Project Fetch: Phase two
Summary
In the second phase of Project Fetch, Anthropic researchers evaluated how newer AI models like Claude Opus 4.7 perform on robotic tasks compared to human teams. While the previous generation required human guidance, Opus 4.7 successfully completed several tasks autonomously, performing them up to 20 times faster than human participants from the initial study. Although the model still struggles with high-precision physical manipulation, such as accurately fetching a beach ball, its ability to identify optimal paths and generate effective code highlights the rapid evolution of AI agents in physical environments.
(Source:Anthropic)