Breaking

Tavus Introduces Griffin, a Real-Time AI Video Model It Says Passed a Face-to-Face Turing Test

October 3, 2026 at 16:42 EDT

Foundation Models
AI Agents
Media Generation

Tavus has introduced Griffin, a real-time video-to-video AI model that the company says can listen, speak, gesture and react during a live face-to-face conversation. In a company-run one-minute study, 26 of 54 participants, or 48%, said they believed they had spoken with a real person.

Oct. 1, 2026 · Tavus

Griffin Tests the Human Line

Tavus says its real-time video model can listen, speak, gesture and react during live calls, with 48% of study participants believing they had spoken to a real person.

48%

said Griffin-Lite was a real person

participants in one-minute calls

3.83

VideoFDB generation score out of 5

A 20x jump over Tavus's prior system

Each stacked block represents 2.4 percentage points in the face-to-face study result.

2.4%

Previous system

48%

Griffin-Lite

What changes in the loop

Sees and hears

Keeps listening

Times a response

Generates face and voice

Unlike turn-based pipelines, Griffin is designed to react during the other person's turn.

Why builders care

Tutoring, coaching, support and training could benefit when an AI can notice hesitation, visual context and interruptions.

Why caution remains

The study was short and company-run, and Tavus says safety and disclosure work must come before customer availability.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.