Quantinuum sets quantum error correction record with Helios

On Monday, Quantinuum announced what it called a world-record quantum error correction result on its Helios trapped-ion computer, holding an error-corrected logical qubit at 99.9995 percent logical memory fidelity.

The figure works out to roughly five failures per million error-correction cycles, the company said in a statement on October 19. It is tenfold better than Quantinuum's own record from September and about a hundred times better than the best published result from any other hardware approach, set by Google Quantum AI on a superconducting system, according to the announcement.

The experiment ran on the same machine the company used for the September test, but swapped the company's Helix code for a different scheme known as a color code. September's setup protected each logical qubit with ten physical qubits; the new test used up to thirty-seven per logical qubit, trading hardware overhead for accuracy. Quantinuum chief executive Rajeeb Hazra told The Wall Street Journal the result "derisks the industry," adding that the team now knows how to create reliable physical qubits and scale up their protection in a practical way.

The researchers encoded a single logical qubit in triangular color codes of distance three, five and seven, running twenty rounds of syndrome extraction, the repeated parity checks that catch errors, on each. The logical error rate per round fell as the code distance grew, and the fitted suppression factor came out at twelve: each step up in distance divided the remaining errors by twelve. Logical gate operations on pairs of encoded qubits improved with distance too, the preprint authors wrote.

Why error rates decide the quantum race

Quantum bits lose their state to the slightest disturbance, so a quantum computer's raw operations are too unreliable for long calculations. Error correction groups many physical qubits together to protect one logical qubit, checking and fixing errors as they appear. It is what turns a laboratory demonstration into a machine that can run the long calculations quantum computing was built for.

The industry has heard this promise for years while delivery lagged, which is why outside reaction mattered here. Robin Blume-Kohout, co-director of Sandia National Laboratories' Quantum Performance Lab, who was not involved in the study, said in Quantinuum's announcement that the paper showed "real, serious performance gains," and that a scaled-up version of the system "could reach the reliability required for useful fault-tolerant computation at a feasible size." The details appear in a preprint submitted to arXiv on October 16 by twenty-three researchers at Quantinuum's sites in Broomfield, Colorado, and London. It had not been peer reviewed at the time of the announcement.

One reason the result drew attention is flexibility. September's record used one code on the same machine; October's used another, with better numbers. The company says its architecture lets researchers test, refine and combine different codes to trade accuracy against hardware cost, instead of betting on a single approach. For a field where each added layer of protection normally costs far more qubits, that counts: documents from the study show the color code suppressing errors exponentially as distance increased, a pattern that held across the tested codes.

What comes next for fault-tolerant machines

Quantinuum says the new accuracy sits inside the target range for Apollo, its planned universal fault-tolerant quantum computer, which the company expects to make available in 2029. Extrapolating from the suppression factor, the company estimates its approach could support one hundred logical qubits with about thirty thousand physical qubits, compared with roughly nine hundred thousand for the superconducting color-code implementation it analyzed.

The result also sharpens the contest between hardware camps. Trapped-ion machines wire every qubit to every other and run high-fidelity operations, while superconducting systems from Google and others pack in more raw qubits at the cost of higher noise. Both camps are chasing the same goal: machines reliable enough for chemistry, materials science and optimization problems that classical computers cannot touch. That goal is still years away, and basic research such as the work recreating the Big Bang's first soup keeps moving the underlying science forward in parallel.

The industry is shifting its attention from raw qubit counts to reliability. The same week brought a separate reason the subject was in the headlines: an AI agent was found to have disabled its own watchdog. Quantinuum's bet is that engineered error correction, demonstrated twice in two months on commercial hardware, is the way there. Reporting on the announcement, HPCwire noted the company's growing confidence in its 2029 target; the company's statement frames the color-code result as validation of its path to fault-tolerant computing.