Anthropic just said Claude Mythos 5 designed quantum circuits that beat the best published ones. Here's what you need to know. The work comes from Anthropic's STEM Fellows Program, with Harvard and other colleagues. Claude was given plain-language descriptions of what each circuit must compute, plus an automated checker to verify its answers. It looped on designs, tested them, and improved them. The search over circuit-generating programs used OpenEvolve with Claude. The result: Claude's designs beat the best published ones on 7 of 8 tasks, with fewer Toffoli gates and fewer qubits. Anthropic says some improvements were an order of magnitude or more. Fewer resources per circuit could bring quantum applications in chemistry and materials science within reach sooner. One run lasted 53 hours with minimal human intervention. It started from a one-paragraph brief. Claude developed a new reversible implementation and proved new algebraic identities in Lean. The headline cases involve elliptic-curve point addition, the core step in quantum attacks on the curves behind Bitcoin and Ethereum signatures, and a quantum decoder for DQI (decoded quantum interferometry). Key numbers: - 7 of 8 tasks beat the best published designs - Up to 39x lower combined gate-and-qubit cost - Point addition: 17x fewer gates, about 12% more qubits - DQI decoder: 12x fewer gates than Google's design - 53 hours for the longest run One researcher, Soubhik Deb, disputes the Pareto-frontier claim. He says ecdsa.fail currently shows the frontier at 843 qubits sitting near 11.947M. Anthropic posted this on Oct 17, 2026, and it is the first Mythos 5 quantum circuit result it has published.