Quantum computing hardware news is usually measured in qubit counts: superconducting chips boast thousands of physical qubits, trapped ions build logical qubits, neutral-atom arrays stretch into the thousands. All assume information lives in ordinary qubits, rescued from errors by error-correcting codes.
On September 25, a paper published in *Nature* took a different path. A team from the University of Chicago's Pritzker School of Molecular Engineering, Harvard, Stony Brook University, and Quantinuum used 54 entangled qubits on the H2 trapped-ion processor to create non-Abelian anyons—and for the first time assembled the complete set of operations needed for universal quantum computation. The paper is titled *Universal gates from braiding and fusing anyons on quantum hardware*, published in *Nature* vol. 655, issue 8123.
Particles That Remember Order
Anyons exist only in two dimensions and cannot be found in nature. They emerge when dozens of ordinary qubits are entangled into a collective state that behaves like a new kind of particle, with rules unlike electrons or photons. Co-author Ruben Verresen of the University of Chicago offered an analogy: these encodings create "little universes," parallel universes that nevertheless reflect properties of our own.
"Non-Abelian" describes their quirk: taking one anyon around another changes its internal state, and the order of exchanges matters. Ordinary particle exchanges are like moving chairs—order is irrelevant. Non-Abelian exchanges are like shuffling a deck—the sequence of moves determines whether you hold a flush or a mess. Information is not stored in any single qubit but spread across the whole entangled state, giving it inherent immunity to certain local perturbations.
2024's Shortcoming, Closed in 2026
In 2024, a team including Verresen created non-Abelian order on Quantinuum's machine using the symmetry group D4 of a square—the first time such an order was realized in hardware. But that experiment had a hole: braiding alone could not supply all operations for universal computation. That particular universe, Verresen recalled, was not powerful enough.
This time the team used S3, the symmetry group of an equilateral triangle (rotations plus mirror flips). They encoded topological qutrits—three-level units with one more level than ordinary qubits—in anyon pairs. Then came the crucial step: combining two types of operations. Braiding contributed one entangling gate; fusion contributed two different measurements. Fusion means bringing two anyons together and measuring their combined state. Together, these three tools can in principle generate all quantum operations—including those braiding alone could never achieve. That is precisely the definition of a universal gate set.
Carlos Mochon, who proposed the idea in 2003 as a student of John Preskill at Caltech, saw 23 years pass between a paper possibility and a machine demonstration.
Cutting the Most Expensive Item on the Error-Correction Bill
The practical value lies in error-correction economics. Quantum computers must correct errors, and the standard approach spreads one logical qubit across many physical qubits. But protected logical qubits can typically execute only a subset of operations; the missing part is supplied by a resource called magic states. Producing magic states usually requires distillation: preparing many copies and repeatedly purifying them, consuming a large share of the machine's qubits.
Henrik Dreyer, head of Quantinuum's Munich office, said non-Abelian encodings are a dark horse in the error-correction race, and that this work demonstrates the first universal gate set within a non-Abelian encoding—meaning fault-tolerant computation could in principle proceed without magic state distillation and cultivation, the two most expensive operations in standard codes. In the S3 scheme, magic states can be produced directly by topological operations, skipping distillation entirely.
A Dark Horse Not Yet at the Finish Line
The numbers are 54 qubits, three-level encoding, one gate plus two measurements. But the next sentence matters: this experiment did not perform active error correction. In Verresen's words, "we have been dodging the error-correction question so far—this is more of a proof of principle." The team validated the components; the full machine is not yet assembled. The next step is combining these operations with active error correction, and Verresen is already working with colleagues on stabilization techniques for non-Abelian quantum memories.
To complete the horse-race metaphor: the dark horse is not outrunning superconductors and trapped ions in qubit counts—it is beating them on the bill. The most expensive item in mainstream error-correction overhead has been shown, in principle, to be avoidable on this route. That is all "dark horse" means—not already leading, but having a lower-effort path in the later stages. What cost this route itself demands, and whether it holds on noisier hardware, the paper does not answer, and cannot yet.
Anasuya Lyons and Chiu Fan Bowen Lo, two graduate students in Harvard's Vishwanath group who helped lead the work, put it plainly: an idea conceived during their PhDs could be realized in the lab thanks to the astonishing progress in quantum hardware over recent years. A proposal from Preskill's student 23 years ago has now taken its first step on 54 qubits.