# Harvard, Anthropic, and QuEra Claim AI-Discovered Subroutine Cuts Quantum Computing Resource Needs 9.9x > Source: > Published: 2026-11-11 02:08:31+00:00 # Harvard, Anthropic, and QuEra Claim AI-Discovered Subroutine Cuts Quantum Computing Resource Needs 9.9x Harvard University, Anthropic, and QuEra Computing have released a preprint reporting that a large language model, paired with an automated verifier, designed quantum circuits that reduce estimated resource requirements for several computational building blocks. A highlighted figure—a 9.9x reduction for simulating the Coulomb potential—comes from a combined metric that multiplies Toffoli-gate count by logical qubit count, not from a demonstrated speedup on hardware. The preprint, "Discovering state-of-the-art quantum circuits with large language models" (arXiv:2610.22367), describes a collaboration 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. The researchers used Anthropic's Claude to propose circuit designs, which an independent evaluator checked for correctness and scored on two numbers: Toffoli-gate count and peak logical qubit count. Both matter for fault-tolerant quantum computing: Toffoli gates dominate error-correction overhead, and logical qubits consume scarce hardware capacity. Across eight building blocks, the AI-assisted search produced circuits that improved on both metrics in seven cases. For the eighth—modular inversion, a subroutine used in elliptic-curve arithmetic—it found a new reversible approach that reconstructs intermediate information rather than storing it, offering a different trade-off: roughly 17x fewer gates than the prior design with the fewest qubits, at the cost of about 12% more qubits. The headline chemistry result concerns the Coulomb potential, the term that describes charged-particle interactions in molecular simulation. The paper reports Toffoli gates falling from 6,355 to 1,140 and qubits from 252 to 142, a 9.9x reduction in the product of the two. A fixed-point Gaussian function showed a 39x reduction in the same combined metric (gates from ~24,000 to 1,583, qubits from 362 to 139), and a Reed-Solomon decoder used in a quantum optimization algorithm dropped from 5.1 million Toffoli gates to about 440,000 while also using fewer qubits. The authors are careful to note that the combined metric is not a runtime speedup. "These results are resource estimates for individual building blocks rather than runs on hardware," the preprint states, "and their effect on complete applications will depend on how those components are used." The work has not been peer-reviewed. The claims come from organizations with a direct commercial interest. QuEra, which is building neutral-atom quantum computers targeting fault tolerance, has a roadmap that places its Libra system on Amazon Braket in 2028. Anthropic gains a showcase for its model's reasoning capabilities. The press release describes the collaboration as "industry-first" and quotes executives framing the results as accelerating the timeline for useful quantum computing. The research builds on an earlier QuEra–Anthropic pilot in which Claude wrote laser-lock recovery software that succeeded in 695 of 700 trials, replacing a manual process that took experts 5–10 minutes. That controller now runs as conventional code without the AI. The new work shifts from hardware control to algorithm design. The evaluator-verifier loop is the methodological contribution. "Because every design was independently checked and scored, Claude could pursue unconventional ideas and we could trust the ones that held up," said Kemeny. The 53-hour search that produced the modular-inversion subroutine was guided by researcher-defined requirements and the automated scorer. Independent assessment is limited. The preprint provides component-level estimates; it does not show integration into full algorithms, nor does it model how the savings translate across different fault-tolerant architectures. Historical precedent suggests caution: the widely studied FeMoco simulation saw its estimated Toffoli count drop roughly 150,000-fold over eight years through successive human-driven advances. Whether AI can accelerate that trajectory remains an open question, not a demonstrated result. What is established is a reproducible search method that produced verifiable circuit designs for specific subroutines. Whether those designs survive compilation, layout, and error-correction overhead in a complete application—and whether the approach generalizes beyond the eight tested blocks—has not yet been shown. ## Sources 1. [Harvard, Anthropic, and QuEra Breakthrough Research ...](https://www.quera.com/press-releases/harvard-anthropic-and-quera-breakthrough-research-uniquely-shows-ai-can-reduce-the-resources-quantum-applications-need-significantly-accelerating-the-era-of-useful-quantum-computing) (quera.com) 2. [Harvard, Anthropic, and QuEra Breakthrough Research ...](https://www.prnewswire.com/news-releases/harvard-anthropic-and-quera-breakthrough-research-uniquely-shows-ai-can-reduce-the-resources-quantum-applications-need-significantly-accelerating-the-era-of-useful-quantum-computing-302934155.html) (prnewswire.com) 3. [ENIAC - Wikipedia](https://en.wikipedia.org/wiki/ENIAC) (en.wikipedia.org) 4. [BlackRock World Technology Fund | A2](https://www.blackrock.com/sg/en/products/230010/bgf-world-technology-fund-a2-usd) (blackrock.com) 5. [Quantum tunnelling - Wikipedia](https://en.wikipedia.org/wiki/Quantum_tunnelling) (en.wikipedia.org) 6. [QuEra, Harvard And Anthropic Use AI To Reduce Quantum Circuit ...](https://domystats.com/advanced-methods/quera-harvard-and-anthropic-use-ai-to-reduce-quantum-circuit-resource-requiremen) (domystats.com)