AI BriefingProject Fetch: Phase Two
Key point
Anthropic's Claude Opus 4.7 controlled a robot without human assistance, working 20 times faster than existing human teams.
Details
Anthropic announced the results of Phase Two of Project Fetch, which tests LLMs' ability to control the physical world using a robodog.
According to the experiment results, Claude Opus 4.7 performed complex tasks—including connecting to the robot's sensors, writing programs, monitoring routes, and detecting and retrieving objects—without human intervention. In particular, Opus 4.7 demonstrated overwhelming efficiency by completing tasks at least 10 times to as much as over 37 times faster than human teams across all tasks.
Although limitations remain in low-level robot control such as precise object manipulation (fetching), this presents an important indicator that AI models are evolving from 'tools that assist humans' into 'agents that carry out tasks on their own.'
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.