The 98 qubit trapped ion machine, where each qubit is a single atom acting as a fragile quantum bit, scores high on the two metrics that now matter most: accuracy, and the ability for every qubit to talk directly to every other qubit (all to all connectivity).
For decades, the quantum computing race has been scored the wrong way. A press release announces a new qubit count, the number goes on the scoreboard, and a 98-qubit machine sounds bigger than a 56-qubit one. A new result from Quantinuum, the British-American company behind the Helios trapped-ion processor, suggests the field is finally switching to a more honest metric. The number that matters now is how accurately each qubit does its job, and how tightly the qubits can talk to one another.
Helios is described in a peer-reviewed Nature paper as a 98-qubit trapped-ion quantum computer whose performance pushes beyond what classical machines can easily simulate, according to coverage in Scientific American. It succeeds Quantinuum's previous System Model H2, which had 56 qubits. The doubling is real, but the more interesting story is what each of those qubits can now do.
For non-beat readers, a qubit is the quantum equivalent of a computer bit: it can be 0, 1, or a fragile combination of both at once. That superposition is the trick that lets quantum computers explore many calculations in parallel. The catch is that qubits are easily disturbed by stray heat, vibration, or stray electric fields. "Trapped-ion" describes how Helios tames that fragility. Each qubit is a single charged atom suspended by electric fields in a vacuum and chilled to a hair above absolute zero, then nudged with precisely tuned laser pulses. The hardware is the opposite of a rack of server chips, and the operating conditions read like a physics experiment, because they are.
That control buys two things. First, accuracy. Oxford Physics reports that the platform behind Helios has set a new benchmark for qubit operation accuracy, with the precise gate-fidelity numbers still best cross-checked against the peer-reviewed paper before being treated as fixed. Second, connectivity. Trapped-ion qubits on this hardware can interact directly with any other qubit on the chip, not just immediate neighbors. Most superconducting processors, the rival technology inside machines from IBM and Google, can only entangle nearest neighbors and need extra operations to route information across the chip. All-to-all connectivity cuts the depth and error cost of many algorithms, which matters once a calculation is long enough that small per-operation errors compound into garbage output.
This is the yardstick shift. Raw qubit count is easy to advertise and hard to compare across architectures, because a superconducting qubit and a trapped-ion qubit are very different beasts. Fidelity and connectivity are harder to summarize in a press release but more directly tied to whether a machine can do useful work. A quantum computer that runs for a microsecond with very high gate fidelity can outperform one that runs for a second with mediocre fidelity on the kind of deep, structured calculations that classical machines struggle with, even if the second machine advertises more qubits.
The classical-simulation boundary is the other piece. Helios's 98-qubit result sits at the edge of what a top-tier classical supercomputer, fed the right algorithms and the right amount of memory, can still check against. Beyond that boundary, classical verification gets harder, and the quantum machine starts to do work that is genuinely outside the reach of today's simulation tools. Crossing that line is not the same as solving a useful problem. It is closer to crossing the threshold where the answer stops being easy to double-check the old-fashioned way. The Conversation's explainer frames the same point: a high-fidelity result at this scale is a real technical milestone and a real signal of progress, not a guarantee of near-term commercial payoff.
The application claims travel with the standard quantum caveats. Promised uses include simulating new materials and chemistry, solving optimization problems that are currently intractable, and breaking or hardening cryptographic systems. All three depend on being able to run very long, very accurate quantum programs, which is precisely the gap Helios is meant to narrow. None of them are next-quarter deployments. Useful fault-tolerant quantum computing still requires further improvements in error correction, logical qubit overhead, and the software stack that compiles abstract algorithms into the specific laser pulses a trapped-ion machine understands. The general theoretical framing for what error correction requires is summarized in quantum information course notes from the University of Cologne.
The honest version of the story is that quantum computing has promised to change the world for thirty years and has delivered several genuine technical milestones, but it has not yet changed any commercial workflow in a way most users would notice. Helios does not break that pattern. What it does is sharpen it: the field is starting to grade itself on the right thing, and the grade is going up. The next signal to watch is whether the same yardstick survives contact with superconducting rivals and other trapped-ion teams, and whether the fidelity and classical-simulation gains hold up as the machines scale beyond a hundred qubits.