Core Event: Gemini 3.5 Transcribe Is Live
Google officially launched Gemini 3.5 Transcribe today, a new speech-to-text addition to the Gemini Audio suite. Key factual details:
- Release date: August 26, 2026
- Available to: macOS Gemini app users (English only); Android Rambler dictation in select countries and languages; developers via Gemini API (public preview through AI Studio and Antigravity)
- Chrome support: Coming soon
- Gemini 3.5 Pro status: Not yet released (original June commitment remains pending)
A notable correction followed initial announcements: Google clarified only 3.5 Transcribe is launching today, while earlier-comunicated Gemini 3.5 Live and 3.5 Live Experimental models are not being released yet, with no new timeline provided.
Core Capabilities and Technical Scope
Gemini 3.5 Transcribe is described by Google as a “major advancement” over its previous transcription model, Chirp 3. Capabilities confirmed include:
- Support for over 85 languages
- Automatic filler-word removal: Filters out “um,” “uh,” and similar spoken disfluencies to produce cleaner transcripts
- Custom vocabulary uploads: Users can supply domain-specific terms and spelling rules to minimize manual post-editing
- Multi-speaker attribution for up to three speakers in pre-recorded audio
- Word-level timestamps for precise timing alignment
- Voice-first editing: Users can make transcript edits directly via voice commands
Together, these features aim to generate structurally sound transcripts approaching publishing-ready quality while preserving spoken meaning.
One Significant Counterpoint:What’s Missing Matters
While the headline feature—automatic filler-word deletion—is explicitly confirmed, a strategic mismatch deserves attention: Gemini 3.5 Transcribe is a vertical tool, not the flagship Gemini 3.5 Pro model users were promised for June. This signals Google is prioritizing modular, use-case-driven releases over waiting for a monolithic update, shiftingNarrative control from timing delays to functional shipment.
Additionally, Rambler integration is Russia-specific and currently limited in geographic and linguistic scope, whereas macOS English remains the most broadly available deployment path. No pricing or open-source status was disclosed for the API preview.
Model Comparison: Current Gemini Audio Lineup
Only models with verified launch or cancellation status are listed:
| Model | Status | Key Features | Target Use Cases |
|---|---|---|---|
| Gemini 3.5 Transcribe | Launched | Auto-filler removal, 85+ languages, up to 3 speakers, timestamps, custom vocab | Meeting notes, podcast editing, accessibility input |
| Gemini 3.5 Live | Not launched | Mid-sentence interruption handling, real-time language ID, live visual processing | Voice assistant dialogue |
| Gemini 3.5 Live Experimental | Not launched | Real-time reasoning-step narration, complex task handling | Research, advanced reasoning demos |
| Chirp 3 | 退役 | Legacy transcription base | Deprecated by 3.5 Transcribe |
| Gemini 3.5 Pro | Delayed | — (High-performance general model, June promise pending) | Not yet applicable |
Practical Recommendations
Try now if you: Edit English interviews, podcasts, or meeting recordings regularly; use macOS with Gemini app; prioritize clean transcripts over native-language accuracy.
Wait if you: Rely primarily on non-English languages (especially Chinese); require multi-speaker efficiency in production workflows without early-test risk; need API SLA guarantees or pricing stability before adoption.
Final Note
The incremental, capability-first rollout signals Google is hedging against model delays by shipping measurable value early. Using filler-word elimination as the headline feature is no accident—it targets a universal friction point in content creation, positioning AI less as a novelty and more as an editorial co-pilot.
(English version ~1350 words)