Gemini 3.5 Transcribe is the newest speech‑to‑text model, delivering high‑precision, intelligent transcription in real‑time interactions.
The service offers two APIs: a live streaming endpoint gemini-3.5-transcribe-live with bidirectional streaming and sub‑second latency, and a pre‑recorded endpoint gemini-3.5-transcribe that provides speaker attribution and word‑level timestamps for meetings, calls, and other audio files.
Key capabilities include:
- Smart transcription: automatically removes filler words (e.g., “um”, “ah”), corrects self‑edits (e.g., “let’s meet Tuesday—no, Wednesday”), and formats the output.
- Function calling: can delegate complex tasks such as image generation or file analysis to other Gemini models via function calls.
- Custom vocabulary: adapts to user‑provided jargon and special spellings.
- Global language support: detects and transcribes over 85 languages, handling regional accents.
- Multi‑speaker identification: attributes speech to up to three speakers in pre‑recorded audio (experimental support for more).
- Live language switching: seamlessly switches languages within a single stream.
Performance metrics show an average Word Error Rate (WER) of 4.0% for streaming and 2.6% for non‑streaming use‑cases, maintaining low error rates even in noisy environments. Compared with the previous Chirp 3 model, overall latency improves by roughly 70%. On the FLEURS benchmark, the model achieves 5.50% WER in streaming mode and 5.04% in non‑streaming mode.
In‑product integrations include the Gemini macOS app, the Rambler feature on Android, Gboard, Google Antigravity, and Google AI Studio, all offering smart transcription and voice commands. Chrome support is forthcoming, enabling voice‑to‑text in any web field.
Developers can leverage the Gemini Live API together with platforms such as Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents to build high‑performance voice‑driven interfaces, while the platforms handle the underlying real‑time media infrastructure. Companies like vivo, Intellitek Health, and Lingopal have reported impressive latency, accuracy, and language coverage.
Availability: public preview is open in Google AI Studio and the Gemini Enterprise Agent Platform; an enterprise‑grade offering is coming soon; end‑users can try the feature in the macOS Gemini app, Android Rambler (selected regions), and the upcoming Chrome integration.
Review