PANews, June 20 – Anthropic has released the Phase 2 experimental results of "Project Fetch," evaluating the capability improvements of its latest model in real-world robot manipulation. The experiment, conducted in August 2025, involved non-robotics-expert Anthropic employees using off-the-shelf quadruped robots to complete a series of complex tasks. It compared performance between "assisted by Claude models" and "relying solely on humans and the internet." The results showed that under fully autonomous operation of the latest model, Claude Opus 4.7, its average speed across all completable tasks significantly surpassed that of human teams, with execution speed improved by at least 10x.

Anthropic stated that this progress is not the result of specialized robot training, but rather an extension of general large model capabilities, and noted that AI is gradually moving from "assisting humans in using tools" to the early stage of "directly manipulating physical tools," similar to the path previously seen in the evolution toward agentic programming in software engineering.