TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Researchers from QuEra Computing, Harvard University and Anthropic used Claude with an automated circuit evaluator to develop lower-resource designs for quantum computing. Their preprint reports reductions in both gate counts and qubit requirements for seven of eight building blocks; the results are estimates, not demonstrations on quantum hardware.
Researchers from QuEra Computing, Harvard University and Anthropic used Anthropic’s Claude model and an automated evaluator to design quantum circuits with lower estimated resource requirements, according to a preprint highlighted by QuEra on November 10, 2026. Across eight computational building blocks, the team reports that seven designs reduced both gate counts and qubit requirements; the work is a set of circuit resource estimates, not a demonstration on quantum hardware.
The preprint, titled Discovering state-of-the-art quantum circuits with large language models (arXiv:2610.22367), examines eight components used in larger quantum algorithms. Claude proposed circuit designs, while a separate automated evaluator checked their correctness and scored them by Toffoli-gate count and logical-qubit count. The model then used that feedback to search for revised designs. The researchers report improvements in both measures for seven components. For the eighth, they found an alternative that creates additional trade-offs between gates and qubits.
Among the reported examples, a Reed-Solomon decoder design fell from about 5.1 million Toffoli gates to approximately 440,000, a 12-fold reduction, while also using fewer qubits. For a Coulomb-potential calculation, the gate count dropped from 6,355 to 1,140 and the qubit count from 252 to 142. The paper reports a 9.9-fold reduction in a combined resource metric for that calculation. A fixed-point Gaussian-function design had a reported 39-fold reduction in the same type of combined metric, with gates decreasing from about 24,000 to 1,583 and qubits from 362 to 139.
The combined metric multiplies Toffoli-gate count by peak logical-qubit count. It is a measure of resource use, not a direct measure of runtime: the reported 39-fold change does not mean a 39-fold speedup. The team also describes a new reversible method for modular inversion, an operation used in elliptic-curve calculations. In a point-addition benchmark, one version used about 17 times fewer gates than a prior design with the lowest qubit count, at the cost of using about 12% more qubits.
Lowering the Cost of Quantum Algorithms
Quantum computers have finite hardware capacity, and fault-tolerant machines are expected to need resources for both the computation itself and the correction of errors. Reducing logical qubits and costly operations in important circuit components could allow a future machine to run larger computations, or make a given computation more practical on a machine with limited capacity.
The reported gains matter most if they carry through from individual building blocks to complete applications. The Reed-Solomon decoder, for example, is described as the most resource-intensive step in decoded quantum interferometry, an optimization approach. The chemistry and numerical-computation examples point to other possible uses. But the study does not establish that these reductions translate directly into lower total costs, shorter execution times or useful results on existing machines. Those outcomes depend on how components are integrated and on the quantum computer’s architecture.
The work also offers an example of AI being used as a design-search tool rather than as part of the quantum computer’s operation. Claude generated candidate circuits, and a separate evaluator tested and scored them. Correctness checks were part of the loop, an important distinction from treating model-generated circuit proposals as verified results on their own.
Popular right now
Dyson V15 Detect Cordless Vacuum
Dyson Ball Animal 3 Upright Vacuum
De'Longhi Rivelia Automatic Espresso Machine with Grinder and Milk Frother, Beige
As an affiliate, we earn on qualifying purchases.
How the Circuit Search Worked
Quantum circuits are assembled from operations, or gates, acting on qubits. The paper tracks Toffoli gates, which can contribute substantially to the overhead of fault-tolerant computations, and logical qubits, which represent usable information protected against errors. These measures capture different resource demands: a design with fewer gates may still use more qubits, so reducing both is not always possible.
That trade-off is central to the modular-inversion result. The researchers say the new reversible approach reconstructs much of the information that other methods store during intermediate steps, creating alternative ways to balance gate use against qubit use. The work was led by Andi Gu through Anthropic’s STEM Fellows Program, with Alek Kemeny of Anthropic and Susanne Yelin and Mikhail Lukin of Harvard; Lukin is also affiliated with QuEra. The findings are presented in a preprint, so they should be read as research reported by its authors rather than as a peer-reviewed result.
The project follows an earlier QuEra-Anthropic pilot involving laser control. In that work, the source material says Claude developed and validated conventional control software that recovered the correct laser lock in 695 of 700 trials. The controller then ran without the AI. That earlier effort concerned operating quantum hardware; the new preprint focuses on designing circuits and reducing their estimated resource requirements.
““What is exciting here is that AI, guided by a rigorous evaluator, found new circuit structures that work more efficiently within that constraint.””
— Mikhail Lukin, QuEra co-founder and chief science officer and Harvard professor, as quoted in QuEra’s release
From Estimates to Hardware Results
The reported figures are resource estimates for individual circuit components, not measurements from running the circuits on a quantum processor. The source material does not establish how much the optimized designs would reduce resource use across complete applications, or how their gate and qubit counts would translate into runtime and physical hardware needs on different architectures.
The results are described in a preprint, and the supplied material does not report peer review or independent replication. It also does not provide enough detail to compare the search process with all other circuit-optimization methods, or to determine how broadly the reported improvements generalize beyond the eight tested building blocks. The 9.9-fold and 39-fold figures refer to the specified combined metric, not speedups.
Testing Designs in Larger Algorithms
The next test is whether these designs can be incorporated into full quantum algorithms and retain their resource advantages under the constraints of a specific fault-tolerant architecture. Further evaluation would need to account for such factors as error correction, circuit execution time and the costs of connecting components—not just the gate and logical-qubit measures reported here.
QuEra said the organizations would continue collaborating, but the supplied release does not set out a schedule for new results or hardware demonstrations. Readers should watch for peer-reviewed publication, independent assessments and follow-up benchmarks that show how the circuit designs perform as parts of larger computations.
Key Questions
What did the researchers use AI to do?
They used Anthropic’s Claude model to propose quantum circuit designs. An automated evaluator checked proposed circuits for correctness and scored their Toffoli-gate and qubit requirements, providing feedback for further design searches.
Did the team run the improved circuits on a quantum computer?
The reported results are resource estimates for circuit components. The source material does not describe a hardware run of those circuits or a demonstration of complete applications using them.
Does the reported 39-fold reduction mean the circuit is 39 times faster?
No. The 39-fold figure refers to a combined resource metric that multiplies Toffoli-gate count by peak logical-qubit count. It is not a reported runtime or speedup.
What remains to be shown?
Researchers would need to establish how the component-level savings carry over to full algorithms and how the designs perform under the constraints of particular quantum hardware. The preprint’s findings also await further scrutiny, including peer review or independent replication as reported in the supplied material.
Source: thequantuminsider.com
