You're 35,000 feet over the Pacific, Wi-Fi is $12 an hour, and you need to transcribe an interview before landing. This used to mean either paying through the nose or waiting until you touched down. Google just made that problem disappear.
The company quietly released an offline dictation app on iOS last week, powered entirely by its Gemma open-weight models. No internet required. No cloud processing. No subscription fee hanging over your head. For anyone who's tried to use voice-to-text on a plane, in a dead zone, or simply without wanting their medical notes, therapist session, or board meeting minutes routed through someone else's servers, this changes things.
The app competes directly with Wispr Flow, the well-regarded dictation tool that's been building a loyal following among journalists, doctors, and productivity obsessives. Wispr Flow charges for its premium tier and routes processing through the cloud. Google's entry undercuts that model on both dimensions—offline by default, and built on an open foundation anyone can inspect and modify. The implications for Whisper, long the default choice for developers building on-device speech recognition, are equally significant. Gemma has been steadily climbing the performance ladder, and this deployment signals Google believes its models can handle real-time transcription at a quality level that users will accept.
The technical picture supports that confidence. Gemma's recent iterations have shown strong results on speech-related benchmarks, and the model's efficiency on edge devices has improved markedly. Running transcription locally means latency stays low and consistent—important for anyone who's watched cloud-based services stutter during poor connectivity. It also means privacy advocates get their win: sensitive recordings never leave the device.
Pricing for the iOS app remains unclear, and Google didn't respond to questions about whether this ships as a standalone product or rolls into an existing service. That's a gap worth watching. If it's free or bundled with Android, it becomes a meaningful differentiator for Google's ecosystem. If it lands at a price point comparable to Wispr Flow, the comparison becomes about raw quality—and Google will need to earn that verdict in practice, not just in benchmarks.
What's not ambiguous is the strategic signal. By picking dictation—a use case where users have concrete, recurring needs and where cloud dependency has been a genuine friction point—Google is stress-testing Gemma in the wild. Dictation is not glamorous. It's not going to generate the viral demos that Gemini image generation does. But it's exactly the kind of task that makes people switch ecosystems. If Gemma nails this, the argument for on-device AI over cloud AI gets a lot harder to dismiss.
The app quietly shipped on iOS last week, available in English with additional languages expected. Android users will have to wait, which means for now, Google's edge AI experiment is running first on Apple's hardware.