Gemini 3.5 Transcribe Live

Low-latency real-time speech-to-text over WebSockets: the Live API endpoint of Gemini 3.5 Transcribe, for bidirectional streaming audio transcription. Status ga because Google's models index badges the single Gemini 3.5 Transcribe card 'New Stable' and that card names both endpoints, gemini-3.5-transcribe and gemini-3.5-transcribe-live; read 2026-09-01. Paid-tier rates per 1M tokens: input $3.50 on audio, output $21.00 on text — both above the unary endpoint, which the sibling row carries. Google also publishes duration-equivalent figures for the same rates (roughly $0.005 per minute of audio in and $0.004 per minute of text out, on an assumption of 25 audio tokens per second and 175 text tokens per minute, blending to about $0.009 per minute for Live Transcribe), a unit this schema has no field for. Free tier is free of charge. Grounding with Google Search is Not Supported. Feature parity with the unary endpoint is partial and Google states the differences: language auto-detection and custom vocabulary biasing yes, but word-level timestamps and speaker diarization NOT supported, and sessions are capped at 10 minutes rather than the 1 hour the file endpoint allows. No context window or max output token count is published, so both fields are null. Released August 2026 per Google's deprecations table — month only, which is all Google states — with 'No shutdown date announced' and the row not gray. Added 2026-09-01 alongside gemini-3.5-transcribe (LEARNINGS #105).

Source: Google official documentation ↗

Explore more

Gemini 3.5 Transcribe Live — questions & answers