Quantinuum's Helios Sets a New Accuracy Watermark: 98 Qubits, All-to-All Connectivity and Why That Matters More Than the Headline Number

[Cheeky Blake]

A paper published in Nature this week describes Helios, the new trapped-ion quantum computer from Quantinuum, as a 98-qubit processor with error rates and performance that push beyond what classical machines can easily simulate. The qubit count alone is notable, nearly double the 56 qubits of the previous record holder, Quantinuum's own System Model H2. But the more important story is accuracy rather than scale. The paper reports an average single-qubit gate error rate of about 2.5 in 100,000 and a two-qubit gate error rate of about 7.9 in 10,000, putting Helios among the best demonstrations recorded anywhere, comparable to Oxford's own world record of around 5 in 10,000.

The distinction matters because quantum operations are cumulative. A useful algorithm might require thousands or millions of operations chained together, and a small error rate per operation compounds quickly across a long calculation. Lower error rates mean more complex, more meaningful calculations become possible before the quantum information collapses. Helios also features all-to-all connectivity, meaning any qubit can in principle interact directly with any other rather than only its nearest neighbours, a property especially valuable for algorithms whose interaction patterns do not map neatly onto a fixed grid.

The hardware achieving this uses barium ions held in a quantum charge-coupled device architecture, essentially a tiny railway system where ions are stored in memory regions and physically shuttled into operation zones for laser-controlled quantum gates to be performed. Software makes real-time decisions about ion routing and gate ordering as a program runs, a sign that trapped-ion quantum computing is maturing from impressive laboratory components into something closer to a full computing system. The paper also reports Helios can run random circuit sampling tasks that are extremely difficult for classical machines to simulate, an important complexity benchmark, though distinct from solving an actually useful problem in chemistry, materials science or logistics.

The Conversation: A New Quantum Computer Sets a High Watermark For Accuracy - Are We on The Verge of a Big Breakthrough?thequantuminsider.com

[HitmanMatt53]

The framing of judging a race by runners at the starting line versus runners who finish is exactly right. Every quantum announcement for years has led with qubit count because it is the easiest number to put in a headline. Error rate at scale is the number that actually predicts whether a machine can do anything useful

[AustinTheory18]

All-to-all connectivity is the underrated part of this announcement. Most superconducting architectures are stuck with nearest-neighbour connectivity, meaning distant qubits need a chain of intermediate swap operations to interact, and each swap adds error. Trapped ions sidestepping that with physical shuttling is a genuine structural advantage

[DarkMatter24]

2.5 in 100,000 for single-qubit gates and 7.9 in 10,000 for two-qubit gates at 98 qubits simultaneously is the headline number that should travel further than it will. Getting low error rates on a handful of qubits is one thing. Maintaining it while scaling to 98 with all-to-all connectivity is a different engineering challenge entirely

[Courier53]

The QCCD railway metaphor is an useful way to picture what is happening physically. Ions sitting in storage, then being moved into an operation zone for a gate, then moved back, with a ring-shaped storage area and junction routing them. It is closer to a real computing architecture than most people imagine when they hear quantum computer

[Myles]

Random circuit sampling being a complexity benchmark rather than a useful-problem benchmark is the caveat that gets lost in most coverage of results like this. It proves Helios is doing something classical computers struggle to replicate. It does not prove Helios can simulate a new drug molecule or optimise a logistics network yet

[BigDog_Fan]

Software making routing and gate-ordering decisions dynamically while a programme runs is the detail that signals real systems engineering maturity. Early quantum demonstrations were static circuits decided in advance. Adaptive control based on mid-circuit measurements is what real algorithms with feedback loops will require

[RayOfLight87]

This being published the same week as Quantinuum's $15.7 billion Nasdaq debut is not a coincidence in terms of investor narrative, but the Nature peer review process moves independently of stock market timing. The science here would matter on its own merits whenever it published