AI company Tavus has introduced Griffin, a Human Interaction Model (HIM) designed to handle real-time, face-to-face video conversations. The company says the system achieved a 48% human success rate in a video Turing test, highlighting progress in AI systems designed for more natural visual interaction.
Unlike conventional conversational AI systems that typically connect speech recognition, language generation and separate voice or video-generation components, Griffin is designed as a full-duplex video-to-video system. This allows it to process visual and audio inputs while simultaneously generating responses and facial movements.
According to Tavus, Griffin is built to see, hear, interpret, speak, move and respond during an ongoing conversation rather than simply producing isolated facial animations.
The system uses a dual-engine architecture aimed at supporting real-time interaction. Tavus also says Griffin incorporates temporal awareness, allowing the model to account for events and changes occurring across a conversation rather than treating every frame independently.
The reported 48% result comes from Tavus’s video Turing test, where human participants evaluated AI-generated interactions. The figure represents the company’s reported test outcome rather than an independent industry benchmark.
Tavus has made an early research-preview version of Griffin available to selected developers. The development reflects a broader shift in AI research toward interactive systems that combine language, vision, audio and real-time physical or facial responses.
