Last updated: 2026-10-25
Released in April 2026 by Google's AI Edge team, Eloquent is a free dictation app for iPhone and iPad built on Gemma speech models. It dictates in many languages, has no limit on dictation length, and by default cleans up speech on the device rather than sending audio to a server.
About Google AI Edge Eloquent
Google AI Edge Eloquent is a dictation app built by Google's AI Edge team that launched on iOS in April 2026. It turns spoken thoughts into polished prose instead of a raw verbatim transcript, cutting filler words like ums and uhs and cleaning up mid-sentence corrections. The app runs on Gemma-based speech models from Google's on-device AI Edge platform (see Gemma 4 12B and Gemma 4 31B for the model family), which sits between dictation tools that transcribe word for word and cloud services that add latency and a privacy tradeoff. It is published by Google and listed in the AI voice dictation category.
Per the App Store listing, all machine learning processing runs locally on the device once the speech model downloads on first launch, and the app works offline afterward, though the listing adds that some advanced optional features require the cloud. A cloud mode for deeper rewrites using Gemini is optional. The app also builds a personal context dictionary from the user's vocabulary, and the listing says users can optionally connect a Google account to seed it, with that data processed on the device.
Version 1.4.0 (13 August 2026) added upload of audio and video files from Photos or Files and removed the limit on dictation length, and the release notes describe no cap on the number of dictations. The August update after it brought in 14 more languages, which the release notes call 15 in total, along with a refreshed interface. In the same review thread Google answered a microphone complaint with improved auto-detection and manual selection in the settings.
Eloquent suits people who think out loud and want a usable first draft: journalists filing quick notes, students working from lecture recordings, and professionals dictating emails on the move. Compare it with Wispr Flow, which dictates into whichever app has focus on Mac, Windows, iOS and Android, with Contextli, which formats dictation for Gmail, Slack and Notion, and with TurboScribe for bulk transcription of uploaded files. Not sure which fits? Answer a few questions in Smart Match for a shortlist.
Where the alternatives win: Eloquent has no system-wide keyboard yet, which the listing flags as on the way, so cleaned text has to be copied into other apps, while Wispr Flow types straight into them. It cannot label speakers, so a two-person conversation comes back as one transcript and is a poor fit for meetings or interviews; browse AI meeting assistants such as Fathom, Fireflies or Granola for that job, or NotebookLM if the goal is working with notes after transcription. It runs on iPhone and iPad (iOS and iPadOS 16.0 or later) and Apple Vision (visionOS 1.0 or later); Google Play turned up no Android listing when this page was checked. Anyone on Android or Windows should look at the alternatives above instead.
Eloquent is free with no subscription, no in-app purchases and no stated usage cap as of October 2026, distributed under Google's AI Edge developer umbrella rather than as a numbered Google One tier. The download is 137.2 MB on the App Store, before the speech model fetched on first launch.
Pricing
No cost to download or use as of October 2026: the App Store lists the app as free with the subtitle "Unlimited, no cost dictation", and the listing says it operates free of charge with no usage limits. It sits under Google's AI Edge developer program rather than a numbered Google One plan, and Google has made no commitment about pricing later. For a paid alternative with a free tier, see Wispr Flow.
Key Features
- On-device speech cleanup: Runs Gemma-based speech models locally on the iPhone to rewrite spoken audio into polished text, cutting filler words and false starts, with some advanced optional features needing the cloud per the App Store listing.
- Audio and video file transcription: Accepts recordings from the Photos or Files app and transcribes and polishes them; version 1.4.0 (13 August 2026) put no cap on file size or dictation length.
- Personal context dictionary: Builds an editable dictionary from the user's vocabulary to improve recognition of names and jargon, and can optionally draw on a connected Google account, processed on the device.
- Optional cloud rewrite mode: Lets users opt into Gemini-powered cleanup for heavier rewrites, while the default stays on-device so audio is not sent to a server unless the user turns the cloud features on.
- 15 dictation languages: Version 1.4.1 (21 August 2026) added Spanish, German, Italian, Portuguese, Chinese, Japanese, Korean, Hindi, Indonesian, Russian, Thai, Turkish and Vietnamese among others, and the release notes cite 15 languages in total.
Pros
- Unlimited use at zero cost, with no usage limits stated as of October 2026, unlike the metered free tier on Wispr Flow.
- Core transcription and cleanup run on the device and work offline after the first-launch model download, unlike cloud-first dictation services.
- Dictates in Spanish, German, Japanese, Korean, Hindi and other languages since the August update, and a personal dictionary improves accuracy on names and jargon.
Cons
- iPhone, iPad and Apple Vision only as of October 2026. Google Play shows no Android listing, so Android and Windows users have no path to the app.
- No system-wide keyboard yet. The App Store copy says "Keyboard coming soon", so text has to be copied and pasted into other apps rather than dictated where the user is typing.
- Cannot separate speakers, so a two-person conversation returns as a single undifferentiated transcript, ruling out meeting or interview use.
Data Handling
- Training-data policy
- On-device processing by default; audio does not leave the device unless the user opts into cloud mode, which uses Gemini for text cleanup.
Frequently Asked Questions
What does Google AI Edge Eloquent cost in October 2026?
Nothing. The listing shows a single free price with no tiers, paid upgrades or in-app purchases. Google has not said whether that changes later, so check the App Store before relying on it long term.
Are there limits on how much I can dictate?
The release notes say there is no cap on the number of dictations, dictation length or uploaded file size. The practical limits are the device and platform: it needs iOS or iPadOS 16.0 or later, and it cannot label separate speakers in one recording.
Does Google AI Edge Eloquent work on Android or Windows?
Not as of this review. The listing covers iPhone, iPad and Apple Vision, and a search of Google Play found no Eloquent listing. Android and Windows users can look at Wispr Flow, which covers Mac, Windows, iOS and Android.
Which languages does Eloquent support?
English plus a long list of others: an August 2026 update widened dictation to European, East Asian and South Asian languages, so most major ones are covered.
Does my audio leave the phone?
By default the speech model runs on the device and the app works offline after its first-launch download. The App Store listing says some advanced optional features need the cloud, and the Gemini cloud rewrite mode only applies if you turn it on.