Tavus Unveils Griffin and Opens Limited Preview

Tavus offers restricted access to Griffin-Lite, a video avatar designed for real-time conversation. A more advanced version is planned after addressing security concerns. According to the company, 48% of tested individuals mistook the avatar for a human after a one-minute call, and Griffin scored 3.83 points in a test by Nvidia, where humans achieved 3.92.

Funding, Origin, and Restricted Access to Griffin-Lite

Tavus, an AI startup founded in 2020 in San Francisco, has raised approximately $64 million. The company initially developed personalized AI videos for sales and marketing, then shifted to live video conversations with digital personas. Griffin-Lite, a preliminary version of its video avatar, is accessible to selected testers as part of a research preview. A more capable version will be offered once security concerns are resolved.

Reported Results: 48% Confusion and 3.83 Points in the Nvidia Test

According to a study conducted by Tavus, after a one-minute video call, 48% of participants believed Griffin was a real person, while previous systems did not exceed 2%. Tavus also cites an independent test conducted by Nvidia to measure the perception of human-like characteristics during a direct audio-video interaction: Griffin achieved a score of 3.83 points. In the same test, human participants scored 3.92 points, and the best previous AI model received 2.80 points.

HIM: A Model for Face-to-Face Dialogue and Its Use Cases

According to Tavus, Griffin represents the first "Human Interaction Model," a new category of model designed to interpret and conduct real-time, face-to-face exchanges. This system analyzes speech, facial expressions, intonation, gestures, and pauses while processing and producing video. Among the use cases envisioned by Tavus are tutoring, training for complex conversations, and camera-based technical assistance. The company provides additional details.