Anthropic’s Claude AI Revolutionizes Quantum Circuits with 17x Reduction in Gates

In a monumental advancement for artificial intelligence and quantum computing, Anthropic’s Claude Mythos 5 has independently reengineered eight quantum circuits. It outperformed past state-of-the-art designs on seven out of the eight benchmark tasks, using up to 17 times fewer gates than previous methods.

The findings, disclosed on October 16, 2026, were the outcome of a 53-hour experiment spearheaded by researcher Andi Gu. This experiment was part of Anthropic’s STEM Fellows Program and included collaborators from Harvard University and quantum computing firm QuEra Computing. Claude operated in a completely closed loop, creating each circuit as Python code, receiving automated scoring, and refining designs without any human intervention between iterations.

The most remarkable result involved a modular inversion architecture for elliptic-curve point addition. Claude’s design used 17x fewer gates, albeit with 12% more qubits than the smallest previous design. The AI also decreased gate counts for a decoded quantum interferometry decoder by 12 times compared to Google’s prior construction. The circuits cover domains such as chemistry, optimization, quantum simulation, number theory, and numerical computing.

Anthropic suggests that these findings could bring practical quantum computers significantly closer by reducing the resource demands of complex quantum algorithms. The paper has been submitted to arXiv (2610.22367).

Source: Interesting Engineering – Claude Solves Quantum Memory Bottleneck with 17x Leaner Circuit