Designing more efficient quantum circuits with Claude

Researchers used Claude to design and optimize quantum circuits by giving it plain-language specifications for eight fundamental routines drawn from chemistry, optimization, simulation, and number theory, and pairing it with an automated Python-based evaluator that checks correctness and scores designs by qubits × gates. Claude Mythos 5 ran in a closed loop - proposing implementations, receiving verifier feedback, and iterating with minimal human input (about 53 hours of work) - and produced novel circuit constructions that used fewer qubits and gates than the best previously published designs, in some cases by an order of magnitude. Improvements came from fresh structural and mathematical insights rather than incremental tuning: examples include exploiting molecular symmetries to reuse computation in iron-sulfur simulations and removing costly components intended for extremely rare inputs. Two highlighted successes are a new reversible algorithm for modular inversion (the bottleneck in elliptic-curve point addition) and a redesign that makes the costliest step of a leading quantum optimization subroutine roughly 12× cheaper.

The project stresses why quantum circuit design is uniquely hard - no classical if-then branching and the need to uncompute scratch space make reversibility and space-time trade-offs central - and shows that an AI can discover new ways to balance those trade-offs. By lowering qubit and gate counts on key primitives, AI-driven redesign can substantially reduce resource estimates for practical quantum applications, accelerating timelines for problems in chemistry, materials, optimization, and cryptanalysis.