Google has shipped Gemini 3.5 Live Translate, a near real-time speech translation system now live in Google AI Studio, Google Translate, and Google Meet. The system targets natural, fluid output rather than the clipped, mechanical cadence that defines most machine translation pipelines.
The Meet integration is the sharpest edge here. Cross-language conversations in video calls have historically required post-hoc captioning or human interpreters. Gemini 3.5 Live Translate pushes that latency close to zero, which changes the practical calculus for multinational teams and live interviews. The underlying model is Gemini 3.5, meaning the same architecture handling reasoning and code generation is now running live audio translation.
The original post is worth reading for the technical framing around what 'natural' means in Google DeepMind's evaluation criteria, how latency is measured across the three platforms, and which language pairs are currently supported at launch. The naturalness problem in speech translation is harder than the accuracy problem, and Google is making a specific claim here that deserves scrutiny.
[READ ORIGINAL →]