Details
- Anthropic announced research from its STEM Fellows Program on using Claude Mythos 5 to design quantum circuits.
- Given plain-language specifications, Claude produced circuits for eight computational primitives spanning chemistry, optimization, physics simulation, number theory, and numerical computing.
- The model operated in a closed loop: it proposed circuit designs, received scores from an automated evaluator, and revised its output with minimal human intervention.
- The resulting designs used fewer qubits and gates than previously published circuits; some improvements reached an order of magnitude or more.
- The work was led by Andi Gu, with mentorship from Anthropic quantum physicist Alek Kemeny and collaboration from Harvard researchers Susanne Yelin and Mikhail Lukin, who is also associated with QuEra Computing.
- More efficient circuits reduce the quantum hardware resources required to run algorithms, potentially improving the practicality of future applications.
- Anthropic said quantum circuits are well suited to this workflow because they can be represented as code that AI systems can generate, execute, evaluate, and refine.
Impact
The announcement shifts AI’s role in quantum computing from assisting with software development toward discovering lower-resource algorithmic designs. It may pressure quantum hardware companies and research groups to pair their systems with automated circuit optimization, although the reported gains still require validation on physical hardware and across broader workloads. The result also supports a wider trend toward evaluator-driven AI research, where models generate candidates and independent programs verify performance rather than relying solely on human judgment.