The starting point
Tavus sells to developers and to the companies that put AI video in front of customers. On X it has a big launch audience. On LinkedIn, where buyers and partners sit, a typical Tavus post drew 74 reactions and 45 comments.
Tavus builds AI video models, including Griffin, the first model it says has passed a video Turing test. When Tavus posted the Griffin announcement on LinkedIn on October the 2nd, our creators moved inside the first hour. Within 24 hours the post had 1,170 reactions, 364 comments and 188 reposts: nearly one comment for every three reactions, and one repost for every six. Two weeks on it has 2,667 reactions, 584 comments and 344 reposts, and it is still climbing.
Tavus sells to developers and to the companies that put AI video in front of customers. On X it has a big launch audience. On LinkedIn, where buyers and partners sit, a typical Tavus post drew 74 reactions and 45 comments.
Today we're introducing Griffin, the first model to pass the real-time video Turing test.
On live video, 48% of people who talked to it thought it was a real person. Previous systems, including our own, had less than a 3% pass rate. It is also #1 on the NVIDIA benchmark for realtime full-duplex AI video.
It's the first Human Interaction Model (HIM): a new class of video-to-video, unified full-duplex models that listen, watch, talk, gesture and react at the same time, like a person.
At Tavus, we want computing to become invisible, to feel as natural as talking to a friend or coworker. You shouldn’t have to turn a messy thought into a perfect prompt, or explain “this” when you’re pointing right at it. A tutor with their eyes closed misses the confusion on your face. And an apology means very little when it comes with a smile. These details are the difference between feeling understood and having to manage the machine.
The brief. Reach the people who build with AI video and the people who buy it: developers, product leads and founders, plus the AI and tech voices they follow. Comments had to come from people with a real take on AI video.
The creators. 85 creators, every one a real person with a real profile, chosen for audience fit rather than follower count. Between them, 25 million followers. For Tavus the network leaned on AI builder and founder voices.
The window. Everything landed inside 24 hours. Reactions and the first comments in the first hour, when LinkedIn decides whether to show a post more widely. Comments with a point of view through the day. Reposts spread out, so the post kept re-entering feeds.
What each creator did. Reacted, wrote a comment in their own words from the brief, and reposted. No two comments were the same.
What we did not do. No bots, no bought engagement, no paid boosting, no copy-and-paste comments. Every account is a real person, so the engagement stays. Platforms remove fake engagement within days. Ours is still there at day 14.
Here is how engagement on this post moved after our delivery window closed.
Based on 1,497 additional reactions and 220 additional comments between the 24-hour reading and 20 October.
By 20 October, 18 days after publication, the post had reached 2,667 reactions, 584 comments and 344 reposts. Since the first 24-hour reading, it had added 1,497 reactions and 220 comments, showing that engagement continued well beyond the launch window.