The qubit leaderboard that changes unit between bars
Showing the misleading chart
Five flagship quantum computers, five published qubit counts, one zero-based linear axis: D-Wave’s Advantage2 at 4,400+ against Quantinuum’s Helios at 98, a 45× lead. Every figure is the maker’s own. But a D-Wave qubit is a node in an analog annealer that runs no gate circuits, an IBM Condor qubit is a scaling test article, and a Helios qubit is one of 98 that yield 48 error-corrected logical qubits — the unit a fault-tolerant algorithm is actually written in.
01The claim
The qubit race has a clear leader. Five companies, five flagship machines, one number each — the qubit count in the manufacturer’s own launch announcement, nothing estimated and nothing modelled, on a single zero-based linear axis with no log scale and no truncation. D-Wave’s Advantage2 carries more than 4,400 qubits. Atom Computing’s array holds 1,180 and IBM’s Condor 1,121, so three machines are now past a thousand. Google’s Willow is at 105 and Quantinuum’s Helios at 98, still in three figures. That is a 45× spread across the field. You don’t need any quantum mechanics to read this chart: one machine has 4,400 qubits and one has 98.
02The trick
Every count is accurate and every axis is honest. The bars start at zero, the scale is linear, the numbers are printed beside the lengths, and the disclosure is even on the slide — in grey, under the last card: “Qubit counts are not comparable across architectures.” The trouble is that the axis says “qubits”, and the word changes meaning between rows. A bar chart makes exactly one promise, that its lengths are commensurable because each is so many of the same thing, and this chart quietly breaks it three times. D-Wave’s 4,400 are annealing qubits: nodes in an analog machine that settles into the minimum of an energy landscape. It runs no gate circuits at all, which D-Wave does not dispute — its own site keeps annealing and gate-model on separate pages, describing gate-model machines as a different thing still to be commercialised. Second, within the gate-model rows, a count of hardware is not a count of working computer. IBM’s own roadmap says Osprey and Condor were built to “test the limits of single-chip processors and controlling large-scale quantum systems”; the processor IBM actually put in front of users at that same December 2023 announcement was Heron, at 133 qubits. Atom Computing’s 1,180 is the number of atoms loaded into a 1,225-site array, announced in October 2023 as a scaling milestone, with no algorithm published across the full array. Third, and largest, a physical qubit is not a logical one. Google spent 101 of Willow’s 105 physical qubits building a single distance-7 surface code, and Quantinuum’s Helios turns 98 physical qubits into 48 error-corrected logical ones. So the chart’s 45× lead is not an exaggeration of a real gap: it is a ranking of a quantity that does not exist. (This exhibit is our own demonstration, assembled from each maker’s published spec sheet in the style of a market-overview slide — no vendor publishes these bars together, and each states its own architecture plainly.)
03The fix
Split the chart along the thing that changed. Three panels, three axes, one unit each. Annealing qubits: D-Wave Advantage2, 4,400+, on an axis to 5,000 — alone in its panel, because nothing else in the field is that kind of machine. Physical gate-model qubits, on an axis to 1,250: the Atom Computing array at 1,180 and Condor at 1,121, both greyed as scaling demonstrations, then Heron at 133, Willow at 105 and Helios at 98. Error-corrected logical qubits, on an axis that runs to 50 rather than 5,000: Helios 48, the 56-qubit H2 with Microsoft 12, Willow 1, none published for Condor or the Atom Computing array, and not applicable for the annealer, which runs no code to correct. Read that way the leaderboard inverts. The shortest bar on the original — 98 physical qubits — is the machine with the most logical qubits and the best published two-qubit gate error, 0.079% with all-to-all connectivity, against Willow’s 0.33% ± 0.18% at an average connectivity of 3.47. Three panels is not a failure to combine them; it is the finding, stated plainly, and the empty space between the axes is where the comparison stops. None of this is a knock on any of the machines: an annealer is a real tool for real optimisation problems, Condor did the scaling job IBM built it for, and Willow’s single logical qubit is the first quantum memory to beat its own hardware, holding its state 2.4× longer than the best physical qubit feeding it. The habit is smaller than any of that, and it costs two seconds on any chart: read the unit under the axis, then ask whether that word means the same thing in every row.