Google Unveils Gemini 3.5 Transcribe
Credit: AI
Google unveiled the Gemini 3.5 Transcribe model, featuring new speech transcription capabilities, support for specialized terms, and support for over 85 languages. The new model comes amid the delay of the flagship Gemini 3.5 Pro.
What’s more interesting than standard transcription is that the model automatically cleans up speech: it removes “ums,” takes into account human corrections on the fly, and can immediately produce well-formed text. Google specifically promotes it as the basis for voice agents, not just a transcription service.
In the FLEURS multilingual test, Google claims a recognition error of about 5%, and the time to final transcription, according to Artificial Analysis, has been reduced by approximately 70% compared to the previous solution, Chirp 3. These are still test results, and the quality in a noisy kitchen, factory, or meeting with poor microphones can vary greatly.
Gemini 3.5 Transcribe with English support is now available to users of the Gemini app for macOS. Additionally, the new Rambler voice input feature has been released in Gboard for Android in some countries and supports multiple languages. The model is also available to developers in public preview via the Gemini API in Google AI Studio and Antigravity. Gemini 3.5 Transcribe support is expected to arrive in Chrome soon.
