AI agents enter the physical world: Claude Opus 4.7 controls a robot dog 20 times faster than humans
We are witnessing a landmark milestone in the evolution of artificial intelligence. Anthropic has unveiled the results of the second phase of its Project Fetch experiment, and they are impressive. The flagship model Claude Opus 4.5 demonstrated the ability to configure and control a four-legged robot (robotic dog) 20 times faster than teams of human engineers did last year.
This is not just about speed, but about a qualitative leap in autonomy. While in the first phase of the experiment (August 2024), the AI acted as an assistant, helping inexperienced employees find solutions faster, Claude Opus 4.7 now worked almost entirely independently. Under minimal researcher supervision, the neural network completed a full cycle of tasks:
- Connected to the robot's video sensors and LiDAR.
- Wrote a program for manual control.
- Created a trajectory monitoring system.
- Configured an algorithm for recognizing target objects.
The performance of the new model is impressive not only compared to humans. Compared to teams using previous versions of AI, Opus 4.7 proved to be 18 times faster. And compared to humans working without chatbot assistance, it was 37 times faster. Moreover, the code generated by the neural network turned out to be 10 times more compact and efficient than human-written code.
Key takeaway for us as analysts: this progress is not the result of specialized algorithms for robotics. Anthropic emphasizes that this is a "side effect" of the general scaling of language models. This means that any improvements in context understanding and code generation are automatically extrapolated to physical tasks.
However, let us not rush to conclusions about the complete replacement of humans. The experiment also revealed a fundamental limitation: Claude is not yet capable of precise physical manipulation in real time. The AI-controlled robot reached its goal but could not gently push a ball to the exact target point. This task requires complex sensorimotor feedback, where humans still have an advantage.
My expert opinion: We are entering the era of "physical AI agents," as Anthropic rightly notes. However, the current gap between AI's cognitive and motor skills shows that the next major breakthrough will be related not to text, but to tactile sensations and fine motor skills. Investors and developers who are currently focused on this area will gain the greatest advantage in the next 2-3 years.
I also recall that against the backdrop of these successes, on June 13, Anthropic was forced to suspend access to its Fable 5 and Mythos 5 models due to new export control directives from the U.S. government. This once again underscores that AI development is becoming a matter not only of technology but also of global security.