OPEN LINK ↗
// 59m agoMODEL RELEASE
Tavus Griffin Passes Video Turing Test
Tavus introduced Griffin, a real-time, full-duplex video-to-video model that sees, hears, interrupts, gestures, and responds during face-to-face conversations. In Tavus’s one-minute study, 26 of 54 participants believed Griffin-Lite was a real person; the research preview is limited to selected testers.
// ANALYSIS
Griffin is a meaningful interface breakthrough, but “passes the Turing test” rests on a small, short study; the more durable achievement is continuous multimodal turn-taking.
- –Griffin unifies perception, conversational decisions, speech, facial expression, and gesture in one streaming system.
- –NVIDIA’s VideoFDB evaluation scored Griffin 3.83 for generation and 3.73 for perception, ahead of reported baselines.
- –The human study improved sharply over Tavus’s Phoenix-4.5 result: 48% recognition as human versus 2.4%.
- –Developers cannot deploy Griffin yet; Tavus says the preview is not customer-available while disclosure and safety work continue.
// TAGS
griffinmultimodalvisionspeechvideo-genevaluationbenchmarkagent
DISCOVERED
59m ago
2026-10-01
PUBLISHED
1h ago
2026-10-01
RELEVANCE
9/ 10
AUTHOR
faststocknewss