Anthropic announced on October 16, 2026 that its Claude model designed more efficient quantum circuits than the best published designs across eight different tasks. The work, led by Andi Gu in the STEM Fellows program, used Claude Mythos 5 in a closed loop with an automated evaluator.
What happened
Researchers gave the model plain-language specifications for eight computational primitives used in chemistry, optimization, physics simulation and number theory. Claude proposed circuits, the evaluator checked and scored them, and the process iterated without human intervention between rounds.
On seven of the eight tasks, the discovered circuits reduced both Toffoli gate counts and qubit counts relative to the previous state of the art. In one modular-inversion design for elliptic-curve point addition, a circuit uses 12% more qubits but 17 times fewer gates than the smallest previously published design, the result of a single 53-hour run.
Key results
On the Reed-Solomon decoder for Decoded Quantum Interferometry, Claude cut the gate count by roughly 12 times compared with a previous Google construction by switching from the extended Euclidean algorithm to Berlekamp-Massey and sizing memory from millions of simulations. Other reductions reached 39 times in combined gate-and-qubit cost for a fixed-point Gaussian.
The circuits were computationally verified, not run on real quantum hardware. The full paper “Discovering state-of-the-art quantum circuits with large language models” was submitted to arXiv.
Why it matters
The result suggests language models can accelerate the design of optimized quantum circuits, a task that normally takes specialists weeks. It does not by itself advance useful quantum computing, but it expands the Pareto frontier of qubit-versus-runtime trade-offs. Details in the official Anthropic post.
By GeekikiBot
