Anthropic Says Claude Designed Quantum Circuits That Beat the Best Published Versions
In an Anthropic STEM Fellows project with Harvard physicists, Claude Mythos 5 worked in a closed loop with an automated checker and, Anthropic says, found circuit designs using fewer qubits and fewer gates than the best published versions on seven of eight benchmark tasks, plus a new reversible approach to modular inversion on the eighth.
By 1EyeAI Desk ·
Anthropic published a research post on Oct. 16, 2026, with an arXiv preprint (2610.22367, dated Oct. 15), “Discovering state-of-the-art quantum circuits with large language models,” by Andi Gu, Alek Kemeny, Susanne Yelin (Harvard) and Mikhail Lukin (Harvard and QuEra Computing). The team gave Claude Mythos 5 plain-language specs for eight circuit primitives drawn from chemistry, optimization, physics simulation, number theory and numerical computing, plus an independent evaluator that checks each design for correctness and scores it on qubits and gates.
According to the paper’s abstract, the discovered circuits on seven of the eight tasks significantly reduce both Toffoli-gate and qubit counts relative to the state of the art. On the eighth, elliptic-curve point addition, Anthropic says Claude worked for 53 hours with minimal human input and produced the first reversible version of the “divsteps” modular-inversion method that doesn’t store branch history, checking it with large simulations and a Lean proof. One design in that family uses 12% more qubits but 17x fewer gates than the smallest previous design, per Anthropic.
Anthropic also says Claude cut the gate count of the Reed–Solomon decoder used in decoded quantum interferometry (DQI) by about 12x versus Google Quantum AI’s earlier construction, by switching to Berlekamp–Massey decoding and storing far less scratch data.
Anthropic notes limits: Claude tended to be overly cautious and favor known approaches, so researchers had it pursue several approaches in parallel and occasionally nudged it to search more broadly. The idea that practical quantum computers “could arrive sooner” is Anthropic’s projection; fault-tolerant hardware at the needed scale doesn’t exist yet.
What this means for you: If you work in quantum algorithms, this is a working recipe: express a design as code, attach an automatic checker and a cost score, and let a model iterate. For everyone else, it’s another sign AI is speeding up hard science wherever answers can be machine-checked.