Claude designs leaner quantum circuits than published bests
Anthropic says Claude Mythos 5, working largely on its own, found quantum circuit designs that beat the best published versions on eight core routines, which would cut the resources useful quantum applications are estimated to need.
Anthropic has published research in which Claude Mythos 5 designed quantum circuits that use both fewer qubits and fewer gates than the best previously published designs. The routines tested were eight computational primitives drawn from chemistry, optimization, quantum simulation, number theory and numerical computing.
The setup was a closed loop. Claude received a plain-language specification of what each circuit must compute and wrote a design as code. An automated evaluator then checked the design for correctness and scored it by qubit count multiplied by gate count, which serves as a proxy for runtime. Claude used that feedback to revise its next attempt, with no human input between rounds. The method works for a common class of algorithms in which each input has one definite output that a classical computer can verify.
Anthropic says some gains were an order of magnitude or more. These include roughly 17x fewer gates for elliptic-curve point addition, via a new approach to modular inversion, and a 12x cheaper decoder for DQI, a quantum optimization algorithm. The company says the gains came from new structural insights, not from tuning existing designs. In one case, Claude exploited a molecule's symmetries to reuse calculations. In others, it removed costly components that existed only to handle extremely rare inputs.
The results are resource estimates, not runs on quantum hardware. Andi Gu led the work through Anthropic's STEM Fellows Program, with Anthropic physicist Alek Kemeny and Harvard's Susanne Yelin and Mikhail Lukin, who is also with QuEra Computing. A paper is on arXiv. The material provided did not include full results for all eight primitives.