Google announced Gemini 3.5 Transcribe on August 26, 2026 — its most precise speech-to-text model yet, built for real-time voice apps and recorded-audio transcription alike. It cuts word error rate to as low as 2.6% for pre-recorded audio and 4.0% for live streaming, replacing the older Chirp 3 model across Google's products with a 70% improvement in time-to-final-transcription latency. It ships as two API surfaces — gemini-3.5-transcribe-live for sub-second real-time streaming and gemini-3.5-transcribe for recorded audio with speaker attribution — both in public preview via Google AI Studio and the Gemini Enterprise Agent Platform.
Source: Google BlogWhat Is Gemini 3.5 Transcribe?
Gemini 3.5 Transcribe is Google's newest speech-to-text model, announced by the Gemini Audio team as designed for "intelligent voice interactions." Unlike conventional speech recognition models that struggle with background noise, complex jargon, and disfluency cleanup, it converts raw audio directly into accurate, polished, formatted text — not just a literal word-for-word transcript.
Consumers reach it through the Gemini app, the Rambler dictation feature on Android, and the Gemini app on macOS. Developers get access through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform.
Two APIs: Real-Time Streaming vs. Pre-Recorded Transcription
Google split the model into two API paths depending on whether audio is live or already recorded:
| API | Model ID | Built for | Key trait |
|---|---|---|---|
| Live API | gemini-3.5-transcribe-live | Interactive voice apps | Continuous, bidirectional streaming with sub-second latency |
| Interactions API | gemini-3.5-transcribe | Recorded audio, meetings, call logs | Speaker attribution and word-level timestamps |
Key Capabilities
- Smart transcription: handles self-corrections, strips filler words, and auto-formats the output text.
- Function calling: can delegate complex tasks to other Gemini models mid-transcription.
- Custom vocabulary: recognizes specialized jargon and unusual spellings you supply.
- Global language coverage: automatically detects and transcribes more than 85 languages.
- Multi-speaker identification: attributes speech reliably for up to three speakers, with support for more still experimental.
Performance: How Much Better Is It Than Chirp 3?
Word Error Rate (WER) measures the percentage of words a model gets wrong — lower is better. Google's own figures for Gemini 3.5 Transcribe:
- 4.0% WER for streaming transcription.
- 2.6% WER for non-streaming (pre-recorded) transcription.
- On the FLEURS multilingual benchmark: 5.50% WER in streaming mode and 5.04% WER in non-streaming mode.
- 70% improvement in time-to-final-transcription latency compared to the previous Chirp 3 model.
Where You'll See It: Rambler, Antigravity, AI Studio, macOS, and Chrome
Gboard's Rambler feature (Android)
Rambler uses Gemini 3.5 Transcribe to turn spoken thoughts into well-formatted text, filtering out filler words automatically, with voice-based editing so you can revise what you dictated by speaking follow-up commands.
Google Antigravity
Pairs screen context and chat history with the transcription model for more accurate recognition of file names and document references while you talk.
Google AI Studio
Available in Build mode for voice-driven coding sessions.
Gemini app on macOS
Lets you issue voice commands paired with screen context to summarize local files, repurpose text across apps, or generate images at your cursor — using just your voice. Currently English-only.
Chrome
Voice-to-text in web form fields is coming soon.
Developer Ecosystem and Availability
Integration partners already building on Gemini 3.5 Transcribe include Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents, with early testimonials from Vivo, Intellitek Health, Lingopal, Stream, and Agora.
Rollout by audience:
- Developers: public preview via the Gemini API in Google AI Studio and Google Antigravity.
- Enterprises: public preview via the Gemini Enterprise Agent Platform, with Gemini Enterprise for Customer Experience coming next.
- General users: live now in the Gemini app on macOS (English) and Rambler on Android (select countries and languages), with Chrome support coming soon.
FAQ
What is Gemini 3.5 Transcribe?
It's Google's newest speech-to-text model, announced August 26, 2026, designed for precise, intelligent transcription of both live and recorded audio. It converts raw speech into accurate, polished, formatted text rather than a literal word-for-word transcript.
How accurate is Gemini 3.5 Transcribe compared to Chirp 3?
Google reports a 70% improvement in time-to-final-transcription latency over its previous Chirp 3 model, along with word error rates of 4.0% for streaming and 2.6% for non-streaming transcription.
What's the difference between gemini-3.5-transcribe and gemini-3.5-transcribe-live?
gemini-3.5-transcribe-live runs through the Live API for continuous, bidirectional, sub-second-latency streaming in interactive voice apps. gemini-3.5-transcribe runs through the Interactions API for recorded audio like meetings and call logs, adding speaker attribution and word-level timestamps.
How many languages does Gemini 3.5 Transcribe support?
It automatically detects and transcribes more than 85 languages without requiring you to specify the language in advance.
Can Gemini 3.5 Transcribe identify different speakers?
Yes. It reliably attributes speech for up to three speakers, with support for larger groups still considered experimental.
Is Gemini 3.5 Transcribe available to developers yet?
Yes, it's in public preview for developers through the Gemini API in Google AI Studio and through Google Antigravity, and in public preview for enterprises via the Gemini Enterprise Agent Platform.
What is Rambler and how does it relate to Gemini 3.5 Transcribe?
Rambler is a dictation feature in Gboard on Android that uses Gemini 3.5 Transcribe under the hood to turn rambling spoken thoughts into clean, formatted text, stripping filler words and supporting voice-based edits.
Further reading:
- What Is RVC (Retrieval-based Voice Conversion) and How to Run It 100% Locally (2026 Guide)
- Context Engineering: The Core Discipline of AI Engineering in 2026
- Evaluation-Driven Development: Why Evals Are Becoming the New Unit Tests
- Multi-Agent Orchestration and Governance: Why Most AI Agent Pilots Never Reach Production