part of EQUORA Institute
HUEN
EQUORA Institute · Closed data set · Evidence quality
Mental Fast Food™ · 20 · Research byproduct

The qubit count went down and the machine got better

IBM's 2023 processor had 1,121 qubits. The 2025 one has 120.

One vendor. Two years. The smaller number is the better machine.

The same number appears in every quantum computing headline, because qubit count fits on one line, is a whole number, and tends to grow year on year. Machine performance, however, is decided elsewhere, and the vendors treat it that way themselves: after 2023 IBM moved down in device count while the depth of runnable circuits went up.

This page does four things. It ranks eight quantum technologies across five different measures, shows that the logical qubit replacing them already misleads in the same way, measures the EQUORA Institute's own 2025 overview against the present state, and closes with a filter that works on any future quantum news item.

01 · The number that travels
Why the qubit count fits in a single headline

A qubit means the system contains a two-state quantum object that can be controlled and read out. That definition omits everything that makes a machine compute: how long it holds its state, how precisely an operation between two qubits can be performed, and how many other qubits it can interact with directly. A qubit that forgets in a microsecond and one that remembers for a second both count as one.

This is how a 98-qubit trapped-ion machine can do more useful work than an annealer with several thousand qubits, and why the field started swapping the old number for a different one. A logical qubit is built from many physical qubits whose errors correct one another, so it is the quantity in which the actual length of an algorithm can be measured.

The swap shows up in the vendor roadmaps too. IBM's 2023 Condor processor carried 1,121 physical qubits and never received broad access; the Nighthawk introduced in 2025 has 120 qubits, 360 across three modules, and runs deeper circuits. The Starling promised for 2029 targets 200 logical qubits and 100 million operations.

TakeawayThe qubit count fits in a single headline. What the machine can do emerges a few numbers later.

02 · The ranking
Five measures, four different winners

Eight quantum technologies compete, and each leads on a different metric. Switch the measure and watch how often the platform in first place changes. In the fainter rows the publicly available data is weak or hard to compare, so their position is uncertain.

Sort by measure

TakeawayA ranking that inverts when the measure changes is a statement about the measure, less so about the technology.

03 · The logical qubit
The same machine reports 94 and 48

As the field moved to logical qubits, announcements followed, and the number already plays the role the physical count once did. Code distance makes the difference: a short-distance code notices the error without being able to correct it, distance-4 corrects, and running Shor at length requires a distance above 25.

Four claims follow, phrased alike. Tap any of them to see what it covers.

TakeawayWhen you see a logical qubit count, code distance is the number missing from the sentence.

04 · Correction ledger
Eleven items on our own document

In 2025 the EQUORA Institute produced a Hungarian-language overview of the eight quantum technologies for a conference roundtable. The ledger below measures it against the state of things in August 2026. Four items were physics errors at the moment of publication, six went stale, and one came true.

The forecast that held is on the ledger deliberately, since a correction ledger that lists only errors distorts in the same way as what it criticises.

TakeawayThe worth of an overview rests on whether its author is willing to go back for it.

05 · An open dispute
Quantum advantage always holds relative to something

In March 2025 D-Wave published in Science that its Advantage2 annealer, with more than 5,000 qubits, had simulated the real-time dynamics of disordered spin glasses, a task it called out of reach for classical machines. In May 2026 physicists from the Flatiron Institute and Boston University published, also in Science, that they had reproduced the same dynamics using three-dimensional tensor networks and a belief propagation method from 1982, across several lattice geometries, on ordinary workstations.

D-Wave's formal response of 26 May holds that this does not overturn the result, because the classical method fails to scale on the hardest instances and the higher-order physical observables. This page leaves the dispute open, since that is where the literature currently stands.

The lesson follows from the shape of the dispute. Quantum advantage is always a claim measured against a particular classical algorithm, so progress on the classical side can revise it retroactively. An advantage stated without a named classical baseline has not yet been measured at all.

What this page does not claim

Quantum computing is a serious and fast-moving field, and the numbers above record real engineering results. This page is about what each number is good evidence for, so it is not about which platform will win: nobody knows that at present, and several of the eight may yet arrive.

Nor does the ledger say the 2025 overview was bad. A snapshot of a fast-moving field stays useful even when parts of it go stale within a year.

TakeawayA number says something once you can name what it is relative to.

06 · The reading filter
Three questions for any quantum headline

The three questions below fit any quantum news item, and most of the answers can be found in the cited paper within minutes. Where the item itself fails to answer one, that is information in its own right.

The filter works outside quantum too. For any technology announcement one can ask whether the published number measures the outcome or merely a component, whether there is a quality threshold behind the unit, and better than what the announced result is.

TakeawayWhere an item states its own denominator, the number can be read. Without it, the number is merely large.

07 · References
Primary sources
  1. Google Quantum AI and Collaborators: Quantum error correction below the surface code threshold. Nature 638, 920–926 (2024). arXiv:2408.13687
  2. QuEra Computing, Harvard, MIT: error-corrected logical qubits on neutral-atom hardware. Nature (January 2026)
  3. Quantinuum: Helios — a 98-qubit trapped-ion quantum computer. arXiv (November 2025)
  4. King, A. D. et al.: Beyond-classical computation in quantum simulation. Science (March 2025)
  5. Flatiron Institute and Boston University: three-dimensional tensor network algorithms for quantum annealing dynamics. Science (May 2026)
  6. D-Wave Quantum Inc.: formal response to the classical simulation result (26 May 2026)
  7. Microsoft Azure Quantum: Majorana 2 — topological gap and parity lifetime, Build 2026 (June 2026)
  8. IBM Quantum: Nighthawk announcement and the Starling roadmap (November 2025)
  9. Papp László (Pölö): Quantum technology overview — the eight current quantum technologies explained. EQUORA Institute, 2025.
Research provenance
This page comes out of a research programme at the EQUORA Institute and records one state of it, in place of a closed institutional position. That state rests on the results available at the moment of publication; anything published later is included only where the page was updated, and the date shows this. AI takes part throughout the research as a thinking partner; responsibility for interpretation and publication remains human.
Published: 1 August 2026
Papp László · EQUORA InstituteHow we research →