The Signal
AI circuits cut a resource metric up to 39-fold, QuEra says
QuEra says AI-designed circuits reduced a combined gate-and-qubit resource measure by up to 39-fold across computational building blocks. QuEra highlighted the research with Harvard and Anthropic, while noting the results are estimates rather than hardware runs.

AI searches for leaner circuits
AI-designed quantum circuits cut a combined gate-and-qubit resource measure by as much as 39-fold in research QuEra highlighted with Harvard University and Anthropic on Nov. 10. The preprint, titled “Discovering state-of-the-art quantum circuits with large language models,” examines eight computational building blocks used in larger quantum algorithms.
In seven of the eight, the AI-designed circuits reduced both gate counts and qubit requirements compared with baseline designs, QuEra reports. In the eighth, QuEra reports that the search found a new approach that offers different choices between the two resource requirements.
The researchers paired Anthropic’s Claude with an independent automated evaluator. The evaluator checked proposed circuits for correctness and scored their resource use; Claude used that feedback to explore and improve its designs. The two measures were Toffoli-gate count and qubit count.
Reported reductions and a new subroutine
For the Reed-Solomon decoder, which QuEra describes as the most expensive step in decoded quantum interferometry, QuEra says the AI design used 12 times fewer Toffoli gates: 5.1 million compared with about 440,000. The AI design also used fewer qubits. QuEra describes decoded quantum interferometry as a promising quantum optimization algorithm.
For a Coulomb-potential calculation used in molecular simulation, QuEra reports a 9.9-fold reduction in the combined measure. Toffoli gates fell from 6,355 to 1,140, while the qubit count went from 252 to 142. For a fixed-point Gaussian function, QuEra reports the largest reduction: 39-fold, with Toffoli gates dropping from about 24,000 to 1,583 and qubits from 362 to 139.
The researchers also report that Claude helped discover a new reversible approach to modular inversion, an arithmetic operation used in elliptic-curve calculations. In a complete point-addition benchmark, one design required about 17 times fewer gates than the prior design using the fewest qubits, while using about 12% more qubits, the researchers report. QuEra says the approach reconstructs much of the information normally stored during intermediate steps, offering new ways to balance gates against qubits.
A 53-hour search, guided by researcher-defined requirements and an automated evaluator, produced the new elliptic-curve circuit approach, QuEra says.
Resource estimates, not hardware runs
The combined measure, which QuEra defines as multiplying the Toffoli-gate count by the peak number of logical qubits, is not a 39-fold speedup for a 39-fold improvement, QuEra cautions. The results are estimates for individual building blocks, not runs on hardware, and their effect on complete applications depends on how the components are used. Runtime and physical hardware requirements also depend on a system’s architecture, QuEra says.
The release links the research to an earlier QuEra-Anthropic pilot on laser control. QuEra says Claude-developed control software recovered the correct laser lock in 695 of 700 trials, in seconds rather than the 5 to 10 minutes a human expert needs. The resulting controller is conventional software that runs without AI. The Libra system is targeted for availability on Amazon Braket in 2028.
Source
Companies in this story
Related stories
- Zapata CTO calls for faster quantum use-case screening
- Study reports higher simulated Max-Cut ratios with cat-qubit QAOA
- DLR researchers report shallower circuits for quantum data loading
- Preprint expects no exponential DQI speedup with local decoders
- Mele and Bittel match channel tomography’s dimension scaling