This AI can make you think you're talking to human
What's the story
San Francisco-based AI research company Tavus has unveiled Griffin, touted as the first human interaction model. Griffin is a full-duplex video-to-video model that can hold real-time face-to-face conversations. It sees, hears, interprets, speaks, moves and reacts continuously to what it sees and hears. The system can adapt words, timing, tone and expressions to suit the context of a conversation just like a human would.
Benchmarking
Griffin scored high on NVIDIA's VideoDB benchmark
Griffin has already shown its capabilities in independent evaluations of face-to-face conversational AI.
On NVIDIA's VideoFDB benchmark, it scored 3.83 out of 5 on generation, just shy of the human reference score of 3.92 and way ahead of the next-highest published system at a score of 2.80.
In another live study, nearly half (48%) of participants thought they were talking to a real person instead of an AI avatar.
Twitter Post
First model to pass the video Turing test
Introducing Griffin, the first model to pass the video Turing test.
— Tavus (@tavus) October 1, 2026
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM). pic.twitter.com/eb11XbC8Xr
Development
Tavus's years of research into human interactions
The development of Griffin was made possible by years of research into different elements of human interaction.
The company's research team claims it has created industry-leading models in perception, conversational timing, and expressive video.
These technologies are already being used by over 150,000 developers and businesses.
Phoenix-4.5 is Tavus's real-time rendering model that creates subtle movements and active-listening behaviors to make face-to-face interactions feel natural.
Innovation
How Griffin works and its potential applications
Griffin takes a different approach by being a single human interaction model focused on the interaction itself. It combines perception, conversational decision-making, speech, and video generation in real time.
This lets the model react both while speaking and listening.
The system can be used in various scenarios such as tutoring students or rehearsing difficult conversations with someone who reacts dynamically.
Future prospects
Griffin-Lite is available for select testers
Tavus is initially offering Griffin-Lite as a research preview for select trusted testers. The company is working on additional disclosure and safety mechanisms before making it widely available.
This is because the same capabilities that make human interaction models powerful interfaces can also make AI more difficult to distinguish from a person.