Claude designs quantum circuits that beat the best published ones

On 16 October 2026 Anthropic published results from a STEM Fellows Program project led by Andi Gu, mentored by Anthropic quantum physicist Alek Kemeny and run with Susanne Yelin and Mikhail Lukin of Harvard. Claude Mythos 5 received plain-language descriptions of eight circuit building blocks and worked in a loop with an evaluator program that checked each design and scored it by qubits multiplied by gates, with no human input between rounds.

On every task Claude found designs that need both fewer qubits and fewer gates than the best published ones. For elliptic-curve point addition it worked for 53 hours and produced the first reversible version of the divsteps method for modular inversion that stores no branch history; one design in the resulting family uses 12% more qubits but 17 times fewer gates than the smallest earlier design. For the decoder in the DQI optimization algorithm, the gate count fell about 12-fold compared with Google's previous construction.

Anthropic also lists limits. Claude was cautious, rechecked well-supported results and leaned toward familiar methods, so the team had it develop several approaches in parallel and sometimes prompted it to search more broadly. The full paper is on arXiv.