Sign InOpen Brain
VercelEngineering PostOfficial Source

Gemini 3.5 Transcribe now available on AI Gateway

Gemini 3.5 Transcribe adds batch and live WebSocket transcription to AI Gateway, with automatic language detection, 85+ languages, and custom vocabulary.

Vercel · Aug 26, 2026
Open Source Open MarkdownOpen JSON
Source Summary

**Gemini 3.5 Transcribe** has separate models for complete recordings and live WebSocket transcription. It supports **85+ languages**, automatic language detection, language switching, and custom vocabulary.

Practical Implication

Use batch transcription for stored agent inputs and **AI SDK V7 streamTranscribe** when an agent needs an updating microphone transcript; declare the raw audio format explicitly for live streams.

Agent-Ready Context
**Gemini 3.5 Transcribe** has separate models for complete recordings and live WebSocket transcription. It supports **85+ languages**, automatic language detection, language switching, and custom vocabulary.

Use batch transcription for stored agent inputs and **AI SDK V7 streamTranscribe** when an agent needs an updating microphone transcript; declare the raw audio format explicitly for live streams.

The material reports no accuracy, latency, speaker-separation, or cost results, and the streaming AI SDK functions shown are marked experimental.
Connected Context · Feed7 Judgment

This turns Gateway’s general streaming-transcription capability into a concrete split between stored-recording and live-input models, with multilingual switching, custom vocabulary, and an explicit raw-audio contract. It broadens deployable voice-input workflows but does not establish a production default because streaming functions are experimental and accuracy, latency, diarization, and cost remain unmeasured.

Context Map
toolsaudio#generative-media#agent-sdks#gateways
Uncertainty
The material reports no accuracy, latency, speaker-separation, or cost results, and the streaming AI SDK functions shown are marked experimental.