Researchers have introduced Realtime-Venus, a proactive full-duplex interaction system designed for natural dialogue in both digital and physical settings. According to the paper, such interaction demands continuous perception and timely responses: spoken dialogue depends on acoustic and linguistic cues, while video interaction additionally requires grounding the conversation in an evolving visual context.
The system is built on two separately trained 9B models. Realtime-Venus-Omni handles audio-visual interaction, while Realtime-Venus-Audio focuses on spoken interaction. Each model acts as a complete conversational frontend, integrating continuous perception, conversational control, and native speech generation through a shared causal timeline that covers user inputs, model outputs, and delegation events.
A dual-loop runtime coordinates live interaction with background reasoning and tool execution. The foreground conversation continues uninterrupted while Realtime-Venus-Harness executes tasks asynchronously and returns results for integration into the ongoing dialogue. Both models follow a common post-training recipe combining offline understanding, proactive full-duplex trajectories, and delegation workflows.
On the video side, Realtime-Venus-Omni reportedly achieves the highest scores among evaluated online models on six of eight benchmarks, including StreamingBench (70.2%), OVO-Bench (64.7%), and Daily-Omni (81.3%). For audio, Realtime-Venus-Audio leads the compared models on MMAU (78.0%), MMAU-Pro (63.2%), Llama Questions (83.8%), and Speech CMMLU (67.8%), while matching the best VoiceBench AlpacaEval score of 4.81.
On Full-Duplex-Bench v1.5, the audio model responds to 75% of user interruptions and achieves continuation rates of 97%, 88%, and 86% under backchannels, other-directed speech, and background speech respectively. The authors state these continuation results exceed those of Gemini 3.1 Live and GPT-4o on all three metrics.