Official documentation

Bring advanced speech-to-text to your app with SpeechAnalyzer

Apple, Apple Developer, WWDC25 session 277, 2025

Read on October 16, 2026

What it is: Apple's WWDC25 session on the SpeechAnalyzer and SpeechTranscriber API, with its transcript. It explains that speech-to-text runs on the device once the language model is downloaded, and how the older SFSpeechRecognizer could use Apple's servers.
How much weight it carries: Primary Apple documentation for developers, so it is the most direct statement of how Apple's speech API works. It describes a developer API, not the Dictation key in System Settings, and the session does not say which engine Mac Dictation itself uses.

Facts taken from this source

Apple says its older SFSpeechRecognizer class worked well for short dictation and could use Apple's servers on devices with limited resources.
“It worked well for short-form dictation and it could use Apple servers on resource-constrained devices” (Transcript, introduction to the speech API history)
Apple says that transcription with its SpeechAnalyzer API is entirely on the device, although the language models must first be downloaded.
“Remember that transcription is entirely on device but the models need to be fetched.” (Transcript, model download passage)