Claude designs more efficient quantum circuits in Anthropic study

Anthropic said Claude found new quantum circuit designs that improve on the best published versions across eight tasks, with some gains reaching an order of magnitude or more. The company said the results could reduce the qubits and gates needed for useful quantum applications, a constraint that has kept many algorithms beyond current hardware.

Evidence and context

In a research note published on 16 Oct. 2026, Anthropic said it gave Claude plain-language specifications for eight circuits spanning chemistry, optimization, physics simulation, number theory and numerical computing, along with an automated evaluator that checked correctness and scored each design. The company said Claude Mythos 5 worked in a closed loop, proposing designs, receiving feedback, and improving them without human input between rounds.

Anthropic said the model produced novel circuits that used both fewer qubits and fewer gates than the best previously published algorithms. In one case involving elliptic-curve point addition, the company said Claude developed a new reversible modular inversion method after 53 hours of work with limited human prompting. Anthropic said one of the resulting designs used 12% more qubits than the smallest prior design but required 17 times fewer gates, while broader variants expanded the Pareto frontier of available trade-offs.

The company also said Claude redesigned a key step in decoded quantum interferometry, a quantum optimization approach. According to Anthropic, Claude replaced the extended Euclidean algorithm with Berlekamp-Massey decoding and reduced the decoder's gate count by a factor of 12 versus Google Quantum AI's previous construction. Anthropic said the combined changes made the most expensive step of the algorithm a little over an order of magnitude cheaper.

Anthropic said the work was led by Andi Gu as part of its STEM Fellows Program, with mentorship from Alek Kemeny, a quantum physicist at Anthropic, and collaboration from Susanne Yelin of Harvard University and Mikhail Lukin of Harvard University and QuEra Computing. The company said its researchers used formal checks, numerical simulation and Lean proofs in parts of the process.

Why the result matters

Quantum computers remain limited by qubit counts and error sensitivity, and Anthropic framed circuit design as one of the places where software-level improvements can matter before hardware scales. The company said its results suggest AI can help redesign quantum circuits, potentially shortening the path to practical applications in chemistry, materials science, fertilizers and related fields.

Anthropic also noted limitations. It said Claude was initially conservative, often second-guessing apparent improvements and preferring familiar methods unless prompted to broaden its search. The company said that behavior was sometimes useful, but it also had to be countered with instructions to explore multiple approaches in parallel.

The report did not specify a next catalyst or release date beyond the 16 Oct. 2026 publication of the research note.