Claude beats published quantum-circuit designs by up to 12x

Anthropic says Claude Mythos 5 designed quantum circuits that outperform the best published alternatives across eight computational tasks spanning chemistry, optimization, simulation, number theory and numerical computing.

The model worked in a closed loop: it proposed a circuit, received automated correctness and cost feedback, and iterated with little human input. According to Anthropic, the gains came from new structural insights rather than small parameter tweaks.

For elliptic-curve arithmetic, Claude developed a reversible version of the divsteps method that avoids storing a full branch history. One design uses 12% more qubits than the smallest prior circuit but requires 17 times fewer gates. For a decoder used in decoded quantum interferometry, Claude switched algorithms and reconstructed temporary state instead of storing it, reducing the gate count by a factor of 12.

Anthropic says the elliptic-curve result emerged from a 53-hour run and was checked with simulations and a formal Lean proof. Across the full set of tasks, the company reports lower gate or qubit costs than previous published designs.

The caveat matters: these are Anthropic’s research results, and better circuits solve only one part of the much larger hardware and error-correction problem. Still, the work suggests AI can compress months of specialized circuit-design iteration into machine-checkable research loops.