Claude Quantum Circuit Design: What Mythos 5 Just Improved

Anthropic says its Claude Mythos 5 model has done something quantum researchers usually measure in years of effort: it designed quantum circuits that use fewer qubits and fewer gates than the best published designs. In a research post dated October 16, the company describes giving Claude plain-language descriptions of eight circuit tasks and an automated checker, then letting the model iterate largely on its own. The results suggest AI-assisted quantum circuit design could bring useful quantum computers closer, because cheaper circuits ask less of hardware that is still scarce and fragile.

• Eight tasks, eight improvements: Anthropic reports Claude beat the best previously published design on qubits and gates in every case, across chemistry, optimization, physics simulation, number theory and numerical computing
• 12x cheaper decoder: a Reed-Solomon decoder for the DQI optimization algorithm uses about 12 times fewer gates than Google Quantum AI's earlier construction
• 17x fewer gates: one elliptic-curve point-addition design uses 12% more qubits but 17 times fewer gates than the smallest previous design, found in a single 53-hour run
• Open paper: the full study is on arXiv, led by Andi Gu of Anthropic's STEM Fellows Program with Harvard's Susanne Yelin and Mikhail Lukin

How Did Claude Design Quantum Circuits?

The setup is a closed loop. Claude wrote each candidate circuit as ordinary Python code. An evaluator program checked whether the circuit computed the right answer, then scored it by multiplying qubits by gates, a rough proxy for how long the circuit would take to run. Claude read the score, revised and tried again, with no human input between rounds.

Quantum circuits make this harder than normal programming. They cannot branch on "if-then" logic the way classical code does, and they cannot simply throw away scratch work, because every step must be reversible. Intermediate results either stay around and eat qubits, or get "uncomputed" at the cost of extra gates. Good circuit design is a constant trade between space and time, which is exactly the kind of structured puzzle large language models are getting good at.

The Two Headline Discoveries

The first came from elliptic-curve point addition, a standard benchmark in the field. Its most expensive step is modular inversion. Classical computers handle it with a method called divsteps, but a straightforward quantum version has to store a long record of every branch it took. Claude reasoned that the record could be decoded from the numbers themselves. It proved several new algebraic identities, derived an exact decoding rule, tested it in large simulations and formalized the claim in the Lean proof assistant. The result is the first reversible divsteps with no stored branch history, plus a family of designs that trade qubits for speed.

The second targeted decoded quantum interferometry, or DQI, one of the strongest candidates for a large quantum speedup on optimization problems. Claude swapped the decoder's underlying method for Berlekamp-Massey decoding, which runs the same steps every round. It then wrote a fast simulator, ran 403 million sample decodings, and found it could reserve memory for just one saved value, 19 qubits instead of a worst-case 665. Gate count fell by roughly a factor of 12.

What Does This Mean for Practical Quantum Computing?

Anthropic frames the work against a familiar benchmark: estimates for simulating FeMoco, the enzyme cluster bacteria use to fix nitrogen, fell from about 50 trillion Toffoli gates in 2017 to about 340 million today. That progress took many research groups about eight years, and Anthropic is clear that it was not Claude's doing. The point is that Claude's elliptic-curve result came from one 53-hour run. If AI can compress that research cadence, catalyst design, battery chemistry and materials science could arrive within reach of real machines sooner.

Two honest caveats apply. This is a preprint, not yet independently reproduced. And Anthropic notes Claude was conservative at first, preferring well-known approaches until the team asked for several parallel strategies and occasional nudges to think more broadly.

Because point addition is also the arithmetic behind future quantum attacks on elliptic-curve cryptography, the work is another reminder that the steady industry migration covered in our post-quantum roadmap reporting is time well spent. For more on how frontier models are reshaping research, browse our AI coverage and our look at the Claude Science research workbench.

Sources: Anthropic Research — October 16, 2026; arXiv — October 2026; Interesting Engineering — October 16, 2026.