The AI model Claude Opus 4.7 has outperformed engineers in controlling a robot dog — with speeds 20 times faster.
Anthropic has unveiled the results of the second phase of the Project Fetch experiment, and they are impressive. The new Claude Opus 4.5 model demonstrated the ability to autonomously configure and control a four-legged robot, completing tasks 20 times faster than a team of experienced human engineers working with previous AI versions.
To recall, in August 2024, Anthropic employees with no robotics experience attempted to program a robot dog using AI. At that time, the model only acted as an assistant. In the new testing phase, Claude Opus 4.5 operated almost entirely autonomously, with minimal researcher oversight.
The neural network independently performed a number of critical operations:
- Connected to video sensors and LiDAR;
- Wrote a manual control program;
- Created a system for monitoring the robot's movement trajectory;
- Configured an object recognition algorithm.
Key performance metrics are impressive: Opus 4.5 proved to be 18 times faster than a team using older AI versions, and 37 times faster than humans working without chatbot assistance. Moreover, the code generated by the neural network turned out to be 10 times more compact than that produced by human teams, indicating significantly higher efficiency.
An important point: Anthropic emphasizes that the progress in robotics is a "side effect" of the general scaling of language models, rather than the result of implementing specialized algorithms for controlling hardware. This confirms the hypothesis that universal AI models can adapt to physical tasks without special fine-tuning.
However, Claude still struggles with precise physical actions. The model successfully guided the robot to its target but failed to gently push a ball to a specific point. Such manipulations require complex real-time feedback, where humans still maintain an advantage.
Anthropic believes we are entering the era of "physical AI agents." In the near future, neural networks will be able to work with standard tools and equipment as effectively as they currently do with software code.
My analysis: The results of Project Fetch are not just another performance record. They signal a fundamental shift: AI models are beginning to transition from the virtual space into the physical world, and this transition is happening faster than expected. However, tasks requiring fine motor coordination and real-time feedback remain the "bottleneck." This is where competition among developers will be concentrated in the coming years.