In short: Google has quietly released an iOS app called Google AI Edge Eloquent, a free, offline-first voice dictation tool that transcribes speech in real time, strips filler words automatically, and transforms raw dictation into polished text without requiring an internet connection. The app runs on Gemma-based on-device ASR models, offers an optional cloud mode using Gemini for text cleanup, carries no subscription and no usage caps, and includes a personal vocabulary dictionary that can import frequently used words directly from a user’s recent Gmail history. It appeared in the App Store on 6 April 2026 with no press release or announcement. An Android version is referenced in the App Store listing but has not yet appeared on Google Play.
Google released a dictation app last Sunday and did not tell anyone. Google AI Edge Eloquent appeared in the iOS App Store on 6 April 2026 with no announcement, no blog post, and no press event, the kind of launch that invites the word “quietly” in every subsequent headline. The app is free, requires no subscription, and places no cap on usage. It runs speech recognition on-device using Gemma-based ASR models, which means recordings do not need to leave the phone. Given that the two most popular premium dictation apps for iPhone currently charge between $85 and $180 a year, that combination of features is not a minor release.
How it works
Opening Eloquent presents a dictation interface with a live waveform. As you speak, the app transcribes in real time. When you pause or stop, it automatically processes the raw speech: filler words , “um”, “ah”, and similar verbal placeholders, are removed, and the surrounding text is smoothed into readable prose. The cleaned transcript is copied to the clipboard automatically, ready to paste wherever it is needed.
A toggle in the top-right corner switches between two processing modes. In fully offline mode, all audio stays on the device and is processed by the Gemma-based ASR model locally, nothing is sent to a server. In cloud mode, the speech recognition still begins on-device, but Gemini models handle the text cleanup in the cloud. The distinction matters for privacy-sensitive contexts: users in regulated industries, or anyone wary of uploading voice data to a remote server, have a credible fully local option. The growing demand in 2026 for AI tools that process data locally rather than sending it to third-party servers has become a primary consideration in enterprise and professional software procurement, and Eloquent addresses it in the first toggle the user sees.
The feature set
Beyond the core transcription, Eloquent includes four text transformation tools: “Key points” extracts the main ideas from the dictation as a bulleted list; “Formal” rewrites the transcript in a more professional register; “Short” condenses it; and “Long” expands it. A history tab retains all previous transcriptions, each deletable individually. Usage statistics track cumulative word count and words per minute, a detail aimed at productivity-conscious users who want to measure how much they are actually dictating.
The most notable secondary feature is a personal context dictionary. Users can manually add names or technical jargon to improve transcription accuracy for domain-specific vocabulary. Optionally, users who sign in with a Google Account can allow Eloquent to import frequently used words from their recently sent Gmail messages, building a vocabulary profile without requiring any deliberate configuration. It is the one point in the app where Google’s wider data ecosystem appears, briefly and optionally, in what is otherwise a self-contained local tool.
iOS first, and what that signals
The release on iOS before Android is an unusual move for Google. Android is Google’s own platform; it is where Google typically demonstrates new capabilities first, using Gemini Nano and the AI Edge SDK that run directly on Pixel and compatible hardware. Releasing Eloquent on iOS first, without a corresponding Android launch, suggests either that the app is an experiment in market positioning rather than a flagship product rollout, or that the iOS version of the underlying Gemma ASR models reached readiness before the Android configuration did. The App Store listing includes a reference to an Android version, so parity is presumably coming. The sequencing, however, means Google has launched a significant competitive product on Apple’s platform before its own.
The competitive impact
The subscription model that has defined the premium end of AI productivity tools looks considerably less defensible when a technology company the size of Google enters the market for free. The two most prominent standalone dictation apps for iPhone, Wispr Flow and Willow, both cost $15 per month and rely on cloud processing, with Wispr Flow routing audio through servers operated by OpenAI and Meta. SuperWhisper, the most popular privacy-focused alternative, costs $85 per year and runs locally but is available only on Mac. Eloquent is free, runs locally on iPhone, and has no usage limit. It does not require a paid tier to access offline processing. For users who have been paying a monthly subscription primarily because no credible free alternative existed, the competitive arithmetic has changed overnight.
The app also undercuts Apple’s own built-in dictation, which is free but offers no filler-word removal, no text transformation, and no vocabulary learning. The practical quality gap between Apple Dictation and a Gemma-powered model running cleanup on top of ASR output is meaningful for anyone dictating more than occasional sentences.
Google AI Edge and the on-device strategy
Eloquent is released under the Google AI Edge brand rather than Google’s consumer product umbrella, the same initiative that provides developers with the tools and SDKs to run AI models locally on Android and iOS devices. That positioning suggests the app has a dual purpose: to demonstrate a real-world application of on-device Gemma capabilities for developers and enterprises evaluating the platform, as well as to serve as a direct consumer product. The broader push by major technology companies to run capable AI models directly on consumer devices has accelerated significantly over the past year, driven partly by privacy demands and partly by the latency and cost advantages of eliminating the server round-trip. Google AI Edge Eloquent is one of the clearest demonstrations yet of what that push produces when applied to an everyday productivity task.
Voice as a primary interface for AI-assisted work has been a persistent promise in enterprise computing for years, consistently underdelivered because the friction of imprecise transcription outweighed the benefit of speaking rather than typing. Removing filler words, offering instant transformation modes, and doing all of it locally without a subscription changes the friction calculation. Whether Eloquent becomes a mainstream productivity tool or remains a developer showcase depends on whether Google continues to update and support it, a history that the company’s track record with iOS apps does not make entirely easy to predict. The app exists. It is free. It works offline. That is more than most of its paying competitors can say simultaneously. A year that established on-device AI as a viable alternative to cloud-dependent tools set the stage; Eloquent is one of the more direct consumer expressions of that shift.
