What happened
Google just rolled out Gemini 3.5 Transcribe, a new AI transcription model that automatically cleans up spoken audio by removing filler words like "um" and "uh" while formatting the text into readable paragraphs. Announced on August 26, 2026, the model is the latest addition to the Gemini Audio lineup, following the earlier launch of Gemini 3.5 Live Translate. It arrives while the entrepreneurial and creator community is still waiting on the delayed Gemini 3.5 Pro model, which Google originally promised for June.
According to Google, 3.5 Transcribe is a major step up from its predecessor, Chirp 3, particularly in multilingual accuracy and word error rates. The model recognizes more than 85 languages and can automatically detect specialized jargon, technical terms, and unusual spellings without manual correction. It also supports voice-based editing, letting users revise a transcript naturally by speaking corrections instead of typing them.
Why it matters
For anyone who records podcasts, sales calls, interviews, or internal meetings, transcription cleanup has always eaten up time. A raw transcript full of "um," repeated words, and false starts is unusable for publishing or documentation without heavy editing. Gemini 3.5 Transcribe automates that entire cleanup pass, which is a meaningful shift for content teams and small businesses that can't afford a dedicated editor.
The model can also attribute speech to up to three speakers in pre-recorded audio and generate word-level timestamps, both of which matter for anyone producing captions, show notes, or searchable meeting archives. Combined with the custom vocabulary feature, which lets users feed in brand names, product terms, or industry jargon so the model stops mangling them, this positions Gemini 3.5 Transcribe as a serious competitor to dedicated transcription services rather than just a Google Assistant feature.
How to use it today
Gemini 3.5 Transcribe started rolling out on August 26, 2026 in English for all macOS Gemini app users. It's also powering the Rambler dictation feature on Android in select countries and languages, and developers can already access it in public preview through the Gemini API via AI Studio and Antigravity. Google says Chrome support is coming soon, though no date has been confirmed.
To get the cleanest results, feed the model a custom vocabulary list before transcribing anything with brand names, acronyms, or niche terminology — this avoids manual find-and-replace after the fact. For teams experimenting with AI-assisted content workflows more broadly, pairing a transcription tool like this with other free AI utilities, such as the ones available at [mykreatool.com](https://mykreatool.com), can speed up the full pipeline from raw audio to a polished, publish-ready post.
Note that Google initially told The Verge that companion updates — Gemini 3.5 Live and 3.5 Live Experimental — were also launching the same day, then walked that back after publication, saying those two models aren't live yet and declining to give a new release date. Only 3.5 Transcribe is confirmed and available right now.
Who benefits
Content creators and podcasters stand to save the most time, since transcript cleanup and speaker labeling are typically the most tedious part of turning audio into blog posts or show notes. Marketers running customer interviews or focus groups get a faster path from recording to usable insights, especially with jargon and product names transcribed correctly on the first pass.
Small business owners and consultants who record client calls or meetings can now generate accurate, readable transcripts without hiring a transcription service. Developers building voice-driven apps also benefit, since the model is already available via the Gemini API, letting them integrate cleaner transcription directly into their own products during the public preview.
Risks
Automatic filler-word removal is convenient, but it also means the transcript is no longer a verbatim record of what was said. For legal depositions, journalism, or any use case where exact wording matters, an edited transcript could create disputes about accuracy — users should keep the raw audio as the source of truth rather than treating the cleaned text as an official record.
The rollout is also inconsistent right now: English-only support on macOS, limited Android countries, and no Chrome availability yet means most non-English or web-based users can't access the feature at launch. Google's same-day walk-back on the Live and Live Experimental models is a reminder that Gemini Audio announcements have shifted before, so treat unconfirmed features with some caution until they're officially live.
Conclusion
Gemini 3.5 Transcribe shows Google prioritizing practical, everyday AI utility — cleaning up messy speech automatically — even as its flagship Gemini 3.5 Pro model remains overdue. With support for 85+ languages, three-speaker attribution, and word-level timestamps, it's a genuinely useful tool for creators, marketers, and developers who deal with audio regularly. The rollout is still limited, but for macOS Gemini app users in English, it's available to try right now.
Comments 0