Tavus has introduced Griffin as what the company calls its first Human Interaction Model, or HIM, built for face-to-face, real-time conversation. The company says Griffin is designed to process live interaction as it happens, using audio and video signals rather than waiting for a person to finish speaking before generating a response.
Tavus says on the same announcement page that Griffin-Lite is a research preview available to a select group of early testers, with a wider release of a more powerful model to follow.
Some of the early attention around Griffin came from more aggressive framing in outside coverage. Digit described the model through a hypothetical scenario where AI could answer job interview or viva questions for a user. In the supplied evidence, though, that is use-case framing around Tavus's announcement, not proof of a documented real-world system replacing people in actual interviews.
How Tavus says Griffin works
On its launch page, Tavus says Griffin combines perception, conversational decision-making, and expressive speech and video generation in one system. The company presents that as an alternative to more turn-based systems that transcribe speech, generate a reply, then turn that reply into voice and video after the user stops talking.
Digit's report, citing Tavus, says Griffin continuously processes what is happening during a conversation so it can react to interruptions, changes in tone, and other visible or audible cues while the exchange is still going on. Tavus similarly says on its own page that perception, deciding when and how to respond, and expressive speech and video generation happen at the same time.
That is also where the practical use cases in the supplied evidence come from. Digit says Tavus believes Griffin could be useful in practice interviews, tutoring, and customer support. In the evidence provided, those are proposed applications, not confirmed deployments.
Tavus's study claim
On its announcement page, Tavus says that in a live study, 48% of participants who spent one minute talking with Griffin thought they had spoken to a real human. Tavus compares that with what it says was a maximum 2% pass rate for its previous systems.
A Progressive Robot write-up summarizing Tavus's materials puts that figure at 26 of 54 participants for Griffin-Lite, versus 1 of 41 for Tavus's previous system. In the supplied evidence, that claim comes from Tavus's own reported one-minute study, as summarized by Tavus and outside coverage drawing on Tavus's materials.
For now, the clearest supported development is narrower than the most attention-grabbing framing: Tavus has unveiled Griffin as what it calls its first Human Interaction Model, and the version currently available is Griffin-Lite in a limited research preview for select testers only.
