Tavus releases Griffin: 54-person video chat for 1 minute, 26 people thought the other side was real.

Beating AI News Flash: AI video company Tavus has released Griffin, focused on real-time face-to-face communication. Tavus calls it the first Human Interaction Model. Griffin can see and listen while speaking, reacting in real time based on the other person's expressions, movements, and pauses. If the user interrupts mid-sentence, it can stop immediately.

Tavus recruited 54 participants from the US and Europe, having them video chat with Griffin-Lite for 1 minute without telling them in advance that the other side was an AI. As a result, 26 people, or 48%, believed they had just been talking to a real person. By comparison, Tavus's previous generation Phoenix-4.5 in similar tests had only 1 out of 41, or 2.4%, misjudge it.

Tavus therefore calls Griffin the first model to pass the "real-time video Turing test." However, this test was organized by Tavus itself, with a sample of only 54 people, and there is currently no larger-scale independent replication.

Griffin-Lite also ranked first in NVIDIA's VideoFDB full-duplex audio-video evaluation. Tavus reported a score of 3.83/5, close to the human reference of 3.92. The previous highest public baseline, Gemini 2.5 + Anam, was 2.80.

Griffin-Lite is currently only available to a small number of testers and has not yet been officially provided to ordinary customers.