Overview
Gemini 3.5 Transcribe Live is a dedicated, low-latency speech-to-text model for bidirectional streaming over WebSockets through the Gemini Live API. It accepts audio and returns incremental text transcription, including interim and final results. Intended uses include real-time captions, microphone transcription and voice applications. (aistudio.google.com)
Capabilities and limits
- Automatically detects and transcribes more than 85 languages, including language switches within a session. It supports utterance-level timestamps and custom vocabulary biasing for up to 1,000 terms. (docs.cloud.google.com)
- Supports automatic voice activity detection with configurable sensitivity and manual endpointing. Smart dictation and formatting, including filler-word removal, is experimental on Agent Platform and requires manual endpointing for live use. (docs.cloud.google.com)
- On Agent Platform, live audio sessions are limited to 10 minutes. Live transcription does not support speaker diarization or word-level timestamps; those are features of the separate non-streaming model,
gemini-3.5-transcribe. (docs.cloud.google.com)
Release
The Gemini API release notes announce general availability on August 26, 2026. Google Cloud's Agent Platform documentation separately lists the Live preview release in August 2026 and the GA version in October 2026; the sources therefore give different platform-specific release timelines. (aistudio.google.com)
Pricing
Prices below are in USD, checked on November 16, 2026, and apply to this Live model, not the separate file-processing model.
- Gemini Developer API, standard paid tier: audio input is $3.50 per 1 million tokens; text output is $21.00 per 1 million tokens. The free tier lists input and output as free of charge, subject to applicable limits. Google estimates a combined approximately $0.009 per audio minute, based on 25 audio input tokens per second and 175 text output tokens per minute; this is an estimate, not a separate fixed-minute tariff. Official Developer API pricing. (aistudio.google.com)
- Google Cloud Gemini Enterprise Agent Platform: audio input is $3.50 per 1 million tokens and text output is $21.00 per 1 million tokens. Google states that only final committed tokens are billed and gives an estimated blended price of $0.008925 per audio minute, using the same token-rate assumptions. Official Agent Platform pricing. (cloud.google.com)