Gemini 3.5 Transcribe
Audio modelCard summarizedOur latest speech-to-text model designed for precise and intelligent real-time transcription
● Re-checked 12 Oct 2026 · 37 facts · 0 corrections · 1 unconfirmedSource: Gemini 3.5 Audio (Live Translate, Transcribe, Transcribe Live) - Model Card
Good for
- Real-time, context-aware transcription
- Agentic voice features
- Low-latency conversational agents
Watch out for
Overview
Google · 02Transcribe, front
3.5 Transcribe
Our latest speech-to-text model designed for precise and intelligent real-time transcription
Typical price
$0.005 / min
Pre-recorded audio, blended estimate
- Max length
- 1 h per file (30 min with diarization or timestamps); Live sessions 10 min
- Languages
- 85+ languages (auto-detected, incl. code-mixing)
- Modes
- speech-to-text, real-time streaming (Live API)
- Released
- 26 Aug 2026
One card covers Gemini 3.5 Live Translate, Transcribe and Transcribe Live ("Gemini 3.5 Audio"), based on Gemini 3 Pro and described as cost-efficient and fast for high-volume, latency-sensitive tasks. Architecture, training data and safety policies defer to the Gemini 3 Pro card; the card itself reports no benchmark numbers. Model information
gemini-3.5-transcribe- Typical price
- $0.005 / min
- Max length
- 1 h per file (30 min with diarization or timestamps); Live sessions 10 min
- Languages
- 85+ languages (auto-detected, incl. code-mixing)
- Weights
- Closed
- Released
- 26 Aug 2026
- speech-to-text
- real-time streaming (Live API)
- speaker diarization (up to 8 speakers)
- word-level timestamps
- custom vocabulary (up to 1,000 terms)
- smart dictation