AI-designed circuits cut resource estimates for key quantum building blocks, QuEra-backed study says
A preprint from Harvard, Anthropic and QuEra researchers reports that circuits found with an LLM-driven search used fewer Toffoli gates and qubits in seven of eight subroutines.
QuEra Computing on November 10 highlighted a research preprint with Harvard University and Anthropic reporting that a large language model, guided by an automated checker, found quantum circuits that need fewer resources than standard designs. The preprint is titled “Discovering state-of-the-art quantum circuits with large language models” (arXiv:2610.22367).
What the study did
According to QuEra’s release, the team paired Anthropic’s Claude with an independent evaluator that verified each proposed circuit and scored it on two numbers: Toffoli-gate count and qubit count. The model used that feedback to refine its designs. The study looked at eight building blocks used inside larger quantum algorithms. In seven, the AI-designed circuits reduced both gate counts and qubit requirements against the baseline designs; in the eighth, the search found a new trade-off between the two.
The work was led by Andi Gu as part of Anthropic’s STEM Fellows Program, with Alek Kemeny of Anthropic, Susanne Yelin of Harvard and Mikhail Lukin of Harvard and QuEra, the release says.
The reported numbers
QuEra’s release cites three examples. For the Reed-Solomon decoder used in decoded quantum interferometry, Toffoli gates fell from 5.1 million to about 440,000, with fewer qubits. For the Coulomb potential used in molecular simulation, Toffoli gates fell from 6,355 to 1,140 and qubits from 252 to 142. For a fixed-point Gaussian function, Toffoli gates fell from about 24,000 to 1,583 and qubits from 362 to 139, a 39x reduction in the combined metric.
Important caveats
The release itself stresses that the combined metric multiplies Toffoli count by peak logical qubits, so a 39x improvement is not a 39x speedup. The results are resource estimates for individual components, not runs on quantum hardware, and their effect on complete applications depends on how the components are used. The paper is a preprint and has not yet been peer reviewed.
Sources
Facts on this page come from the sources below, retrieved by DesignQuantum™.