The AI model Claude outperformed humans by 20 times in controlling a robot dog: a breakthrough in physical robotics
The world of robotics has witnessed a landmark event: a new version of Anthropic's language model, Claude Opus 4.7, has demonstrated the ability to configure and control a four-legged robot 20 times faster than teams of human engineers. This is not just another benchmark—it is a real step toward an era where AI takes on not only digital but also physical tasks.
In August 2024, Anthropic employees with no robotics experience attempted to program a robot dog using AI. At that time, the model acted as an assistant, accelerating the search for solutions. Today's experiment, known as Project Fetch, Phase 2, has radically changed the rules of the game. The Opus 4.7 model worked almost autonomously, under minimal researcher supervision.
What did the AI do independently?
The neural network completed a full cycle of integration tasks: it connected to video sensors and LiDAR, wrote a program for manual control, created a robot route monitoring system, and configured an object recognition algorithm. All this work was completed tens of times faster than by human teams.
The comparative analysis is impressive: Opus 4.7 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 generated code was 10 times more compact and efficient than human-written code. This suggests that the model does not simply copy operator actions but finds fundamentally more optimal solutions.
Physical limit: the ball proved elusive
Despite the triumph in programming, Claude still faces challenges with precise physical manipulations. The model successfully guided the robot to the target but could not gently push a beach ball to the exact spot. This is a complex task requiring real-time feedback and fine motor skills—areas where humans still hold an advantage. As I have repeatedly noted in my analyses, "hardware" and physics remain the last bastion of human intelligence in the AI era.
It is important to emphasize that Anthropic did not implement specialized algorithms for robot control. The progress in robotics was a side effect of the general scaling of language models. This confirms my long-standing hypothesis: universal models trained on vast datasets will inevitably begin to outperform specialized systems in related fields.
The company believes the industry is entering an era of "physical AI agents." In the near future, neural networks will be able to use standard tools and equipment as effectively as they currently work with software code. Recall that on June 13, Anthropic was forced to halt access to the Fable 5 and Mythos 5 models due to a directive from the U.S. government on export controls. This shows how seriously regulators perceive the potential of such technologies.
My analysis: The breakthrough with the robot dog is not just a demonstration of speed. It is a signal to the market that investments in foundational LLM models are paying off in the most unexpected sectors. I expect that within 12-18 months, we will see a wave of startups attempting to apply similar approaches to industrial automation and logistics. However, physical precision will remain the "bottleneck"—and this is where the greatest investment value lies for companies developing tactile sensors and feedback systems.