Tavus Unveils Griffin AI Model, Ranking First on NVIDIA Benchmark

San Francisco-based AI startup Tavus has introduced Griffin, a "Human Interaction Model" capable of real-time video and audio interaction. The model achieved a score of 3.83 out of 5.0 on NVIDIA’s VideoFDB benchmark, the highest among published systems. Unlike traditional chatbots that process data in sequential steps, Griffin uses a video-to-video architecture to reduce latency, allowing it to react, interrupt, and interact in real time. Tavus claims the model passed an internal video-based Turing test, with 48% of participants believing they were interacting with a human.