HomeAIGoogle DeepMind Releases Gemini 3.5 Tr
AI

Google DeepMind Releases Gemini 3.5 Transcribe Model

Google DeepMind announced a speech-to-text model featuring sub-second streaming latency, automated filler cleanup, and support for over 85 languages.

WHAT YOU NEED TO KNOW
  • Gemini 3.5 Transcribe records an average Word Error Rate of 4.0% for streaming and 2.6% for non-streaming audio.
  • The model cuts time to final transcription by 70% compared to the earlier Chirp 3 system.
  • Developers can access the model in public preview through the Gemini API in Google AI Studio and Google Antigravity.

Google DeepMind introduced Gemini 3.5 Transcribe on Wednesday, releasing a speech-to-text model designed for real-time and pre-recorded audio transcription across more than 85 languages.

The system converts raw audio into formatted text and is available across two separate interfaces. The Live API provides bidirectional streaming with sub-second latency using gemini-3.5-transcribe-live, while the Interactions API handles pre-recorded audio processing through gemini-3.5-transcribe. According to tests by Artificial Analysis, the model achieves an average Word Error Rate (WER) of 4.0% in streaming mode and 2.6% for non-streaming audio. Compared to Google's previous Chirp 3 model, time to final transcription improved by 70%. On the multilingual FLEURS benchmark, the model posted a 5.50% WER in streaming and 5.04% in non-streaming use.

Google DeepMind designed the architecture to handle natural speech disfluencies. The model strips out filler words like "ums" and "ahs", formats text automatically, and resolves verbal self-corrections on the fly. It processes alphanumeric strings including order IDs and postal codes, adapts to custom vocabulary lists, and attributes timestamps for up to three speakers in recorded audio. Speaker identification for more than three voices remains experimental.

Integrations and developer access

Google integrated the model into multiple existing tools. On Android, Gboard uses the technology through a feature called Rambler in select countries and languages, allowing speakers to dictate edits and adjust writing styles. The Gemini application on macOS uses the model in English, enabling voice commands that delegate background tasks like image creation and file summarization to other Gemini models.

Developers can access Gemini 3.5 Transcribe in public preview through the Gemini API in Google AI Studio and Google Antigravity. Streaming infrastructure platforms supporting the model include Agora, Fishjam, LangChain, LiveKit, Pipecat, Vercel, and Vision Agents. Google said the model will roll out in the future to web fields in Chrome and to Gemini Enterprise for Customer Experience.

Xentir Media
Xentir Media NewsroomSource-backed AI and technology coverage, drafted by Xentir's automated editorial system under fixed human-set rules. See our editorial policy and AI usage policy.
J
Jomon · Founder & EditorFounder and editor of Xentir Media. Sets the editorial rules the newsroom system runs under, and is accountable for its corrections. About Jomon · [email protected]
The Xentir Brief
The developments worth knowing — one useful email.
Get the Brief →