Google's Gemini 3.5 Transcribe 70% Faster

Google's Pixel 11 already runs on Gemini 3.5 Transcribe, quietly handling voice input through the Gboard "Rambler" feature. Now the company is confirming what that model can do, and preparing to spread it across the rest of the Google ecosystem. The pitch centers on speed and accuracy: Google says the new model processes voice-to-text about 70 percent faster than Chirp 3, its previous transcription engine, while cutting the live-speech error rate to 5.5 percent from 7.32 percent.

That accuracy gain is modest on paper, but Google is betting the real selling point is elsewhere. Gemini 3.5 Transcribe isn't just faster at capturing words, it's designed to clean up how people actually talk. Filler words, self-corrections, rambling asides, the model is built to smooth all of it into readable text in real time, adjusting on the fly if a speaker backtracks mid-sentence. It also draws on custom vocabulary lists to handle jargon specific to a user's field, and supports 85 languages, with the ability to distinguish up to three speakers in pre-recorded audio.

Whether that editing instinct actually matches what a speaker meant to say, rather than just what sounded cleaner, is the open question Google's numbers don't answer.