Google releases offline dictation app for iOS powered by Gemma

Google has released a voice dictation app for iOS that works fully offline, sending no data to servers. It is an unusual move for a company whose business model is built on cloud data processing, and the first public product to use Gemma directly on the user's device in real time.

The app appeared quietly in the App Store with no press release and no broad announcement. It is positioned as a tool for voice text input in any app: you speak, and the text appears right where the cursor is. The key difference from competitors is full autonomy: speech recognition happens on the device and no internet connection is required.

Under the hood run Gemma models, Google's open language models, which the company began releasing in 2024. Gemma was designed from the start to run on mobile devices and edge hardware: the models are more compact than the main Gemini models, but powerful enough for tasks that don't need global context. Voice dictation is a typical use case for such models: the task is computationally limited, but benefits from the minimal latency that local execution provides.

The direct competitor Google is going after is Wispr Flow, a popular AI dictation app that has won an audience among professionals who prefer to work by voice. Wispr Flow focuses on transcription quality and integration with work tools, but has historically depended on cloud processing. Google's offline alternative calls into question one of its rival's key architectural decisions.

Offline mode has concrete practical advantages: working on a plane, in the subway, or in areas with poor coverage all becomes possible without losing functionality. For corporate users it is also a privacy question: data does not leave the device, which is critical when handling confidential materials. This argument may prove decisive for the enterprise audience, which has always wondered where employees' voice recordings end up.

It is telling that the app launched without a marketing campaign. It is either a market test ahead of a broad launch, or part of a strategy to promote Gemma as a developer platform, since the app shows clearly what can be built with on-device models. Either way, the quiet App Store release proved louder than many official announcements: media picked up the news precisely because Google doesn't usually do things this way.

Offline AI on mobile devices is ceasing to be an exotic feature and becoming an industry standard. Apple Intelligence, Samsung Gauss, and now Gemma in dictation: the big players are steadily moving computation onto the device. Competition will shift from cloud infrastructure to the quality of the models themselves and the depth of integration into the operating system. Google has made exactly that move.