Anthropic says Claude Mythos 5 designed improved quantum circuits
The model cut qubit and Toffoli gate counts in seven of eight basic operations, and spent 53 hours on one new calculation method.
Anthropic has published research in which its Claude Mythos 5 model designed quantum circuits, with researchers from Harvard University taking part, according to aitimes.com. The work covered eight basic operations used in chemistry, optimization and physics simulation.
In seven of the eight tasks, the AI reduced both qubit counts and Toffoli gate counts compared with the previous best designs. In one elliptic curve arithmetic design, Claude used 12% more qubits than the previous minimum-qubit design, but cut the gate count to one seventeenth. In a quantum algorithm for optimization problems, its design cut the gate count to one twelfth of the existing design.
The setup paired Claude's code-writing with a separate evaluation program that checked accuracy and resource use, then fed the results back for redesign. On one problem the model worked for 53 hours with initial instructions and occasional directional hints before producing a new calculation method.
Design figures, not hardware runs
The reported numbers compare specific circuit designs rather than full runtime on real quantum hardware. The paper is posted on arXiv.
The new circuits were measured against earlier published designs, which served as the baseline for efficiency. Anthropic did not state how the designs would perform on physical quantum processors.
Quick answers
What did Claude Mythos 5 do in Anthropic's research?
It designed quantum circuits, reducing qubit counts and Toffoli gate counts compared with previous best designs in seven of eight basic operations.
How long did the model work on the new calculation method?
Claude worked for 53 hours with initial instructions and occasional directional hints before producing a new calculation method.
Were the circuits tested on real quantum hardware?
The figures compare specific circuit designs, not full runtime on real quantum hardware.