Voice texts often come out as a jumble of “ums” and false starts. Google says its new Gboard Rambler voice typing feature is built to clean that up. It runs on Gemini 3.5 Transcribe, a speech-to-text model Google announced on August 26, 2026.
Rambler at a glance
| Item | What’s reported |
|---|---|
| Price | Not named in reports |
| Availability | Select countries and languages (FoneArena) |
| Platforms | Android in Gboard; the Gemini app on macOS (English only); Chrome support described as coming soon |
| Developers | Public preview through the Gemini API |
| Model announced | August 26, 2026 |
9to5Google reports that the model handles self-corrections and removes filler words. Say “meet me at, uh, no wait, seven instead of six,” and a self-correcting model should keep the final version. That example is ours, not Google’s, and results will vary.
Google’s developer documentation describes two modes. The default “verbatim” mode keeps filler words and false starts. A “smart” mode removes disfluencies and applies formatting. 9to5Google says the model also auto-formats text and supports natural voice editing.
The model detects more than 85 languages, according to Google’s documentation. It can label up to three speakers in pre-recorded audio with timestamps, but that’s a developer feature for recordings. Gboard doesn’t offer it while you dictate.
On speed, 9to5Google, citing Google and the benchmarking firm Artificial Analysis, reports that time to a final transcript is 70% faster than Google’s Chirp 3 model from 2025. That compares two speech models. It doesn’t mean your messages arrive 70% faster. Google also hasn’t published how much time Rambler saves a user, so we make no claim about minutes gained.
| Measure | Streaming | Non-streaming | Source |
|---|---|---|---|
| Word error rate (Artificial Analysis) | 4.0% | 2.6% | 9to5Google, Aug 26, 2026 |
| Word error rate (FLEURS benchmark) | 5.50% | 5.04% | 9to5Google, Aug 26, 2026 |
Word error rate is the share of words a transcript gets wrong. A 4% rate means roughly 4 wrong words in 100, on the test material used. To see whether Rambler is offered to you, check the Gboard settings on your phone.
Why It Matters to You
Picture a single dad cooking dinner while helping with homework who’d rather speak a text than type it. (He’s an illustration, not a real person.) That’s the moment Rambler is meant for.
- Parents and commuters who dictate while their hands are busy
- People who find small phone keyboards painful, including some with arthritis or limited hand mobility
- Non-native speakers who talk more fluently than they type
- Anyone who’d like fewer filler words in a sent message, as long as you read it first
Why the numbers need a grain of salt
The error rates come from benchmarks, and your accent, background noise and phone may give different results, so read every message before you send it. Consumer access depends on product, language and country, per FoneArena, and Chrome support hasn’t shipped. The developer features are a separate matter: speaker labeling beyond three speakers is experimental, Google’s documentation says some options can’t be combined (smart mode with speaker labels, for one), and it warns that word-level timestamps may reduce overall accuracy. Finally, the figures come from Google and one benchmarker. We found no fully independent review of Rambler itself.
Rambler shares an idea with Google’s Gemini Live hands-free update: use your voice instead of a keyboard.
What is Gboard Rambler? A voice typing feature in Gboard on Android. It uses Gemini 3.5 Transcribe to filter filler words and handle voice edits, according to 9to5Google and FoneArena.
Is it available everywhere? No. FoneArena reports Rambler is available in select countries and languages. The macOS Gemini app supports English only.
Does it work in Chrome? Not yet. Reports say Chrome support is coming soon.
How accurate is it? Google’s reported error rates are 4.0% (streaming) and 2.6% (non-streaming) on one benchmark. Real-world accuracy can differ, so proofread before sending.
Sources
- 9to5Google: Gemini 3.5 Transcribe (Aug 26, 2026)
- FoneArena: Google rolls out Gemini 3.5 Transcribe with real-time transcription and 85+ language support (Aug 27, 2026)
- Speech Technology Magazine: Google Releases Gemini 3.5 Transcribe (Aug 31, 2026)
- Google AI for Developers: Gemini 3.5 Transcribe documentation (accessed Oct 13, 2026)
Google’s consumer announcement page was not available, so Rambler details rest on the outlets above.
