Tavus, a San Francisco-based startup founded in 2020, has introduced Griffin, a Human Interaction Model designed for real-time face-to-face conversations. The system processes speech, facial expressions, tone, gestures, and pauses while generating video output.
In a company study, 48 percent of participants identified the AI avatar as a real person after a one-minute call, compared to 2 percent for previous systems. In an independent Nvidia test, Griffin scored 3.83 points on human-likeness, while actual humans scored 3.92 and the best previous AI model reached 2.80.
The company has raised approximately $64 million in funding to date. Initially focused on personalized marketing videos, Tavus has now shifted its development toward live digital personas for applications such as tutoring, technical support, and communication training.
A research preview version, Griffin-Lite, is currently available to select testers. The company plans to release a more capable version of the model once it addresses outstanding safety concerns.
