Tavus Announces Griffin: AI Model Passes Video Turing Test with 48% Human Pass Rate
On Oct 1, 2026, Tavus unveiled Griffin, a unified Human Interaction Model (HIM) for real-time, full-duplex video-to-video AI. In a live study, 48% of participants in 1-minute video calls believed they interacted with a real human, versus <3% for prior systems. Limited "Lite" version in testing.
TLDR
Griffin represents a potential inflection point in human-AI interaction, moving beyond voice agents to video-based conversations handling vision, audio, gestures, and dialogue simultaneously. The high pass rate and practical demos (e.g., interactive coaching) fuel both excitement and concern about deepfakes, AI alignment, and the future of human-computer relations. Tavus's official post generated 14k+ likes and thousands of reposts, with influencers amplifying demos. It exemplifies how AI milestones that blur human-AI boundaries capture rapid social amplification.
Combined views
25.3K
1 Source, first seen 10h ago
