Claude Cuts Quantum Circuit Costs Up to 39-Fold, Study Finds
By @sharedot · · 6 pages
- AI
- Quantum Computing
- Research
AI-designed quantum circuits from Harvard, Anthropic, and QuEra cut combined gate-and-qubit resource costs by up to 39-fold, a preprint reports.
AI Designs Leaner Quantum Circuits
QuEra Computing, Harvard University, and Anthropic announced research showing that AI can design quantum circuits requiring far fewer computational resources, according to PR Newswire.
The preprint, "Discovering state-of-the-art quantum circuits with large language models" (arXiv:2610.22367), examined eight computational building blocks used in larger quantum algorithms.
In seven of the eight, AI-designed circuits reduced both gate counts and qubit requirements versus baseline designs, and in the eighth the search surfaced a new approach offering additional trade-offs between qubits and gates.
The work, an industry-first collaboration between AI and quantum leaders, was led by Andi Gu through Anthropic's STEM Fellows Program, with Alek Kemeny of Anthropic, Susanne Yelin of Harvard, and Mikhail Lukin of Harvard and QuEra.
Why the Gains Are Surprising
The headline result is a 39-fold reduction in a combined resource metric for a fixed-point Gaussian function, with Toffoli gates falling from about 24,000 to 1,583 and qubits from 362 to 139, per the announcement.
Optimization of the Reed-Solomon decoder — the most expensive step in decoded quantum interferometry — cut Toffoli gates 12-fold, from 5.1 million to about 440,000, while also using fewer qubits. A chemistry benchmark for the Coulomb potential saw a 9.9-fold combined reduction.
The metric multiplies Toffoli-gate count by peak logical qubits, so a 39x improvement is not a 39x speedup, but reversibility normally makes many efficient classical methods costly on quantum hardware — and here AI, guided by a rigorous evaluator, found new circuit structures that work efficiently within that constraint, Lukin said.
A New Subroutine and the Verifier Model
Beyond optimizing existing designs, Claude helped discover a new reversible approach to modular inversion, an arithmetic operation central to elliptic-curve calculations, PR Newswire reports.
The approach reconstructs much of the information normally stored during intermediate steps, creating new options for balancing gates against qubits: in the complete point-addition benchmark, one design needed about 17 times fewer gates than the prior fewest-qubit design while using roughly 12% more qubits.
Anthropic's Kemeny credited the independent automated evaluator, which checked every proposed circuit for correctness and scored its Toffoli-gate and qubit counts, letting Claude pursue unconventional ideas that could be trusted if they held up.
A 53-hour search guided by researcher-defined requirements produced the new elliptic-curve approach.
The Stakes for Useful Quantum Computing
The research addresses the algorithm-efficiency curve: useful quantum computing arrives where hardware capability and algorithm resource requirements meet, so progress on either moves that meeting point earlier.
QuEra's Takuya Kitagawa said AI-driven reductions in the resources required for key computations expand what future fault-tolerant systems can deliver, with the company's Libra system targeted for Amazon Braket availability in 2028.
The findings build on QuEra's earlier Anthropic pilot, in which Claude developed laser-relock software that succeeded in 695 of 700 trials, recovering in seconds what takes a human expert 5 to 10 minutes, per PR Newswire.
The results are resource estimates for building blocks rather than hardware runs, and their effect on complete applications will depend on how components are used.
The announcement lands amid a broader wave of AI-assisted scientific discovery — Tom's Guide reports AI is similarly accelerating personalized cancer research, from Moderna and Merck's melanoma vaccine results to DeepMind's AlphaFold protein predictions.