Tavus Griffin and Griffin-Lite Human Interaction Model
Tavus introduced Griffin on 1 October 2026, which it describes as the world's first Human Interaction Model (HIM), a full-duplex video-to-video model that understands and generates face-to-face real-time human interaction.
Tool Overview
Features, steps and FAQ below
Features
- ✓ Tavus says Griffin is the first model to pass the real-time video Turing test: in its study, 48% of participants who had a one-minute video call thought they had talked with a real human, while previous systems (including Tavus's own conversational video interface) had a maximum pass rate of about 2%
- ✓ Tavus says Griffin is a full-duplex video-to-video system that listens while it talks and pays attention to expressions and pauses, not just words; it combines a Continuous Conversational Modeling engine with an Audio-Visual Generation engine (Streaming Speech Generation plus Streaming Video Generation) running concurrently
- ✓ Tavus says Griffin generates every pixel in every frame in real time from one reference image, controlling not just the face, arms and fingers but also things like the movement of the chair, the shadows cast and the background
- ✓ Tavus says Griffin ranks number one on NVIDIA's independent VideoFDB benchmark for face-to-face AI and is 37% ahead of the next best AI at reacting in the moment
- ✓ Tavus says Griffin-Lite is a research preview available only to select trusted testers, with a wider release of a more powerful model to follow; Tavus stresses Griffin is not available to customers yet
How to Use
- Visit the official Tavus Griffin page (tavus.io/griffin) to learn about the model's capabilities, technical approach and safety notes
- Tavus says Griffin-Lite is a research preview for select trusted testers; request access via the form on the page
- While you wait, developers can build face-to-face AI experiences through the Tavus API using its existing PAL models (Phoenix for the face, Raven for visual understanding and Sparrow for knowing when to talk)
- Watch for Tavus updates on safe disclosure features and the wider release so you can integrate when the full model opens up
FAQ
What is Griffin?
Tavus's first Human Interaction Model (HIM), introduced on 1 October 2026, which Tavus describes as a new class of model designed to understand and generate face-to-face real-time human interaction using a full-duplex video-to-video system.
Does Griffin pass the video Turing test?
Tavus says it is the first model to pass the real-time video Turing test: in its study, 48% of participants believed they were talking to a real human, versus a maximum pass rate of about 2% for previous systems.
Can I use Griffin today?
Tavus says Griffin-Lite is a research preview available only to select trusted testers and not available to customers, with a wider release of a more powerful model to follow after safety measures are addressed.
What is the difference between Griffin and Griffin-Lite?
Tavus says Griffin-Lite is a preview of its first Human Interaction Model, with a more powerful, full Griffin model to follow once safety evaluation is complete.
What models did Tavus have before?
Tavus says its earlier conversational video interface was powered by Phoenix 4.5 (the face), Sparrow-2 (knowing when to talk) and Raven-1 (visual understanding), whose research Griffin unifies into one full-duplex system.