Anthropic has published research showing that its Claude Mythos 5 model developed quantum circuits requiring up to 17 times fewer computational gates than the smallest previously published designs. The findings, dated October 16, follow Google's recent quantum computing breakthrough, which reduced errors during experiments involving its Willow processor.

According to Anthropic's research, the AI model improved eight circuit designs covering quantum chemistry, physics simulations, optimization, and number theory. The results came from automated design tests, and researchers have not yet demonstrated the circuits on quantum hardware.

Fewer Gates for Elliptic-Curve Operations

The project was conducted through Anthropic's STEM Fellows Program, led by Andi Gu with support from physicist Alek Kemeny. Harvard researchers Susanne Yelin and Mikhail Lukin also contributed. The work examined computational tasks that serve as building blocks for larger quantum algorithms.

Claude Mythos 5 received written instructions describing eight circuit design problems and worked toward solutions using an automated evaluator. The system checked each proposal for correctness and measured efficiency through qubit requirements and quantum gate counts. Gate counts are a closely watched efficiency measure in quantum computing because every gate is an operation that hardware must execute, and on machines that still struggle with qubit stability, each added operation increases the error-correction workload.

One experiment focused on elliptic-curve point addition, an operation that requires modular inversion to complete its mathematical steps. Claude developed a reversible version of the divsteps algorithm that avoids storing branch history during calculations. Reversibility is a standing requirement in quantum computing, since quantum operations cannot discard intermediate values the way classical ones can. The resulting design reduced quantum gate requirements by approximately 17 times compared with the smallest previously published circuit, though it required 12% more qubits — a trade-off between memory and computational operations.

The findings add to growing research into quantum hardware efficiency and the resources required for complex cryptographic calculations. Earlier Google quantum computing research examined how future machines could challenge the elliptic-curve encryption used in Bitcoin. Because Bitcoin's digital signatures rest on that same elliptic-curve arithmetic, the cost of operations like point addition is one of the inputs that shapes estimates of how large a future quantum machine would need to be.

Anthropic's experiment ran for approximately 53 hours, during which Claude repeatedly tested and refined its mathematical approach. Researchers verified the modular inversion method through numerical simulations and formal mathematical proofs using Lean software, an open-source proof assistant that checks each step of a mathematical argument.

12x Improvement in Quantum Error-Correction Decoding

Claude Mythos 5 produced another improvement while working on Reed–Solomon decoding within the Decoded Quantum Interferometry optimization algorithm. The model replaced an extended Euclidean approach with the Berlekamp–Massey algorithm, reducing the number of operations required during decoding. Berlekamp–Massey is a classical decoding technique dating to the 1960s and remains a standard tool for Reed–Solomon codes, which are widely used in storage and communications systems.

The redesigned circuit cut gate requirements by approximately 12 times compared with Google's previous construction. Across 403 million test samples, Claude reduced reserved storage from 665 qubits to 19 under a specified error allowance.

These are improvements to circuit designs rather than completed calculations on physical quantum processors. Current hardware still faces challenges involving qubit stability, error correction, and the resources needed for extended computations.

Crypto Sector Prepares for Quantum Threats

The research arrives as cryptocurrency companies prepare for possible future quantum attacks against existing digital signature systems. Earlier research on quantum threats to cryptocurrency examined how advances in artificial intelligence could influence cryptographic security timelines.

Meanwhile, Coinbase recently urged Bitcoin developers to begin preparing quantum-resistant upgrades, warning that implementation could take years. The exchange's recommendations covered alternative signature systems and wallet designs that could support future transitions.

Anthropic has continued developing Claude models for scientific research and technical problem-solving, including coding and mathematical applications. Its earlier Claude Opus 5.5 release focused on coding performance and professional applications.

The latest findings remain subject to further testing, particularly on physical quantum computers. Anthropic has not announced a hardware demonstration or a deployment schedule for the newly designed circuits. Until such a demonstration takes place, the reported gate savings exist in simulation and formal proof rather than on a machine, and whether Bitcoin's quantum-resistant transition moves beyond recommendations remains to be seen.