Who has the best quantum computer right now?

Why is it so hard to name the best quantum computer?

There is no universal quantum-computer leaderboard today because the leading machines are being built to win different contests.

A raw qubit ranking would put D-Wave first with more than 4,400 annealing qubits, followed by Atom Computing’s gate-based AC1000 with more than 1,200 neutral-atom qubits. Quantinuum’s Helios would appear much farther down with 98 trapped-ion qubits. Yet those numbers describe very different machines. D-Wave’s qubits solve a specialized class of sampling and optimization problems, while Atom and Quantinuum can run general sequences of quantum gates.

Even among general-purpose systems, size tells only part of the story. Google’s Willow runs extremely fast gates on a fixed superconducting grid. Quantinuum moves ions around the processor and can directly connect any pair. Atom Computing starts with a huge neutral-atom array, although its operations are currently less accurate than those reported for Helios or Willow.

The experiments also pull in different directions. Google’s Quantum Echoes work produced a verifiable calculation that Google estimated was 13,000 times faster than the best classical approach. USTC’s Zuchongzhi 3 made a much larger speed claim for random-circuit sampling. Helios recently showed that a 98-qubit trapped-ion machine could run random circuits beyond practical classical simulation while keeping its component error rates low. Each result is impressive, but each asks the hardware a different question.

What should “best quantum computer” mean right now?

The best quantum computer today should be the strongest complete machine, rather than the system with one record-breaking specification.

We give the most weight to five things. The gates must be accurate. The machine needs enough qubits to run meaningful circuits. Those qubits should communicate without excessive routing. Published results should show that the whole processor works, not just a carefully selected corner of it. Researchers or companies also need a realistic way to use the system.

That definition rewards consistency. A computer that scores near the top across several categories is more convincing than one built around a single dramatic experiment. It also keeps future roadmaps separate from machines already operating.

Commercial value is still too early to settle the debate. No provider has publicly shown a valuable business problem that its quantum computer solves faster, better and more cheaply than every serious classical alternative. For now, hardware quality and demonstrated scientific capability remain the fairest tests.

Meaning of “best”	Current leader	What supports the claim
Best all-around general-purpose machine	Quantinuum Helios	Accuracy, full connectivity, logical computation, deep circuits and commercial access
Strongest verifiable beyond-classical algorithm	Google Willow	Quantum Echoes ran an estimated 13,000 times faster than the leading classical approach
Largest programmable gate-based machine	Atom Computing AC1000	More than 1,200 neutral-atom qubits
Strongest public platform	IBM Quantum	Large hardware fleet, broad access and the Qiskit ecosystem
Best specialized annealer	D-Wave Advantage2	More than 4,400 annealing qubits and a mature cloud service
Strongest raw sampling claim	USTC Zuchongzhi 3	An 83-qubit, 32-layer experiment with a claimed 10¹⁵ classical speed gap

If you want more recent data on this point, please see our latest quantum computing market report.

Does the quantum computer with the most qubits win?

The machine with the most qubits rarely wins today because a noisy qubit adds little once the circuit becomes too deep to finish correctly.

D-Wave’s Advantage2 has more than 4,400 qubits, but those qubits are designed for quantum annealing. We cannot compare them one-for-one with the programmable qubits inside Helios, Willow or AC1000. The machine is highly relevant for problems that fit D-Wave’s model, though its large qubit count says little about its ability to run a general quantum algorithm.

Atom Computing offers the more interesting comparison. AC1000 has over 1,200 programmable physical qubits, roughly twelve times the number in Helios. Atom reports single-qubit fidelity above 99.9% and two-qubit fidelity above 99.6%. Helios reports average fidelities of 99.9975% and 99.921%, respectively. The testing methods differ, so a direct decimal-by-decimal comparison would be misleading. The broad gap still shows why Helios can compete with a machine more than ten times its size.

Errors pile up quickly. In a simplified calculation where each operation succeeds independently, 1,000 gates at 99.921% fidelity would all succeed about 45% of the time. At 99.6%, the equivalent probability falls below 2%. Real circuits behave differently because errors can be correlated, detected or corrected, but the order of magnitude is revealing.

Atom has the larger canvas. Helios currently makes cleaner use of the space it has.

Which quantum computer has the best physical hardware today?

Quantinuum Helios currently has the strongest published combination of gate accuracy, connectivity and whole-system performance.

Helios contains 98 trapped barium-ion qubits. Ions are physically moved between memory and operating zones, allowing the computer to connect any pair of qubits. Its recent Nature paper reported average errors of 0.0025% for single-qubit gates, 0.079% for two-qubit gates and 0.033% for state preparation and measurement.

Google’s Willow is extremely close on accuracy and much faster per operation. Google reports 99.97% single-qubit fidelity, 99.88% entangling-gate fidelity and 99.5% readout fidelity across the 105-qubit chip. Its gates take tens to hundreds of nanoseconds, while trapped-ion operations are generally slower. Willow can therefore repeat circuits and collect samples at a remarkable rate. Google says its Quantum Echoes project involved around one trillion measurements.

Connectivity gives Helios a practical advantage. Willow’s qubits sit on a grid, so two distant qubits may need extra routing operations before they can interact. Helios can bring any chosen pair together. That flexibility reduces the number of gates needed for many programs, although physically moving ions also costs time.

The recent Helios paper strengthens the case because the component numbers predicted the performance of the complete machine. Quantinuum ran random circuits across its processor in a regime the authors described as beyond practical classical simulation. The system stayed accurate as it grew from Quantinuum’s earlier 6-qubit architecture to 98 qubits, which is exactly where many quantum designs begin to struggle.

If you want more recent data on this point, please see our latest quantum computing market report.

Who is leading the logical-qubit race?

Quantinuum leads in the number and variety of logical qubits used for computation, while Google leads in proving that stronger error correction can produce a more reliable quantum memory.

A logical qubit stores information across several physical qubits. The extra hardware lets the computer detect or correct mistakes before they destroy the calculation. The real goal is a logical error rate below the physical error rate, sustained across long and useful computations.

Google’s Willow reached a crucial milestone with a surface-code memory built from 101 physical qubits. When Google increased the code’s size, the logical error rate fell by a factor of about 2.14. The larger logical memory recorded an error rate of 0.143% per correction cycle. That result showed that adding more protection could genuinely improve reliability, rather than simply creating more places for errors to occur.

Quantinuum has gone wider. Its latest work extracted 48 error-corrected logical qubits and 64 error-detected logical qubits from Helios’s 98 physical qubits. Some configurations produced as many as 94 error-detected logical qubits. The company then used 64 encoded qubits for a quantum-magnetism simulation and reported error improvements ranging from roughly tenfold to one hundredfold over physical implementations in the tested circuits. A separate Nature study reported improvements ranging from 11 times to 800 times with smaller error-correcting codes.

These achievements solve different parts of the problem. Google has the clearest proof that a surface-code memory improves as more protection is added. Quantinuum has shown more logical qubits working together on actual computations. For the most capable machine currently available, we give Quantinuum the edge.

Did Google Willow prove it is the most powerful quantum computer?

Google Willow proved that it can beat classical supercomputers on an unusually demanding and verifiable algorithm, but that result alone does not make it the best machine overall.

Quantum Echoes ran on Willow and studied how a disturbance spreads through a complex quantum system. Google estimated that Willow completed the relevant calculation 13,000 times faster than the best known classical method on a leading supercomputer. The work appeared in Nature, and the result can be checked by repeating the protocol on comparable quantum hardware. That makes it more meaningful than a sampling experiment whose output is difficult to verify directly.

Google also ran a proof-of-principle molecular experiment on structures containing 15 and 28 atoms. The quantum results agreed with traditional nuclear magnetic resonance measurements and revealed additional information within the experimental setup. This remains early research, though it begins to connect the hardware with chemistry rather than an abstract benchmark alone.

The limitation is scope. Quantum Echoes was carefully designed around a type of dynamics that suits quantum hardware. It has yet to beat classical computing on a production chemistry problem, a commercial optimization workload or a widely used scientific calculation.

Willow owns the strongest verifiable algorithmic demonstration. Helios has produced a broader collection of high-quality results across benchmarking, programming, logical qubits and external use.

If you want more recent data on this point, please see our latest quantum computing market report.

Why does Helios still edge out Willow overall?

Helios beats Willow in our overall ranking because it is more complete and more usable today, even though Willow has the stronger flagship experiment.

Helios offers full connectivity, highly accurate gates, mid-circuit measurements and classical control while a quantum program is still running. Developers can use loops, conditional branches and dynamically allocate qubits. These capabilities make it behave more like a programmable computer and less like a fixed experimental sequence.

Access also separates the two systems. Quantinuum makes Helios available through its cloud service and offers dedicated installations. Customers announced at launch included Amgen, BMW, JPMorganChase and SoftBank. Google has started opening Willow to selected research collaborations, including a programme with the UK’s National Quantum Computing Centre, although it remains less broadly available.

Willow remains stronger in speed, surface-code error correction and the Quantum Echoes result. Helios has the better overall balance: accurate hardware, flexible programs, encoded computations and a realistic route for users to run their own work.

That is enough to put Quantinuum first without pretending one benchmark settles the whole contest.

Does IBM have the best quantum-computing platform?

IBM currently has the strongest public quantum platform, although its best individual processor still trails Helios in our hardware ranking.

IBM’s advantage comes from availability and continuity. Its platform offers a fleet of processors with more than 100 qubits, a mature cloud interface and Qiskit, one of the field’s most widely used software environments. Researchers can move from tutorials to real hardware without joining a small invitation-only programme.

Nighthawk is IBM’s latest major hardware step. The processor has 120 qubits and 218 couplers, giving it denser connections than the previous Heron design. IBM first made a Nighthawk system available in the United States and later added a second in Europe. The European machine cut the median two-qubit gate time from 138 nanoseconds to 68 nanoseconds while retaining a median coherence time close to 350 microseconds.

Nighthawk remains an exploratory processor. IBM’s documentation lists temporary limitations around dynamic circuits and mid-circuit measurements. The company’s goal of running up to three linked 120-qubit modules and 7,500 gates belongs to its roadmap, not its demonstrated current performance.

For a university or company that values reliable access, documentation and a large developer community, IBM may be the sensible choice. Pure hardware leadership is a tougher claim, and the latest evidence still favors Quantinuum.

If you want more recent data on this point, please see our latest quantum computing market report.

Is Atom Computing’s 1,200-qubit machine already ahead?

Atom Computing has solved more of the scaling problem than any other gate-based contender, but AC1000 still lacks the full-machine evidence needed to rank first.

AC1000 uses more than 1,200 neutral atoms as qubits. Lasers hold and move those atoms without the dense wiring required by superconducting chips. The machine supports mid-circuit measurement, reset, qubit reuse and real-time decisions. Its size is already well beyond the 98 to 156 physical qubits found in many leading trapped-ion and superconducting systems.

Atom has also made real progress on error correction. In recent work, the company repeatedly extracted error information from a toric code for as many as 90 cycles. It replaced lost atoms from a nearby reserve and showed lower logical error rates when it used the larger of two tested codes during shorter runs. The experiment tackles one of the biggest weaknesses of neutral atoms: atoms can disappear from the array during a long calculation.

The missing piece is a difficult calculation that uses a large share of those 1,200 qubits. We have yet to see AC1000 run a deep, broadly programmable circuit across hundreds of qubits with accuracy comparable to Helios, or a verifiable beyond-classical algorithm comparable to Quantum Echoes.

The hardware scale is genuine. The usable computational scale remains harder to judge.

Will neutral-atom quantum computers take the lead next?

Neutral atoms look like the strongest candidates to take the hardware lead, provided they can raise gate accuracy without sacrificing their huge qubit advantage.

The attraction is easy to understand. Atom Computing already has a four-digit qubit count. Other neutral-atom systems can rearrange their qubits, replace missing atoms and create different interaction patterns without redesigning the physical chip. This gives the architecture a cleaner path toward large error-correcting codes.

Accuracy remains the bottleneck. Atom reports two-qubit fidelity above 99.6%, while Helios averages 99.921%. Those figures come from different benchmarking methods, yet the gap is still large enough to shape deep circuits. A neutral-atom machine can afford more physical qubits for error correction, although those codes must overcome errors from gates, movement, measurement and atom loss.

Recent experiments show movement in the right direction. Atom’s toric-code work extended repeated error correction to 90 cycles and included reloading the atom reserve. Separate research on moving neutral atoms reported 99.86% entangling-gate fidelity while demonstrating operations relevant to error correction.

AC1000 already provides the scale. Better fidelity and a convincing whole-machine computation would turn that scale into leadership. No other architecture has such an obvious route from hundreds of qubits to thousands.

If you want more recent data on this point, please see our latest quantum computing market report.

Does D-Wave have the best quantum computer for optimization?

D-Wave has the best commercial quantum annealer, but classical solvers still make the broader optimization crown impossible to award.

Advantage2 has more than 4,400 qubits, over 40,000 couplers and up to 20 connections per qubit. Compared with D-Wave’s previous generation, the company reports 40% higher energy scales, twice the coherence time and four times lower noise. Customers can already access the machine through the Leap cloud service.

The machine works best when a problem can be expressed as an energy landscape that the annealer searches for low points. Some scheduling, sampling, materials and constrained-optimization problems fit this model. Many others require costly reformulation, and one business variable can consume several physical qubits after embedding.

Classical optimization is brutally competitive. A quantum annealer may beat one solver on one family of instances and lose to another solver after a new heuristic, GPU implementation or problem-specific shortcut appears. The 4,400-qubit headline therefore gives D-Wave a clear lead within annealing without settling the wider question.

D-Wave deserves its own category. Comparing Advantage2 directly with Helios or Willow would confuse two different types of computing.

Could China’s Zuchongzhi be the real leader?

USTC’s Zuchongzhi 3 is the leader in raw superconducting random-circuit sampling, though the available evidence remains too narrow for an overall first place.

Zuchongzhi 3 has 105 qubits and 182 couplers. USTC reported parallel fidelities of 99.90% for single-qubit gates, 99.62% for two-qubit gates and 99.13% for readout. The team ran an 83-qubit circuit with 32 layers and claimed a speed advantage of 10¹⁵ over the best classical simulation on a leading supercomputer.

That is the largest published speed claim among superconducting random-circuit experiments. USTC also deserves credit for testing old quantum-supremacy claims against improved classical algorithms. Its own researchers showed that Google’s famous 2019 Sycamore task, originally estimated at thousands of classical years, could later be reproduced in seconds with better methods and modern GPUs.

That history should make us cautious about every sampling record, including USTC’s. Classical simulation keeps improving, and a huge estimated speed gap can shrink after researchers discover a better algorithm. Sampling also tells us little about error-corrected logical qubits, flexible programming or outside access.

A newer Zuchongzhi 3.1 experiment created one-dimensional cluster states across 95 qubits and two-dimensional states across 72, showing that the platform can do more than random sampling. Even so, Google and Quantinuum currently provide stronger evidence on error correction and broader computation.

Does Microsoft Majorana 2 change the ranking?

Microsoft’s Majorana 2 is one of the most interesting hardware developments lately, but it is still a component breakthrough rather than the best complete quantum computer.

Microsoft says its latest topological qubits are 1,000 times more reliable than the previous generation. The company reports a mean qubit lifetime of 20 seconds, occasional lifetimes close to one minute and operation times around one microsecond. Conventional superconducting-qubit lifetimes are generally measured in microseconds, so the difference is striking.

Topological qubits could dramatically reduce the amount of hardware needed for error correction. Their quantum information is designed to be protected by the underlying physics, which could make errors less frequent before software correction even begins.

Majorana 2 has yet to show a large programmable processor, a useful multi-qubit algorithm or a logical computation that can be compared with Helios, Willow or Atom’s systems. Microsoft now targets a scalable commercial machine by 2029. That is an aggressive and potentially transformative roadmap, though it remains a roadmap.

Microsoft may have found a better foundation. It still needs to build the house.

Has any quantum computer delivered a real commercial advantage?

No quantum computer has yet produced a public, repeatable commercial win that survives comparison with the best classical methods.

A credible commercial advantage needs more than a large theoretical speedup. The machine must solve a real problem, produce a useful answer, beat the strongest classical competitor and do so after counting preparation, repeated runs, error handling and access costs.

Google’s Quantum Echoes result comes closest to combining classical inaccessibility with a scientifically meaningful task. Its molecular experiment points toward chemistry and materials research, but it was still a proof of principle. Helios has run increasingly difficult simulations and generated certified randomness over the internet, while D-Wave customers use annealing and hybrid systems for applied experiments. None has yet established a decisive economic advantage.

Rankings built from revenue announcements, customer logos or isolated pilot projects are not enough. Companies will happily test promising technology years before it becomes the best tool for the job.

Quantum computers already have research value. They let scientists test error correction, develop algorithms and explore physical systems that are becoming difficult to simulate. Commercial superiority remains an open target.

Who has the best quantum computer right now?

Quantinuum has the best quantum computer right now because Helios offers the strongest overall mix of accuracy, connectivity, logical computation, programmability and real access.

The recent peer-reviewed hardware results settle much of the uncertainty. As explained above, Helios combines 98 fully connected qubits with average two-qubit fidelity of 99.921%, and its whole-system random-circuit results reached beyond practical classical simulation. Quantinuum has also used the same processor to create dozens of logical qubits and run encoded scientific calculations.

Google comes closest. Willow has faster gates, the best surface-code memory result and the strongest verifiable beyond-classical algorithm. A ranking focused mainly on long-term fault tolerance could reasonably put Google first.

Atom Computing is the biggest threat to the current order. AC1000 starts with more than 1,200 qubits and now has evidence of repeated error correction. IBM remains the platform leader. USTC holds the largest raw sampling claim. D-Wave dominates annealing. Microsoft could eventually change the architecture of the entire race.

Those narrower victories do not add up to the best complete machine today. Helios does.

Category	Current winner	Direct judgment
Best quantum computer overall	Quantinuum Helios	The strongest complete and accessible general-purpose system
Best verifiable algorithmic result	Google Willow	Quantum Echoes produced the clearest beyond-classical computation with a scientific path
Best scalable logical memory	Google Willow	Larger surface codes measurably reduced logical errors
Broadest logical computation	Quantinuum Helios	Dozens of encoded qubits have already worked together
Largest programmable gate-based machine	Atom Computing AC1000	More than 1,200 physical qubits, with whole-machine performance still less proven
Best public platform	IBM Quantum	The broadest mix of hardware access, software and developer adoption
Best quantum annealer	D-Wave Advantage2	The mature leader for annealing-compatible problems
Strongest raw sampling machine	USTC Zuchongzhi 3	The largest reported superconducting sampling speed gap
Most disruptive long-term architecture	Microsoft Majorana 2	Exceptional reported qubit lifetimes, but no leading full computer yet

Who is the author of this content?

NEW MARKET PITCH TEAM

We track new markets so founders and investors can move faster

We build living "market pitch" documents for emerging markets: AI, synthetic biology, new proteins, and more. Instead of outdated PDFs or hallucinated LLM answers, our clients get a clean, visual, always-updated view of what's really happening: key players, deals, regulations, and signals that matter. Learn more about us.