An open audiovisual model that keeps perceiving while it talks and chooses every 0.71 seconds whether to speak or wait. Weights are public; timing is not yet evaluated.
Editorial summary · not the author abstract. Read the linked paper for the complete abstract and claims.Figure 1: illustrative full-duplex interaction scenarios for the Reciprocal World Model. Original figure · paper authors ↗
Release evidence
Apache-2.0 repository with offline and streaming inference code, but no training code. Duo-1.0 and Duo-1.2 checkpoints are public on Hugging Face; no training data was found.