Tavus Griffin — Full-Duplex Video-to-Video AI

Announced October 1, 2026, Griffin is a full-duplex video-to-video “Human Interaction Model.” It sees and hears a caller while generating speech and moving video at the same time, so listening, watching and speaking run concurrently rather than as a speech-recognition → language-model → speech-synthesis → avatar relay.

In a vendor-run study—not an independent Turing test—26 of 54 participants (48%) judged their one-minute video-call partner to be a real person, versus 2.4% for Tavus’s previous system.

Griffin-Lite is a research preview for selected trusted testers only. Tavus has published no pricing and no release date, and says wider release will wait on safety and disclosure work.

Availability: Not generally available; current use is prospective. Best for businesses and developers tracking the future of conversational video AI.

Related: Runway · ChatGPT

Pros

• A full-duplex video-to-video model: it sees and hears a caller while generating speech and moving video at the same time.
• Listening, watching and speaking run concurrently, not as a speech-recognition to language-model to speech-synthesis to avatar relay.
• In a vendor-run study, 48% of participants judged a one-minute video-call partner to be a real person, versus 2.4% for Tavus's previous system.
• Points to a future where conversational video AI feels genuinely interactive.

Watch out for

• Not generally available: Griffin-Lite is a research preview for selected trusted testers only, so you cannot sign up and use it today.
• Tavus has published no pricing and no release date, and wider release will wait on safety and disclosure work.
• The realism study was vendor-run, not an independent Turing test — treat the 48% figure as indicative, not proven.

Best for

Businesses and developers tracking the future of conversational video AI. Who should skip it: anyone needing a shippable conversational video product today — this is prospective, not available.

Pricing

No pricing has been published and there is no release date — check the official pricing page and Tavus's announcements for any update.

FAQs

What is Griffin? Announced October 1, 2026, it is a full-duplex video-to-video Human Interaction Model from Tavus.

How is it different from older avatar pipelines? It sees and hears a caller while generating speech and moving video at the same time, rather than chaining recognition, language modelling, synthesis and an avatar in sequence.

Can I use it now? No. Griffin-Lite is a research preview for selected trusted testers only; it is not generally available.

Is there pricing yet? No. Tavus has published no pricing and no release date.