Realtime-Venus: A full-duplex interaction system with asynchronous delegation
The system keeps a live conversation running while separate reasoning and tool work happen in the background.
Realtime-Venus pairs separately trained 9B audio-visual and audio models with a shared causal timeline for user input, model output, and delegation events. Its runtime lets foreground dialogue continue while Realtime-Venus-Harness executes asynchronous tasks and feeds results back into the conversation. The paper reports leading scores for Realtime-Venus-Omni on six of eight video benchmarks, and strong audio results for Realtime-Venus-Audio across understanding, spoken QA, and interruption tests. HF Daily Papers' note
Realtime-Venus pairs separately trained 9B audio-visual and audio models with a shared causal timeline for user input, model output, and delegation events. Its runtime lets foreground dialogue continue while Realtime-Venus-Harness executes asynchronous tasks and feeds results back into the conversation. The paper reports leading scores for Realtime-Venus-Omni on six of eight video benchmarks, and strong audio results for Realtime-Venus-Audio across understanding, spoken QA, and interruption tests. HF Daily Papers' note
score 6