Realtime-Venus is a full-duplex conversational AI system built from two separately trained 9B models — one for audio-visual interaction and one for spoken interaction — that integrate continuous perception, conversational control, and native speech generation on a shared causal timeline. A dual-loop runtime lets the system continue foreground conversation while a harness executes longer background tasks asynchronously and folds results back into the ongoing dialogue. The system leads compared online models on multiple video and audio benchmarks and outperforms Gemini 3.1 Live and GPT-4o on continuation rates after user interruptions.
