Anthropic released research on October 16 showing that its Claude Mythos 5 artificial intelligence model created quantum circuits that surpass the best published algorithms across eight benchmark tasks. The system cut gate counts by up to 17 times with very little human assistance.
The eight benchmarks included physics simulation, chemistry, number theory, optimization, and numerical computing. For every challenge, Claude took a simple text description and operated inside an autonomous loop. The model suggested Python circuit designs, took feedback from an evaluator, and updated its strategy without any human help between iterations.
Elliptic curve point addition, an essential component in quantum cryptography, saw the biggest enhancement. Operating for 53 hours, Claude generated the initial reversible version that keeps zero branch history. Consequently, this new layout uses 17 times fewer gates than the prior top design while requiring only 12% more qubits.
Quantum systems demand massive resources, making circuit efficiency vital. For instance, simulating the FeMoco molecule, which is critical for biological nitrogen fixation, previously demanded about 50 trillion gates. Human scientists brought that down to roughly 340 million over eight years. The method used by Claude might speed up similar efficiency gains for other challenges.
Claude also lowered gate counts by 12 times for a quantum optimization decoder known as decoded quantum interferometry. Google Quantum AI previously created the top model for this circuit. Claude accomplished this reduction by changing the decoding strategy and rebuilding temporary values directly from the active state of the circuit instead of saving them.
Anthropic described the potential impact:
Rather than designing each component on its own, AI could optimize all of them jointly, from the algorithm down to the hardware, to build an entire error-corrected quantum computer tailored to a specific problem. Such designs could also tell hardware builders which capabilities matter most for a given application. That could put problems such as designing catalysts and understanding new materials within reach of the first fault-tolerant machines.
During the elliptic curve task, Claude additionally demonstrated new algebraic identities and proved its assertions using Lean, which is a mathematical proof language. The model also executed 403 million simulated decodings to test its optimization decoder layout.
The project was spearheaded by Andi Gu through the STEM Fellows Program at Anthropic. Alek Kemeny, an Anthropic quantum physicist, served as mentor. The group worked alongside Mikhail Lukin and Susanne Yelin from Harvard University, alongside QuEra Computing.
According to Anthropic, automated circuit creation could bring practical quantum computing closer. At the same time, the firm points out that circuit design solves only a single piece of fault-tolerant quantum computing, as decoders, error-correcting codes, and hardware control require upgrades as well. Potential uses involve better batteries, new medicines, catalyst design, and cleaner fertilizer manufacturing.
