Tavus unveils Griffin, an AI model for real-time video conversations

San Francisco-based Tavus has launched Griffin, the first Human Interaction Model (HIM), capable of real-time video conversations with natural speech, facial expressions and gestures. The company says 48% of people who spoke with it live thought it was human, and it scored 3.83 out of 5 on NVIDIA's VideoFDB benchmark versus 3.92 for a human reference. Only Griffin-Lite is available to select testers for now.

• Griffin scored 3.83 of 5 on NVIDIA's VideoFDB benchmark; human reference scored 3.92
• 48% of live test participants believed the model was a real human
• The model reads speech, tone, facial cues and interruptions in real time
• Griffin-Lite is available to eligible testers only