sets its own minimum operating system.

Core ML models and compressed choices

WhisperKit downloads its models from the project's Core ML collection on Hugging Face during first use. The README recommends the compressed large-v3-v20240930_626MB variant across iOS and macOS for accuracy, and large-v3-v20240930_turbo on macOS for speed and accuracy. The collection also contains smaller multilingual and English-only models. Pick a compatible model deliberately instead of assuming every Whisper-format file can be loaded directly.

Decode the model name before comparing sizes

In WhisperKit naming, v20240930 identifies the large-v3-turbo checkpoint, while the _turbo suffix denotes a Core ML variant. Those markers are not interchangeable. The collection includes compressed turbo checkpoint variants labeled 626MB, 547MB and, with the Core ML turbo suffix, 632MB. These are size labels in names, not runtime memory measurements. WhisperKit itself is released under the MIT License.

How WhisperJot uses this

WhisperJot's macOS Jot Local Pro is WhisperKit-based. Its default large-v3-turbo download is approximately 1.5 GB; selectable alternatives are compressed large-v3-turbo at approximately 626 MB and small.en at approximately 217 MB. Non-English transcription requires large-v3-turbo because the .en option is English-only. WhisperKit's full catalog offers additional models beyond this product selection. Jot Local Pro runs on-device by default and works offline after the model download.

Explore WhisperJot →

Frequently asked questions

Does WhisperKit use the Apple Neural Engine?

Yes. WhisperKit is built on Core ML to run Whisper models on the Apple Neural Engine. It is a Swift framework for Apple platforms. The actual app embedding it can impose its own operating-system requirements and model choices. Check the application's requirements too.

Does WhisperKit need internet every time?

Its README says models are downloaded during first use. Local inference then uses the downloaded model without needing a network connection. Complete the initial model download before expecting an offline session, and distinguish inference from other network features an embedding app might offer.

Is every WhisperKit turbo variant large-v3-turbo?

No. The _turbo suffix denotes a Core ML variant. The v20240930 marker identifies the Whisper large-v3-turbo checkpoint in the names discussed here. Read both parts of the name before comparing model families, compressed sizes or application defaults. The suffix alone is insufficient.

Sources

Primary documentation checked 2026-09-05. Specifications and runtime requirements can change.

Related explainers

Whisper large-v3-turbo

Understand Whisper large-v3-turbo: fewer decoder layers, reference speed, translation limits, compressed variants and its role in WhisperJot Jot Local Pro.

whisper.cpp

Explore whisper.cpp, its GGML model disk sizes, integer quantization and Apple acceleration options, plus how WhisperJot uses GGML on Windows and Linux.

Run Whisper locally

Plan local Whisper transcription with the official VRAM table, whisper.cpp and Python routes, or WhisperJot's desktop models and honest download sizes.

See also: Whisper glossary definition · System requirements · Pricing