The Griffin model allows users to engage in face-to-face conversations with a synthetic agent that can see and hear simultaneously. This tool, which Tavus calls the first "Human Interaction Model," adjusts its tone, facial expressions, and movements in response to interruptions and live cues. In a study by the company, 48% of participants believed they were speaking with a real person, a sharp increase over the less than 3% pass rate of previous systems.

Griffin scored 3.83 out of 5 on Nvidia's VideoFDB benchmark, compared to 3.92 for humans and 2.80 for the next best published system. Tavus is limiting access to a "Griffin-Lite" version for trusted testers while it implements additional safety and disclosure measures. The company intends the tool for professional practice, including sales calls, podcast preparation, and mock job interviews.