Google's Gemini 3.5 Transcribe

Gemini 3.5 Transcribe is available via the Gemini API and Google AI Studio, replacing Chirp 3 with 70% faster time to final transcription. Artificial Analysis measured a Word Error Rate of 4.0% for streaming and 2.6% for non-streaming; Ars Technica reported a live-speech error rate of 5.5%, down from 7.32% for Chirp 3. Rather than producing literal word-for-word output, the model handles filler word removal, self-correction, and auto-formatting, and supports custom vocabulary across 85+ languages. Developer access runs through two API endpoints: the Live API for real-time WebSocket streaming with sub-second latency (sessions capped at 10 minutes) and the Interactions API for pre-recorded audio up to one hour, with timestamps and speaker identification. On consumer surfaces, it powers Gboard's Rambler feature on Pixel 11 and is set to expand to the Gemini app on macOS, Google Antigravity, and Chrome.