Anthropic has updated its Project Fetch experiment. The Claude Opus 4.7 model completed tasks related to configuring and managing a robodog 20 times faster than human engineering teams.
New Frontier Red Team blog: Phase 2 of Project Fetch, where we test how well Claude can program a robodog.
Opus 4.7, on its own, was ~20x faster than last year’s best human team aided by Opus 4.1. (The robodog, alas, still failed to fetch a beach ball.)https://t.co/CgbBtRf85e
— Anthropic (@AnthropicAI) June 18, 2026
In August 2024, company employees without robotics experience attempted to program a quadruped robot. At that time, AI only assisted humans in finding solutions faster. In the new testing phase, the Claude Opus 4.7 model operated almost autonomously under minimal researcher supervision. The neural network independently:
- connected to video and lidar sensors;
- wrote a program for manual control;
- created a robot path monitoring system;
- configured an object recognition algorithm.
The Opus 4.7 model was 18 times faster than teams using older AI versions and 37 times faster than humans without chatbot assistance. The neural network wrote more efficient code, which was 10 times smaller than that of human teams.
The authors noted that progress in robotics has become a byproduct of the general scaling of language models. Anthropic did not implement specialized algorithms for hardware control.
Despite the success, Claude still faces challenges with precise physical actions. The model managed to guide the robot to the target but failed to gently nudge a ball to the desired point. This requires complex real-time feedback, where humans currently outperform AI.
Anthropic believes the industry is entering an era of “physical AI agents.” In the future, neural networks will be able to use standard tools and equipment as effectively as they currently handle software code.
Earlier, on June 13, Anthropic halted access to the Fable 5 and Mythos 5 models due to a U.S. government directive under export control regulations.
