Quantum Essentials · Module 2 · Chapter 11 of 17
Which quantum computing approach is winning?
Every few months a headline announces a quantum record. How do you tell which machine is really ahead?
The short answer (tap to open)
No single approach is winning, because they lead on different things. Trapped ions lead on accuracy, superconducting circuits on speed and error-correction evidence, neutral atoms on scale. Early "quantum supremacy" claims were often matched later by ordinary computers. To judge any claim or roadmap, ask about error rates, logical qubits, independent checks and whether companies kept earlier promises.
Predict
Which single number tells you most about how useful a quantum computer is?
After this page you can
- compare quantum computers on more than qubit count;
- explain what the main benchmarks measure;
- read a "quantum advantage" headline critically;
- read a company roadmap, and say what to check.
Part A · The scorecard
What should you compare?
- Qubit count: how many physical qubits.
- Accuracy: how often a two-qubit operation succeeds (99.9% means one failure in a thousand).
- Lifetime: how long qubits keep their state, compared with how long an operation takes.
- Connectivity: which qubits can talk directly (chapter 9's puzzle).
- Speed: how many operations per second.
- Logical qubits: how many error-corrected qubits it can run, and how reliable they are.
- Scalability: a believable route to millions of qubits, with manufacturing and cooling that work at that scale.
- Availability: can customers actually use it, over the cloud, today?
- Superconducting
- Photonic
- Neutral atoms
- Silicon spin
- Trapped ions
- TopologicalNot scored
- AnnealerNot scored
With these weights, Superconducting tops the ranking.
Scores simplify dated figures. Real comparisons need the full numbers.
The scores, and four example weightings
| Accuracy | Scale | Speed | Manufacturing | |
|---|---|---|---|---|
| Superconducting | 4 of 5: 99.93% best two-qubit accuracy (an IQM-led team, 2025) | 3 of 5: 105 to 120 qubits (2024 to 2025) | 5 of 5: Gates tens of billionths of a second long | 4 of 5: Chips made with tools the chip industry already has |
| Trapped ions | 5 of 5: 99.921% across all pairs (Quantinuum Helios, November 2025) | 2 of 5: About 100 qubits (2025) | 1 of 5: Gates much slower than superconducting ones | 3 of 5: Trap chips made with standard semiconductor processes |
| Neutral atoms | 3 of 5: 99.56% in a commercial system (Atom Computing, 2024) | 5 of 5: 6,100 atoms held at once (Caltech, September 2025) | 2 of 5: Each round of operations is slow | 2 of 5: Built from lasers, optics and vacuum chambers |
| Photonic | 3 of 5: 99.22% for fusions (PsiQuantum Omega, February 2025) | 2 of 5: 12 qubits over 35 chips (Xanadu Aurora, January 2025) | 3 of 5: Light is fast, but gates succeed only some of the time | 5 of 5: Chips made in existing semiconductor foundries |
| Silicon spin | 3 of 5: Above 99% (Diraq and imec, 2025) | 1 of 5: A handful to a dozen or so qubits | 3 of 5: Still early; few comparable figures yet | 5 of 5: Made in ordinary chip factories |
| Topological | Not scored: no confirmed working qubit (October 2026). | |||
| Annealer | Not scored: a different kind of machine, for optimisation only. | |||
- Weight accuracy only: With these weights, Trapped ions tops the ranking. Trapped ions, Superconducting, Neutral atoms, Photonic, Silicon spin.
- Weight scale only: With these weights, Neutral atoms tops the ranking. Neutral atoms, Superconducting, Trapped ions, Photonic, Silicon spin.
- Weight speed only: With these weights, Superconducting tops the ranking. Superconducting, Photonic, Silicon spin, Neutral atoms, Trapped ions.
- Weight manufacturing only: With these weights, Photonic and Silicon spin share the top of the ranking. Photonic, Silicon spin, Superconducting, Trapped ions, Neutral atoms.
Scores simplify dated figures. Real comparisons need the full numbers.
Where does each lead today?
As of October 2026:
- Trapped ions: the most accurate operations (Quantinuum's Helios, 99.921% across all pairs, November 2025) and the most error-corrected logical qubits run so far (48).
- Superconducting: the fastest operations, and the clearest evidence that error correction gets better as codes grow (Google's Willow, 2024).
- Neutral atoms: the most qubits held at once (6,100 at Caltech, 2025) and a full set of fault-tolerance ingredients on 448 atoms (Harvard, MIT and QuEra, 2025).
- Photonic: a bet on manufacturing and networking, with large funding and few large demonstrations yet.
- Silicon spin: early, but built in real chip factories.
- Topological: unproven and disputed (chapter 10).
- Annealing: a separate kind of machine for optimisation.
Part B · Benchmarks, decoded
What do the benchmark names mean?
Companies quote several standard tests:
- Quantum volume: the largest "square" test circuit (as many layers as qubits) a machine runs reliably. Record: 2²⁵, about 33.5 million, Quantinuum, September 2025.
- Algorithmic qubits (#AQ): IonQ's own measure of how many qubits are usable across a set of programs. Record: 64, IonQ, September 2025.
- Logical qubits: how many error-corrected qubits run at once. Records: 48 fully error-corrected, and 94 with error detection only, both Quantinuum, November 2025.
Treat any benchmark a company designed itself with extra care, and prefer measures other groups can repeat.
Benchmark to meaning
Qubit count
Quantum volume
Algorithmic qubits (#AQ)
Logical qubits
Random circuit sampling
How many physical qubits a machine has
The answers, in words
- How many physical qubits a machine has: Qubit count. The most quoted figure, and the most misleading on its own.
- The largest "square" test circuit (as many layers as qubits) a machine runs reliably: Quantum volume. Record: 2²⁵, about 33.5 million, Quantinuum, September 2025.
- One company's own measure of how many qubits are usable across a set of programs: Algorithmic qubits (#AQ). Record: 64, IonQ, September 2025.
- How many error-corrected qubits run at once: Logical qubits. Records: 48 fully error-corrected, and 94 with error detection only, both Quantinuum, November 2025.
- Raw power on a task chosen to suit the machine: Random circuit sampling. Random circuit sampling shows raw power on a task chosen to suit the machine. It isn't useful work.
Part C · Quantum advantage: the claims and the comebacks
What is "quantum advantage"?
A quantum computer doing something no ordinary computer can do in reasonable time. In 2012 John Preskill called this quantum supremacy; many now prefer "quantum advantage". There are two kinds of claim: advantage on a made-up task chosen to suit the machine, and advantage on a useful task. Companies have claimed the first several times. The second is the real prize.
How have the claims held up?
Did it hold?
Still stands, so far
Matched by ordinary computers
Contested
Too early to say
2019, Google Sycamore: 53 qubits; Google estimated 10,000 years for a supercomputer.
The answers, in words
- 2019, Google Sycamore: 53 qubits; Google estimated 10,000 years for a supercomputer.: Matched by ordinary computers. IBM estimated about 2.5 days within days; by 2022 and 2024, improved methods on ordinary hardware matched or beat it.
- 2020 to 2023, USTC Jiuzhang (light).: Matched by ordinary computers. In 2024 a classical method matched or beat its later versions and Xanadu's similar Borealis.
- 2024, Google Willow: a sampling task in 5 minutes, against Google's estimate of 10²⁵ years classically.: Still stands, so far. Still stands, so far, as a company estimate on a made-up task.
- March 2025, D-Wave: simulating a magnetic material beyond ordinary computers.: Contested. A 2026 Science paper matched many cases; D-Wave says its hardest ones weren't.
- October 2025, Google "Quantum Echoes": a measurement 13,000 times faster than the Frontier supercomputer, and checkable.: Still stands, so far. Still stands, so far. No ordinary-computer reproduction yet.
- 2026: IBM and partners post three advantage claims to an open public tracker; USTC reports Jiuzhang 4.0 with 3,050 photons.: Too early to say. Too early to say.
The pattern: early claims on made-up tasks often fall to cleverer ordinary methods. That's how science should work. Claims on useful tasks are only now appearing.
Part D · How to read a roadmap
Do companies hit their roadmaps?
Sometimes. Some recent examples, all dated:
- Delivered: Quantinuum promised Helios in 2025 and delivered it in November 2025. IBM reported its fast error-correction decoder a year early.
- Late: Rigetti promised a 100-plus-qubit system at 99.5% accuracy by the end of 2025; its 108-qubit system arrived in April 2026 at about 99.1%.
- Moved: Pasqal's 2024 roadmap promised 10,000 qubits in 2026; it now says 10,000 by the end of 2029.
- Not yet: Google's roadmap milestone of a "long-lived logical qubit" (one error in a million operations) is still ahead.
QubitCo's roadmap (a made-up company)
Year: 2025
2025 · 100 qubits
Arrived in 2025.
2026 · Errors of 1 in 1,000
Not arrived yet.
2027 · First logical qubit
Not arrived yet.
2028 · 10,000 qubits
Not arrived yet.
2029 · 100 logical qubits
Not arrived yet.
The roadmap, in words
QubitCo's roadmap (a made-up company)
- 2025: 100 qubits. Arrived in 2025. Delivered on time.
- 2026: Errors of 1 in 1,000. Arrived in 2027. Late.
- 2027: First logical qubit. Arrived in 2027. Delivered on time.
- 2028: 10,000 qubits. Not arrived yet. Not yet.
- 2029: 100 logical qubits. Not arrived yet. Not yet.
Real promises, real outcomes
Delivered
Late
Moved
Not yet
Quantinuum promised Helios in 2025.
The answers, in words
- Quantinuum promised Helios in 2025.: Delivered. Delivered in November 2025.
- IBM's fast error-correction decoder.: Delivered. Reported a year early, in November 2025.
- Rigetti promised a 100-plus-qubit system at 99.5% accuracy by the end of 2025.: Late. Its 108-qubit system arrived in April 2026 at about 99.1%.
- Pasqal's 2024 roadmap promised 10,000 qubits in 2026.: Moved. It now says 10,000 by the end of 2029.
- Google's roadmap milestone of a "long-lived logical qubit" (one error in a million operations).: Not yet. Still ahead.
What should you ask about any roadmap?
- Physical or logical? "A million qubits" means little without saying which.
- What error rate? Logical qubits at one error in a thousand, or in a million?
- Shown, or promised? Is there a peer-reviewed result, or only a press release?
- Track record? Did the company hit its last three milestones?
- Independent check? Has anyone outside the company verified it?
QubitCo to deliver 1 million qubits by 2030! Today we unveiled our roadmap to the world's most powerful quantum computer.
The answers, in words
QubitCo to deliver 1 million qubits by 2030! Today we unveiled our roadmap to the world's most powerful quantum computer.
- Physical or logical? It just says qubits.
- What error rate? No error rate anywhere.
- Shown, or promised? It answers this one: it's a promise, for 2030.
- Track record? Nothing about earlier milestones.
- Independent check? No outside group is mentioned.
Now try it on the next real headline you see.
Is anyone checking independently?
Yes. The US defence research agency DARPA runs the Quantum Benchmarking Initiative, which tests whether any approach can build an industrially useful quantum computer by 2033. In November 2025 it moved eleven companies to its second stage, across superconducting, ion, atom, photonic and silicon approaches. In October 2026 it moved four of them into its final stage, where its own team tests the hardware: Atom Computing (neutral atoms), Diraq (silicon), IBM (superconducting) and IonQ (trapped ions). They join Microsoft and PsiQuantum, and DARPA expects more to follow. It's an evaluation programme, not a race with one winner.
Part E · So, who's winning?
So, who's winning?
Nobody yet, and anyone who tells you otherwise is selling something. Today's honest summary:
- Several approaches have shown working error correction on small numbers of logical qubits.
- Company roadmaps aim for the first machines with hundreds of good logical qubits around 2028 to 2030 (IBM's Starling, Quantinuum's Apollo, QuEra's Libra, all company plans).
- The first clearly useful results are most likely in simulating physics and chemistry (chapter 14).
- It's quite possible that more than one approach succeeds, for different jobs, or that machines combine them.
Myths and media
Media check
Right, half right or wrong? Call it first.
Card 1 of 3
A 2019 headline: "Google achieves quantum supremacy."
Check
Three quick ones
Collect
Your aha card
No winner yet, and no single score.
Judge by error rates, logical qubits, independent checks and kept promises, not by qubit counts or headlines.
Go further
- 11a · Real-world caseReading a roadmap: promised against delivered
- 11b · A step deeperBenchmarks, and why qubit count misleads
- 11c · People storyJohn Preskill: the physicist who named "quantum supremacy" (2012) and "NISQ" (2018)
Company information is for education only. It isn't investment advice. Figures carry dates.