The AI model Claude Opus 4.7 taught a robot dog new tricks 20 times faster than humans.
Experiments in integrating artificial intelligence with physical robots are reaching a fundamentally new level. Anthropic has presented updated results from its Project Fetch, and the numbers are impressive: the Claude Opus 4.7 model completed tasks for configuring and controlling a four-legged robot 20 times faster than teams of human engineers working with last year's AI versions.
In 2024, the neural network acted only as an assistant, helping employees without robotics experience find solutions faster. Now, the situation has changed dramatically. In the new testing phase, Claude Opus 4.7 worked almost autonomously. Under minimal researcher supervision, the neural network completed a full cycle of tasks:
- connected to video sensors and LiDAR;
- wrote a program for manual control;
- created a robot path monitoring system;
- configured an object recognition algorithm.
A comparative analysis shows a massive performance gap. The Opus 4.7 model proved to be 18 times faster than a team using older AI versions and 37 times faster than people working without chatbot assistance. Moreover, the code generated by the neural network was 10 times more compact and efficient than human-written equivalents.
Physical Limitations Remain
The experiment's authors emphasize an important nuance: progress in robotics has become a side effect of general language model scaling. Anthropic did not implement specialized algorithms for controlling hardware — the model itself found ways to interact with the physical environment. However, despite successes in navigation and programming, Claude still struggles with precise physical actions. It managed to guide the robot to a target, but the model could not gently push a ball to a specific point. Such tasks require complex real-time feedback, where humans still outperform AI.
Anthropic states that the industry is entering an era of "physical AI agents." In their view, future neural networks will be able to use standard tools and equipment as effectively as they currently work with software code.
Expert opinion: The results of Project Fetch are not just another benchmark. We are witnessing AI's transition from a purely digital space into the physical world. The fact that the model independently learned to control a robot without specialized training indicates the emergence of true adaptive logic in neural networks. However, the problem with precise motor skills is a "bottleneck" that separates us from fully functional robot assistants. Until AI learns to sense the physics of interacting with objects, it is too early to talk about the widespread adoption of such systems.