Tavus ships Griffin-Lite as a research preview

Tavus published Introducing Griffin, describing Griffin as its first Human Interaction Model. The page presents Griffin as a full duplex video to video system that listens while it talks.

Griffin-Lite is the preview build. Tavus says it is available to a select group of early testers, with a wider release of a more powerful model to follow.

The page says Griffin combines perception, conversational decisions and expressive speech and video in one real-time loop.

Tavus and early safety testers

Tavus built and evaluated Griffin-Lite. Hassaan Raza, speaking for the company on the page, frames the goal as machines that meet people in face-to-face conversation.

Access today is limited to trusted testers. The page says Griffin-Lite "will not be available for use for customers at this time" and points to an access request form.

A face-to-face study at 48%

Tavus says it ran a live study through an independent research platform. Participants had a one-minute video call about what they were looking forward to that year. Their partner was a Griffin-Lite PAL, generating face, voice and replies in real time.

Of 54 participants, 26 believed their partner was a real person, which Tavus reports as 48%. The page says those who said real averaged 79% confidence and those who said AI averaged 81%.

On an earlier Phoenix-4.5 stack, Tavus reports 2.4% of participants (n=41) believed the partner was real. It calls Griffin the first model to pass what it labels the real-time video Turing test.

Tavus also reports component scores on VideoFDB and streaming video metrics. The Frontier found no independent replication of the study or those boards.

Self-reported study, closed customer access

The study is Tavus's own design. The page does not publish raw transcripts, a preregistration or an outside auditor's report.

Customer use is withheld. Tavus says further alignment and safety procedures are required, and that it is working on safe disclosure features.

Voice cloning from about 10 seconds of audio is described as a Griffin-Lite capability. That raises misuse risk the company says it is still addressing.

Safety reviewers and realtime video teams

  • Labs building conversational video should treat the 48% figure as a vendor claim until an outside group repeats the protocol.
  • Buyers evaluating Tavus should wait for customer access terms and safety disclosures before planning a production rollout on Griffin-Lite.
  • Trust and safety teams can use the study design as a template for their own face-to-face deception tests.