Google Releases Offline-First AI Dictation App Using Gemma Models

Google has quietly launched an offline-first dictation application for iOS that uses the company's Gemma AI models to transcribe speech without requiring an internet connection, entering competition with apps like Wispr Flow. The app processes audio entirely on-device using quantized versions of Google's open Gemma models, allowing users to dictate text in environments without connectivity or in privacy-sensitive situations. The system supports continuous transcription with punctuation and formatting. The release marks Google's continued push into on-device AI capabilities following its earlier announcements about running large language models on mobile hardware. Competitors in the space include Wispr Flow, which has raised significant venture funding for similar voice-to-text technology. The app represents a departure from Google's traditional cloud-based approach to speech recognition, which has historically required sending audio to remote servers for processing. Google has not publicly announced the app, which appeared on the iOS App Store without fanfare.