THE SCOREBOARD FOR WHAT'S NEXTAI + QUANTUM, SCORED. COMPARED. EXPLAINED.
AIQUANTUMSCORE.
← AI & Quantum News

Quantum / NEWS BRIEF

Quantinuum reports a tenfold improvement in logical error correction

A Helios experiment achieved a reported logical memory error rate of roughly five per million correction cycles and stronger suppression as the color-code distance increased.

By AIQuantumScore · · Event 2026-10-19

The measured result

Quantinuum researchers ran triangular color codes at distances three, five and seven on the 98-qubit Helios trapped-ion processor. Their paper reports a lowest logical memory error rate of 5 × 10^-6 per syndrome-extraction cycle, with a stated uncertainty range of 3 × 10^-6 to 9 × 10^-6. That central value corresponds to about five errors per million cycles, or 99.9995% logical memory fidelity for the measured operation. Quantinuum says the error rate is ten times below its own September result.

Evidence of below-threshold scaling

The central technical result is not one isolated fidelity number. Logical memory errors fell by a reported factor of about 12 each time the code distance increased, evidence that added redundancy was suppressing errors rather than creating more of them. The researchers also tested encoded two-qubit Clifford operations: the distance-five code reached a reported error rate of 7 × 10^-5 per logical Clifford, with uncertainty of roughly 4 × 10^-5 to 10 × 10^-5. Neural-network decoders ran faster than the experiment's syndrome-extraction rate, according to the paper.

A milestone, not a finished fault-tolerant computer

OUR ANALYSIS

Below-threshold scaling is essential because useful fault-tolerant machines must become more reliable as their codes grow. The result therefore strengthens the case for trapped-ion error correction. Important limits remain: the study is a Quantinuum-authored preprint, the best memory result used up to 37 physical qubits for one logical qubit, and it does not demonstrate a large useful algorithm. Quantinuum's claim that this validates a path to its planned Apollo system by 2029 is a company forecast, not an independently established delivery date. The experiment does not by itself justify a benchmark score change.

More to follow