Gemini 3.5 Transcribe Enters Public Preview: Separate APIs for Live Streaming and Recorded Audio

Google announced Gemini 3.5 Transcribe on August 26, 2026, in public preview. It splits into gemini-3.5-transcribe-live for real-time streaming and gemini-3.5-transcribe for recorded audio, with average word error rates of 4.0% streaming and 2.6% non-streaming as measured by Artificial Analysis.

Google Gemini speech-to-text API voice-agents

Google announced the speech-to-text model Gemini 3.5 Transcribe on August 26, 2026¹. It arrives in public preview through the Gemini API in Google AI Studio and the Gemini Enterprise Agent Platform, split into two APIs by use case: gemini-3.5-transcribe-live on the Live API, which streams bidirectionally with sub-second latency for interactive voice apps, and gemini-3.5-transcribe on the Interactions API, which transcribes recorded audio with speaker attribution and word-level timestamps¹.

On accuracy, Google cites measurements by Artificial Analysis: an average word error rate of 4.0% for streaming and 2.6% for non-streaming use cases¹. The model automatically detects and transcribes over 85 languages and handles self-corrections and filler-word removal, though speaker attribution in pre-recorded audio covers up to three speakers, and Google notes that support for 3+ speakers is experimental¹. It already runs behind Gboard’s Rambler on Android and the Gemini app on macOS; Chrome is listed as coming soon, and the post gives no pricing¹.