Quantinuum cuts logical error rates on Helios
A preprint reports five logical errors per million correction cycles and stronger protection as the code grows, but the test covers only part of a useful fault-tolerant computer.
Quantinuum builds trapped-ion quantum computers and the software used to program them. Quantum information is fragile, so useful machines need logical qubits that spread information across several physical qubits and correct errors before they ruin a calculation. A new company-authored preprint reports a logical error rate of 5 per million error-correction cycles on its 98-qubit Helios system.
The experiment used triangular color codes at distances 3, 5 and 7, requiring 13, 37 and 73 physical qubits respectively for each logical memory test. As the code distance increased, the remaining logical error rate fell by about a factor of 12 at each step. That is the important part: adding protection improved reliability faster than it introduced new failure points, which is known as operating below the error-correction threshold.
The lowest measured rate corresponds to 99.9995% logical fidelity. Quantinuum says this is ten times better than its own September result and 100 times below the best published result on another hardware type. Those comparisons are company claims. The underlying paper is a preprint, not peer reviewed, and its strongest number comes from a logical-memory experiment. It also tested a limited family of two-qubit Clifford operations, not a long commercial algorithm.
The authors say a remaining step is to show the same below-threshold behaviour for the non-Clifford operations needed for universal computing. Quantinuum plans to make its fault-tolerant Apollo system available in 2029, but that date and the required scale remain forward-looking. The result strengthens the technical path because larger codes worked better on commercial hardware; it does not yet show that a useful fault-tolerant application can run end to end.