Banger paper from Anthropic, Harvard, and colleagues.

They find that LLM-proposed quantum circuits beat the best-known designs on seven of eight tasks, with fewer Toffoli gates and fewer qubits.

They search over these circuit-generating programs with OpenEvolve, using Claude Mythos 5 to propose architectures and program rewrites.

LLMs propose circuit designs, an independent evaluator verifies correctness and scores the cost, and the best designs seed the next round.

The tasks cover numerical computation, chemistry, optimization, quantum simulation, and number theory.

On the eighth task, the search found a new reversible architecture for prime-field modular inversion. It makes divsteps, a classical algorithm impractical for reversible arithmetic, usable and improves the Toffoli-qubit trade-off.

The evaluator verifies every circuit before scoring, so each reported gain is a checked circuit.

https://arxiv.org/abs/2610.22367

Chat with Paper: https://academy.dair.ai/papers/discovering-state-of-the-art-quantum-circuits-with-large-language-models-2610.22367