- Google released “Google AI Edge Eloquent” on iOS on April 6, 2026 — a free, offline-first dictation app powered by on-device Gemma ASR models.
- The app automatically strips filler words and mid-sentence corrections, producing polished text after each dictation pause.
- An optional cloud mode routes text cleanup through Gemini models; disabling it keeps all processing fully local.
- An Android version is referenced in the App Store listing but has not yet launched.
What Happened
Google on April 6, 2026 quietly published a free iOS app called “Google AI Edge Eloquent” to the App Store without a formal announcement, according to TechCrunch. The app performs speech-to-text transcription using on-device Gemma-based automatic speech recognition (ASR) models, meaning no internet connection is required once those models are downloaded. It enters a market occupied by established tools including Wispr Flow, SuperWhisper, and Willow, which have attracted users who want more refined transcription output than native OS tools provide.
Why It Matters
AI-assisted dictation has matured into a competitive product category, with apps like Wispr Flow gaining traction among professionals seeking automatic cleanup of spoken speech rather than verbatim transcription. Google’s entry introduces a major platform vendor with established on-device model infrastructure into a space previously occupied by smaller independent developers. The Gemma family of open-weight models has been deployed across other edge-inference applications; Eloquent represents one of the first consumer-facing iOS apps to ship Gemma-based ASR inference directly on device for a productivity use case.
Technical Details
Once Gemma-based ASR models are downloaded to the device, all transcription runs locally without a network connection. After each dictation pause, the app automatically removes filler words — specifically “um,” “uh,” and mid-sentence self-corrections — before presenting cleaned output. Google’s App Store description states: “Unlike standard dictation software that transcribes stumbles and filler words verbatim, Eloquent utilizes AI to capture your intended meaning. It automatically edits out ‘ums,’ ‘uhs,’ and mid-sentence self-corrections, outputting clean, accurate prose.”
Post-processing transforms labeled “Key points,” “Formal,” “Short,” and “Long” allow users to reshape transcript content after the fact. An optional cloud mode routes text cleanup through cloud-hosted Gemini models rather than on-device Gemma; disabling it returns all processing to local execution. The app can import domain-specific vocabulary — names, technical terms, and jargon — directly from a connected Gmail account, or accept manually entered custom words. The interface also surfaces session history with full search, per-session word counts, and words-per-minute speed metrics.
Who’s Affected
Direct competitors — Wispr Flow, SuperWhisper, and Willow — face competitive pressure from a vendor with significantly larger distribution reach and an existing model ecosystem. iOS users who currently pay for third-party dictation apps now have a free, on-device alternative backed by Google’s infrastructure. Android users are not yet served: the App Store listing references “seamless Android integration” and a floating-button feature comparable to Wispr Flow’s Android implementation, but no release date for an Android version has been confirmed. Enterprises with data-residency requirements may find the fully local processing mode relevant, though Google has not made formal claims about the app’s compliance posture.
What’s Next
Google has not published a blog post or press release about Eloquent; TechCrunch reported it had reached out to Google for comment but had not received a response by the time of publication on April 6. The App Store description’s reference to Android integration — including the ability to set Eloquent as a system-wide default keyboard across any text field — indicates a broader rollout is planned beyond the current iOS-only release. No timeline for that Android launch has been disclosed in any available documentation.
