Anthropic reports that Claude improved designs for eight quantum-circuit tasks, including one elliptic-curve circuit using 17 times fewer gates.
Anthropic says its Claude Mythos 5 model designed improved circuits for eight quantum-computing tasks spanning chemistry, optimization, physics simulation, number theory and numerical computing. The company describes a workflow in which the model received plain-language specifications, proposed circuit designs in code, and used an automated evaluator to check correctness and score resource costs.
The most striking reported result concerns elliptic-curve point addition, a mathematical operation used in algorithms for the discrete logarithm problem. Anthropic says Claude developed a reversible approach to modular inversion that avoids storing a long history of intermediate steps. After 53 hours of work with minimal human input, the model produced a design that the researchers say was checked with large-scale simulations and formalized in Lean. Compared with the smallest previously published design, one circuit used 12% more qubits but 17 times fewer gates.
The company also reports that Claude reduced the gate count of the costliest step in a quantum optimization algorithm by a factor of 12 relative to a prior construction. These results concern circuit designs and resource estimates; they do not demonstrate that existing quantum computers can run the target algorithms or that practical quantum advantage is imminent. Quantum hardware remains constrained by qubit count, noise and error correction.
Anthropic’s work was led by Andi Gu through its STEM Fellows Program, with collaborators from Harvard University and QuEra Computing. The company says the designs improved on prior published circuits, but the findings are presented by the model developer and should be interpreted as research results requiring scrutiny and follow-up.
The broader significance is methodological: circuits expressed as code can be tested against explicit correctness checks and resource measures, creating a feedback loop in which AI proposes and revises technical designs. Anthropic notes that the model sometimes preferred familiar approaches and needed prompts to explore more varied ideas. The results suggest a possible new research tool for quantum algorithm design, while leaving open how broadly the approach generalizes and how its proposed circuits perform on real hardware.