Google today announced the launch of Gemini 3.5 Live Translate, a new audio model engineered for continuous, speech-to-speech translation. This release marks a significant evolution from the company’s pioneering machine learning initiatives that began two decades ago, which currently process over a trillion words monthly for billions of users worldwide. The model supports automatic detection across more than seventy languages. Unlike traditional sequential translation systems that require speakers to pause between exchanges, Gemini 3.5 Live Translate generates output continuously. This architecture maintains synchronization with the original speaker by balancing contextual accuracy with immediate audio generation, resulting in fluid delivery that typically lags only a few seconds behind the source. The system preserves original vocal characteristics, including intonation, pacing, and pitch. From a development standpoint, the model processes streamed audio natively, eliminating the need for manual language configuration. It incorporates enhanced noise robustness, allowing reliable operation in loud or acoustically challenging environments. These capabilities position the model for immediate integration into multilingual customer service calls, remote meetings, educational broadcasts, and live interpretation workflows. The technology is rolling out today across the Google product ecosystem. Developers and enterprise users can begin integrating the system to facilitate seamless, low-latency communication across language barriers, reinforcing Google’s continued expansion of real-time natural language processing capabilities.