Build voice-first apps with Gemini 3.5 Transcribe
Google launches Gemini 3.5 Transcribe, an LLM-based transcription model with 70+ language support
“This model is an LLM based transcription model. So it can do really, really well in transcribing alphanumerics or email addresses, anything that normal transcription models might struggle with.”
Google has launched Gemini 3.5 Transcribe, its first Gemini-based transcription model, available on both the Interactions API and Live API for real-time use. Unlike traditional ASR models, it leverages LLM reasoning to accurately transcribe edge cases like email addresses, phone numbers, and alphanumerics. The model supports 70+ languages despite being configured for English, highlighting Gemini's broad audio understanding capabilities.