Google launches Gemini 3.5 Transcribe

Gemini 3.5 Transcribe’s real advantage is intent-aware transcription rather than raw word accuracy, positioning speech recognition as a developer-ready layer for voice agents and productivity apps.

• –Live API streaming targets interactive voice experiences with sub-second latency.
• –Interactions API supports recorded audio, speaker attribution, and word-level timestamps.
• –Custom vocabulary and dialect support should make it more practical for specialized domains.
• –Artificial Analysis measures word error rates of 4.0% for streaming and 2.6% for non-streaming use cases, though developers should benchmark their own audio.
• –Broad API availability could pressure dedicated transcription providers, especially for teams already using Gemini.