Claude designed quantum circuits that outperform the best published versions. It worked largely on its own using plain-language specs and an automated evaluator. Improvements reached an order of magnitude in some cases. The work reduces quantum resources needed for practical applications. It may accelerate the arrival of useful quantum computers. Two examples include a new reversible algorithm for modular inversion. Another made a key quantum optimization step about 12x cheaper. The research was led by Andi Gu with mentorship from Alek Kemeny. Harvard’s Susanne Yelin and Mikhail Lukin also collaborated. The full paper is on arXiv.
What changed
Best published quantum circuits were the most efficient known.
What it unlocks
Designing quantum circuits that use fewer qubits and gates than prior best-known versions.
- 53 hours of autonomous work
- 12x cheaper for modular inversion
- 17x fewer gates in one design
- 12% more qubits in one design
What you need to act on it
- Claude Mythos 5
- plain-language specifications for circuits
- an automated evaluator for correctness and scoring
Sources