AI cuts quantum circuit costs by 17x

A team from Harvard and Anthropic, in preprint arXiv:2610.22367, combined LLM Claude Mythos 5, evolutionary selection, and independent classical verification to find quantum circuits cheaper than known ones, including a 17-fold reduction in gate count for elliptic curve point addition.

Compiled by Sergey KostenchukPublished 2026-10-20Updated 2026-10-20

2026-10-20 Research Anthropic

🤖 AI cuts quantum circuit costs by 17x

A team from Harvard and Anthropic, in preprint arXiv:2610.22367, launched an evolutionary search: LLM Claude Mythos 5 proposes quantum circuits, while a classical evaluator checks correctness and the TQ-Cost metric. For 7 out of 8 tasks, the circuits are cheaper than known ones, and a new reversible architecture was found for the eighth. Elliptic curve point addition became 17 times cheaper, and the search took 53 hours.

🌍 For the first time, LLMs are improving basic quantum primitives in a cycle with independent verification: operations for post-quantum cryptography are becoming cheaper, and circuit optimization is moving toward automated search.

👤 The preprint (46 pages, CC BY-NC-SA 4.0) and Roemmele's article on the history of reversible computing from Landauer to Toffoli are open. Important: the gain is demonstrated only at the circuit level and has not been verified on real hardware.

Source 1: https://arxiv.org/abs/2610.22367v1 Source 2: https://brianroemmele.substack.com/p/the-gate-that-costs-forever-a-forgotten

Sources