Quantinuum said on October 19, 2026, that an error-corrected logical qubit on its Helios trapped-ion computer had an error rate of 5 in a million per round of error correction, a result the company described as a world record. Quantinuum published the figures in a preprint, “Quantum error correction well below the color code threshold,” by 23 researchers at its sites in Broomfield, Colorado, and London. The paper had not been peer reviewed at the time of the announcement.
The team measured a much steeper fall in logical errors with code size than Google has reported on superconducting hardware. With each step up in code size, Quantinuum’s logical errors fell about 12-fold, against 1.56-fold in Google’s color-code experiment, first posted in December 2024, and 2.14-fold for the surface code on Google’s Willow chip.
Quantinuum’s team showed that suppression for one stored logical qubit, with the error decoding done after the runs. It has not yet run logical gates at the largest code size, which needs more qubits than Helios has.
The experiment
The researchers encoded one logical qubit in triangular color codes of distance 3, 5, and 7. A code with a larger distance can correct more errors but needs more qubits: the three codes use 13, 37, and 73 physical qubits, including measurement qubits.
They ran 20 rounds of syndrome extraction, the repeated parity checks that detect errors, on Helios, a 98-qubit machine that uses barium-137 ions. The logical error per round fell from 5.5 in 10,000 at distance 3 to 3.8 in 100,000 at distance 5 and to 5 in a million at distance 7.
The researchers fitted a suppression factor, written Λ, of 12. Logical errors fell about 12-fold each time the code distance grew by 2. In its 2024 Nature paper on Willow, Google reported a Λ of 2.14 for the surface code.
At distance 7, the team recorded three logical failures in 29,000 runs of 20 rounds each. The authors reported a 68% confidence range of 3 to 9 errors in a million per round.
The researchers also benchmarked logical two-qubit Clifford gates, a standard class of logic operations, on pairs of encoded qubits. The error per gate was 7 in 10,000 at distance 3 and 7 in 100,000 at distance 5, compared with 9.5 in 10,000 for the same circuits on unencoded physical qubits. Gate circuits at distance 7 need more qubits than are available on Helios, the authors wrote.
Leakage and decoding
The authors attributed the scaling to two measures against leakage, in which an ion leaves the two energy levels that encode the qubit. Quantinuum added a mid-circuit “repump,” a sequence of laser pulses that returns leaked ions to the qubit levels, and paired it with a circuit that regularly swaps each data qubit with a freshly reset measurement qubit. According to the paper, the repump lowered logical error rates fourfold at distance 3 and eightfold at distance 5 compared with the swap circuit alone.
Decoding, the classical computation that infers which errors occurred from the parity checks, was done after the experiments had finished. For the headline figures, the authors used Tesseract, a search-based decoder published by Google Quantum AI researchers. The authors described Tesseract as near-optimal but generally too slow for real-time use.
Two neural-network decoders the team also tested averaged 0.22 to 0.52 milliseconds of computing time per round on an NVIDIA H100 graphics processor. The authors wrote that real-time decoding at these speeds is already viable.
Company claims and outside reaction
Quantinuum said the logical memory error was 10 times lower than the record it set in September with its Helix code on the same machine, and 100 times lower than the best published result on other hardware. In a blog post, the company said a machine at the scale of Apollo, the universal fault-tolerant computer it plans to offer in 2029, could reach fault-tolerant quantum advantage with about 30 times fewer physical qubits than other leading approaches. Quantinuum sketched that calculation in the blog post, using Google’s color-code result as its baseline, but has not published a paper on it.
Chief executive Rajeeb Hazra said in the company’s press release that the result showed Quantinuum has the error-correction architecture and trapped-ion technology needed for universal fault tolerance. In the same release, Robin Blume-Kohout, co-director of the Quantum Performance Laboratory at Sandia National Laboratories, who was not involved in the work, described the logical qubit as “nearly 100x better than its component physical qubits.”
Craig Gidney, a quantum software engineer at Google Quantum AI, wrote on X that the scaling implied practical error rates at a code distance of about 21. He also called a 20-millisecond round time “brutally slow,” then corrected the figure to 200 milliseconds. The preprint does not state a round time in its text. A round on Willow takes about 1.1 microseconds.
Next steps
Quantinuum said the accuracy it measured is within the target range for Apollo. The authors described the natural next step as a below-threshold computation that includes non-Clifford T gates, with decoding in real time.
