Designing More Efficient Quantum Circuits With Claude

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Anthropic says Claude Mythos 5 independently designed more resource-efficient circuits for eight quantum computing tasks, beating the best published designs on the measures tested. The results are from an Anthropic-led research project and do not show that quantum hardware is ready sooner or that the circuits have been demonstrated on a quantum computer.

Anthropic says Claude Mythos 5 designed more resource-efficient quantum circuits for eight computational tasks, improving on previously published designs according to an automated evaluator. The research suggests AI systems can help reduce the number of quantum resources some algorithms require, but it does not establish that the circuits have run successfully on quantum hardware or that practical quantum computing will arrive sooner.

The Anthropic-led team gave Claude plain-language specifications for circuits used in chemistry, optimization, physics simulation, number theory and numerical computing. Claude proposed designs as code, while an external program checked whether each design produced the required output and assigned a score. According to Anthropic, the model could use that feedback in a closed loop, revising its proposals without human input between rounds.

Anthropic says the resulting designs used fewer qubits and fewer gates than the best previously published algorithms for all eight tasks. The project measured resource use with a score based on qubit count multiplied by gate count, a proxy for circuit cost rather than a direct measurement of runtime on a functioning quantum computer. The report describes some gains as an order of magnitude or more, but the supplied material does not provide the individual results or the detailed comparisons for every circuit.

The research was led by Andi Gu through Anthropic’s STEM Fellows Program, with mentorship from quantum physicist Alek Kemeny. Susanne Yelin of Harvard University and Mikhail Lukin of Harvard and QuEra Computing collaborated. Anthropic says Claude’s improvements came from identifying mathematical structure, including symmetries that let a calculation be reused and components that could be removed when they handled only rare inputs.

Reducing the Cost of Quantum Algorithms

Quantum machines remain limited by both the number of qubits available and the time those qubits can maintain reliable states. They are also vulnerable to environmental noise, which can introduce errors as a computation proceeds. A circuit that needs fewer gates and qubits could, in principle, reduce demands on hardware and lower the exposure to accumulated errors.

The report matters because it offers a potential way to improve quantum computing without waiting for hardware advances: redesign the computation itself. Anthropic’s results, if independently validated and applicable beyond the tested examples, could help researchers search a larger range of circuit designs. But resource savings on selected circuits are not the same as a practical application running faster or more reliably on a real machine.

Anthropic links the work to difficult scientific problems such as modeling FeMoco, an iron-sulfur cluster involved in nitrogen fixation. Better simulations could eventually help researchers understand the chemistry behind industrial catalysts. That is a long-term possibility, not an outcome demonstrated by this circuit-design study.

Why Circuit Resource Counts Matter

Quantum algorithms are expressed as sequences of operations called gates, applied to qubits. Some gates are more costly than others; the source report notes that Toffoli gates are a common way to estimate the cost of certain computations. Reducing gate counts can lower estimated computational demands, though actual performance depends on hardware, error correction and how the circuit is implemented.

The source uses FeMoco to illustrate how much algorithm design can affect estimates. It says a 2017 estimate put the work needed to calculate the molecule’s ground-state energy at about 50 trillion Toffoli gates. Later algorithmic improvements brought the estimate to roughly 340 million, a reduction of about 150,000 times, according to Anthropic. Those figures refer to successive estimates for a particular computation, not to a quantum computer already performing it.

Anthropic chose tasks whose outputs can be checked efficiently by a conventional program. That made it possible to test Claude’s candidate circuits automatically and provide an objective score. The approach therefore concerns a defined class of circuit-design problems, not unrestricted quantum programming.

“Claude Mythos 5 could successfully work in a closed loop, proposing a circuit design, receiving feedback and a score from the evaluator program, and using that feedback to improve its next attempt, with no human input between rounds.”

— Anthropic

Evidence Beyond the Evaluator

The supplied report does not establish that the circuits have been run on quantum hardware. It also does not provide enough detail here to assess each task’s exact resource savings, the full comparison methods, or whether the results have undergone independent replication. The paper is linked by Anthropic as an arXiv preprint, so its findings should be treated as research claims rather than settled results.

It is also unclear how broadly the method will work. The experiment focused on eight specified computational primitives that could be checked by an external evaluator. The reported resource score is a proxy, and a lower score does not by itself show that a circuit will outperform alternatives on a particular device, where hardware layout, noise and error correction affect performance.

Anthropic’s suggestion that such reductions could bring practical quantum computing closer is an interpretation of possible downstream benefits. The work does not establish a timeline for useful quantum machines or show that any specific application is now feasible.

Paper Review and Further Tests

The research team points readers to its arXiv paper, where the circuit designs and evaluation details can be examined. Further work would need to establish how the methods compare under independent review, reproduce the results, and test whether the resource savings remain meaningful under realistic hardware and error-correction constraints.

For now, the confirmed development is a reported demonstration of automated circuit search on eight evaluator-checkable tasks. The next significant evidence would be independent validation and a clearer account of how these designs translate from lower estimated resource counts to useful computations on quantum devices.

Key Questions

What did Claude do in the study?

Anthropic says Claude Mythos 5 proposed and revised quantum circuit designs for eight computational tasks, using feedback from a program that checked correctness and scored resource use.

Did the circuits run on a quantum computer?

The supplied report does not say that the designs were executed on quantum hardware. It describes evaluation by an automated software checker.

What does “more efficient” mean here?

Anthropic compared designs using qubit and gate counts, including a score based on qubits multiplied by gates. That is a resource-cost proxy, not a direct hardware runtime measurement.

Does this mean practical quantum computers will arrive sooner?

Not on the evidence provided. Lower resource estimates could help future quantum applications, but the study does not show that a useful application has run or provide a timeline for practical quantum computers.

Who conducted the research?

The work was led by Andi Gu through Anthropic’s STEM Fellows Program, with mentorship from Alek Kemeny and collaboration from Susanne Yelin and Mikhail Lukin.

Source: hn