Anthropic's Claude Mythos 5 model has developed a quantum circuit for elliptic curve point addition that requires 17 times fewer gates than the most compact previously published version, despite using 12% more qubits. The project was led by Anthropic's Andi Gu in collaboration with researchers from Harvard University and QuEra Computing.
The model optimized eight fundamental quantum operations used in fields such as chemistry, physics modeling, and number theory. Claude generated circuit designs, while an automated system evaluated their correctness and efficiency. The model improved its iterations autonomously without human intervention between rounds.
For the elliptic curve point addition, Claude created a reversible implementation of divsteps that eliminates the need to store branching history. By deriving new algebraic identities, the model developed a formula to reconstruct intermediate data, significantly reducing the gate count required for these cryptographic calculations.
Claude also optimized the Decoded Quantum Interferometry (DQI) algorithm by replacing the extended Euclidean algorithm with the Berlekamp-Massey method. This reduced the gate count for the Reed-Solomon decoder by approximately 12 times compared to previous designs from Google Quantum AI.
The model completed the optimization tasks in 53 hours. While the results are currently published as a preprint and await peer review, researchers suggest that AI-driven mathematical discovery could accelerate the development of practical quantum computing hardware.
