Google quietly launched AI dictation for iOS that works without internet

Google quietly released a voice dictation app for iOS that works entirely offline. The new app uses Gemma language models — Google's own line of compact AI models optimized to run directly on the device. This is a direct blow to the voice input market, which has so far been controlled by third-party developers.

Without fanfare or loud announcements, Google placed the new app in the App Store. It's positioned as a next-generation dictation tool: the user speaks — the model recognizes, processes and outputs text, with everything happening locally on the iPhone. No voice fragments are sent to remote servers.

The app is built on Gemma models — a family of open language models that Google has been actively developing since early 2024. Unlike heavy cloud models, Gemma is designed to run on consumer devices with limited resources: smartphones, laptops and tablets. The company has already built Gemma into several Android features, and now has brought a similar approach to iOS — a platform where Google has historically lagged behind Apple.

The main competitor to the new app is Wispr Flow, which quickly became a favorite among the tech community: startups, developers and content creators use it for voice input in any app on Mac and Windows. Wispr offers a convenient overlay, dictation in any input field and integration with work tools. Google's product is more modest in functionality for now, but it has other advantages.

The first is privacy. Offline processing means that voice data never physically leaves the device. For many users this is fundamental: dictation often contains personal information, work documents, confidential conversations.

Cloud services, even with good privacy policies, remain a point of vulnerability. Local processing eliminates the question entirely. The second is speed.

The absence of network requests eliminates the delay between speech and text. In practice this is especially noticeable with unstable connections — in the subway, airplane, weak coverage areas. The app works equally reliably regardless of internet quality.

The third is traffic savings. Dictation to the cloud creates a constant stream of audio data. For users with limited data plans, this is a real problem.

Offline mode eliminates any mobile data consumption for AI processing. The iOS launch looks like a strategic signal. Google is actively promoting Gemma as an open solution for on-device AI and demonstrating its capabilities through its own products.

The dictation app is a clear proof of concept: a small model handles real-time speech recognition without cloud support. The voice input market is becoming increasingly competitive. Apple improves built-in dictation with each iOS version.

Microsoft integrates voice features into Windows through Copilot. Startups — Wispr Flow, Superwhisper and others — build products on top of open models like Whisper from OpenAI. Now Google is joining this race with its own stack and strong brand.

A quiet release without a press release — typical Google tactic for new experimental products. The company releases, watches for audience reaction, iterates. If the app gains organic traction, we should expect a wider announcement, integration with Google Docs, and possibly an Android version.

For users, this is in any case good news: competition drives prices down and pushes quality up.