Tavus released Griffin, its first 'Human Interaction Model' (HIM), on October 1, 2026, saying it can understand and carry on face-to-face conversations in real time. It is the company's first major update in its AI video avatar development since expanding into live video conversations with digital personas.
Tavus reported Griffin scored 3.83 points in an independent Nvidia test measuring how human an AI feels in direct audio-video conversation. That compares with actual humans scoring 3.92 and the previous best AI model hitting 2.80.
Griffin is built on Tavus' advanced AI platform and targets use cases such as tutoring, practicing difficult conversations, and camera-based tech support. Availability begins with a research preview for select testers, initially for those interested in early access.
"Griffin processes speech, facial expressions, tone of voice, gestures, and pauses while both receiving and generating video," said Matthias Bastian, Tavus' founder. Tavus said the model's ability to mimic human interaction is a key innovation in AI video communication.
The announcement follows Tavus' expansion into live video conversations with digital personas. Tavus said the development reflects its commitment to creating more natural and immersive AI interactions.
Tavus did not say when the more capable version of Griffin will be available, and raised questions about safety concerns that need to be addressed. Tavus said it will continue to refine Griffin based on user feedback and testing results.
Source: thedecoder
