Concept

Quantum advantage: what it means, every major claim, and what has held up

Quantum advantage means a quantum computer doing some task faster, cheaper or more accurately than any classical method can. It has been demonstrated for tasks built to test it, such as random circuit sampling, boson sampling and a verifiable physics measurement on Google's Willow chip, and the latest of those results still stand. It has not yet been demonstrated beyond dispute on a problem someone needed solved: each claim of useful advantage so far was matched by an improved classical method, is contested, or is still under review. Below is every major claim, how it was checked and whether classical computers caught up.

Updated

The short answer

Has quantum advantage been achieved?
Yes, for tasks designed to test it: sampling from random circuits and from photonic interferometers, certified random numbers, and one verifiable physics measurement. Not yet beyond dispute for a task anyone needed solved.
What happened to most claims
Classical algorithms improved and caught up, often within months. Google's 2019 experiment, IBM's 2023 utility experiment and a 2026 materials simulation all went this way[4][14][33].
Strongest results still standing
Random circuit sampling on Willow and Zuchongzhi 3.0[7][12], the Quantum Echoes measurement on Willow[22], and a provable separation in a sampling game on Quantinuum H2[34].
Open and under review
D-Wave's annealing simulation of spin glasses[16] and three IBM partner experiments from July 2026[38], tracked against classical methods on a public tracker[26].
Where useful advantage is expected first
Simulating quantum physics and chemistry first, then problems that need error-corrected machines. Machine learning on classical data is not among the early candidates.

What quantum advantage means

The idea is older than the hardware. In 2012 John Preskill called it quantum supremacy: a quantum device doing a task that no classical computer can do in any reasonable time, whether or not the task is useful[1]. Most researchers now say quantum advantage or beyond-classical computation for the same thing, and the claims since then differ less in the words than in what exactly they say was beaten.

IBM's current definition is a useful yardstick: quantum advantage is the point at which a quantum computer is shown to be faster, more accurate or more cost-effective than all classical-only methods for some problem, in a way that can be rigorously validated[27][25]. Two words do the work. All classical methods means the best ones, including those not yet invented when the claim is made. Validated means there must be a reason to trust the quantum output when no classical computer can reproduce it.

The words in use, and what each one claims

TermWhat it claimsExample
Quantum supremacyA task beyond any classical computer in reasonable time, useful or not[1]Google's 2019 random circuit sampling[2]
Quantum (computational) advantageThe same claim in the words most groups now use; sometimes also used for a narrower speed-upUSTC's Jiuzhang boson sampling[8]
Quantum utilityReliable results at a scale beyond brute-force classical simulation, without claiming to beat every classical method[13]IBM's kicked Ising experiment on Eagle
Practical or useful advantageBeating the best available classical tool on a problem of known commercial or scientific value[31]Q-CTRL's Fermi-Hubbard simulation, later matched classically[33]
Verifiable advantageThe output can be checked, by a classical computer, by another quantum computer, or against nature[23]Google's Quantum Echoes[22]
Unconditional advantageA separation proven mathematically, not resting on the belief that some problem is hard for classical computers[34]Quantinuum's complement sampling game

Every major quantum advantage claim, and where it stands

Status as of November 2026. "Matched" means a classical method later reproduced the result as well or better, in comparable time. "Standing" means no published classical method has done so yet. How the hardware types in the table differ is on the superconducting, trapped-ion and neutral-atom comparison.

YearTeam and hardwareTaskHow it was checkedStatus
2019Google, Sycamore (superconducting, 53 qubits[2])Random circuit samplingFidelity measured on smaller and simplified circuits, extrapolated to the full onesMatched: sampled classically in about 15 hours on 512 GPUs[4]
2020USTC, Jiuzhang (photonic, 100 modes[8])Gaussian boson samplingStatistical tests against the ideal distribution and against simpler classical modelsMatched for the early experiments by samplers that exploit photon loss[9][10]
2023IBM, Eagle (superconducting, 127 qubits[13])Expectation values of a kicked Ising model (utility)Compared with exact classical results in the regimes where they existMatched within months by tensor networks[14][15]
2024Google, Willow (superconducting, 105 qubits[7])Random circuit samplingSame method as 2019, at larger sizeStanding; classical estimate 102510^{25} years[7]
2025USTC, Zuchongzhi 3.0 (superconducting, 83 of 105 qubits[12])Random circuit samplingSame method as Google'sStanding; classical estimate 5.9×1095.9 \times 10^{9} years[12] on Frontier
2025D-Wave, Advantage2 (quantum annealer)[16]Quench dynamics of spin glassesAgreement with Schrödinger-equation solutions at sizes classical methods reachContested: tensor networks and variational Monte Carlo reproduce part of it[18][19]; D-Wave says the hardest cases stand[20]
2025JPMorgan Chase, Quantinuum and US national labs, H2-1 (trapped ion, 56 qubits[21])Certified random numbersChecked on supercomputers at 1.1×10181.1 \times 10^{18} operations per second[21]Standing, under a restricted adversary model
2025Google, Willow (65 qubits[23] for the hardest data)Second-order out-of-time-order correlators (Quantum Echoes)Expectation values that another quantum computer could reproduce; nine classical algorithms triedStanding[24]
2025 to 2026BlueQubit, on Quantinuum H2 and IBM ibm_boston[29]Peaked circuits (sampling with a known answer)The answer is known when the circuit is builtMatched: new classical methods ran in seconds to an hour[27]
2026USTC, Jiuzhang 4.0 (photonic, 8,176 modes[11])Gaussian boson samplingTests against all known classical spoofing methods, including loss-exploiting onesStanding
2026Q-CTRL with IBM (superconducting, 120 qubits[30])Dynamics of a 1D Fermi-Hubbard modelAgreement with a classical tensor network where it convergedMatched: reproduced classically faster than the quantum run[33]
2026Quantinuum, H2 (trapped ion, up to 55 qubits[34])Complement sampling gameScore above a proven classical ceiling, checked efficientlyStanding, by proof; a test of quantumness, not a useful computation
2026IBM with UChicago, Qedma and Algorithmiq, Heron (superconducting)[38]Hard sampling with a fidelity certificate; Floquet magnet dynamics; operator Loschmidt echoError-detection codes, cross-checks on other quantum computersOpen: on the public tracker[26]

Random circuit sampling: Sycamore, Willow and Zuchongzhi

Random circuit sampling (RCS) asks a quantum computer to run a circuit of randomly chosen gates and report the bit strings it measures. A noiseless device would produce strings in a pattern set by interference across all qubits, and computing that pattern classically takes time exponential in the circuit's size. In 2019 Google ran such circuits on 53 qubits[2] of its Sycamore chip and drew a million samples in 200 seconds[2], a task it estimated would take a supercomputer 10,000 years[2].

The check was indirect. The fidelity of the largest circuits cannot be computed, because that needs the very simulation that is out of reach. Google measured it on smaller and simplified circuits that a computer can simulate and extrapolated[2]. Classical responses came quickly. IBM argued within days that disk storage on the Summit supercomputer would bring the simulation down to a matter of days[3]. Pan, Chen and Zhang then sampled the Sycamore circuits with a tensor network in about 15 hours on 512 GPUs[4], at a higher fidelity than the experiment's. Google's own later estimate put the 2019 task at about 6 seconds[6] on the Frontier supercomputer, and a USTC team reported a GPU simulation 7 times faster[5] than the original experiment.

That is not the end of RCS. The cost of a classical simulation grows exponentially with the number of qubits and the depth, while quantum hardware has kept improving. Google's 2024 experiments at 67 qubits and 32 cycles[6] also mapped the noise level below which the experiment stays out of classical reach. Willow, a 105-qubit[7] chip, ran its benchmark in under five minutes[7] against an estimated 102510^{25} years[7] on Frontier. USTC's Zuchongzhi 3.0 drew a million samples from an 83-qubit, 32-cycle[12] circuit in a few hundred seconds[12], a task estimated at 5.9×1095.9 \times 10^{9} years[12] on Frontier. No classical simulation of these has been published.

The honest reading: RCS at today's sizes is beyond classical computers, and the 2019 claim was the starting gun rather than the finish. RCS is also useless in itself and not efficiently verifiable, which is why the next round of claims went after both problems.

Boson sampling: Jiuzhang

USTC's Jiuzhang sends squeezed light through an interferometer and counts the photons at the outputs, a task called Gaussian boson sampling. The 2020 experiment used 50 squeezed states[8] in 100 modes[8], saw up to 76 photon clicks[8] and claimed a sampling rate 101410^{14} times[8] faster than a supercomputer.

Photon loss turned out to be the weak point. Real optics lose most photons, and a lossy experiment is closer to a classical state. Oh, Liu, Alexeev, Fefferman and Jiang built a classical sampler that exploits loss and scored better than the largest experiments on the benchmarks that had been used as evidence of advantage[9]. A 2026 preprint by Drummond's group reports a sampler closer to the exact answer than the experiments up to 1,152 modes[10].

USTC answered with scale and lower loss. Jiuzhang 4.0 injects 1,024 squeezed states[11] into 8,176 modes[11] and records up to 3,050 photons[11], and its authors test it against every known spoofing method, the loss-exploiting one included. Their preprint puts that method at more than 104210^{42} years[11] on the El Capitan supercomputer, against 25.6 microseconds[11] per sample on the device. It stands for now. Like RCS, it is a sampling task with no use of its own.

Quantum utility: IBM's Eagle experiment and its classical simulations

In 2023 IBM measured the magnetisation of a kicked Ising model on its 127-qubit[13] Eagle processor, used error mitigation to remove the effect of noise, and called the result evidence of utility: reliable answers at a scale beyond brute-force classical simulation[13]. It was a deliberate step away from sampling towards expectation values, the kind of number physics and chemistry need.

Within months, several groups reproduced it on classical hardware. A belief-propagation tensor network shaped like the chip gave results significantly more accurate and precise than the quantum processor[14], and other approximate methods ran orders of magnitude faster than the experiment[15]. The tensor networks page explains why: the circuit's correlations stayed tree-like enough for a well-chosen network to follow. IBM's later candidates are on the public tracker described below.

D-Wave's annealing simulation of spin glasses

In 2025 D-Wave and collaborators published in Science that its annealers simulate the quench dynamics of spin glasses, in two-, three- and infinite-dimensional lattices, in close agreement with the Schrödinger equation, and that leading tensor-network and neural-network methods could not reach the same accuracy in reasonable time[16]. D-Wave states that matching its largest simulations with matrix product states would take nearly a million years[17] on Frontier. Unlike RCS, the observable is one a physicist might want: how a magnet's correlations grow as it is driven through a transition.

Classical replies followed. Tindall and colleagues at the Flatiron Institute simulated the same models on two- and three-dimensional lattices with belief-propagation tensor networks at modest cost, published in Science in 2026[18]. Mauron and Carleo used time-dependent variational Monte Carlo on up to 128 spins[19] of a diamond lattice with correlation errors below 7%[19]. D-Wave's response is that neither work attempted the infinite-dimensional (biclique) lattices, the low-precision inputs where correlations grow fastest, the longest times or the full set of observables[20][17]. Both sides agree the classical work narrowed the claim. Whether anything beyond classical reach remains depends on the cases not yet simulated. The quantum annealing page covers what these machines do in optimisation, where no such claim has held.

Certified randomness: the first beyond-classical task with a use

A random number from a remote server is only as trustworthy as the server. JPMorgan Chase, Quantinuum and US national laboratories used RCS to fix that: a client sends random challenge circuits to a 56-qubit[21] Quantinuum H2-1 over the internet, the server must answer faster than a classical computer could fake it, and the client spot-checks the answers on supercomputers running at a combined 1.1×10181.1 \times 10^{18} operations per second[21]. The run certified 71,313 bits[21] of entropy.

It is the first beyond-classical computation with an application, but a narrow one. The security proof covers a restricted class of adversaries and rests on further assumptions, and the authors say the bit rate and those assumptions limit production use for now[21].

Verifiable advantage: Google's Quantum Echoes

In October 2025 Google published an experiment designed to fix RCS's two weaknesses. Quantum Echoes runs a random circuit forward, nudges one qubit, runs the circuit backward, and measures an out-of-time-order correlator (OTOC), a number that describes how a disturbance spreads through a quantum system[22]. An expectation value is not a random bit string: any quantum computer of similar quality, or a natural system with the same dynamics, should give the same number, so the result can be checked. That is what Google means by verifiable[23].

The hardest data used 65 qubits[23] of Willow. The experiment took about 2 hours[23], which Google estimates would take 13,000 times longer[23] on Frontier, after about 10 person-years[23] of trying to beat it with nine[23] classical algorithms. A 2026 study by a team that includes Google researchers found that the belief-propagation tensor networks that matched IBM's and part of D-Wave's experiments cannot feasibly simulate it[24].

Two caveats keep it in perspective. The verification was a promise at publication: no second quantum computer had yet reproduced the hardest data. And the benchmark circuits are random, so the advantage is shown on a physics quantity, not yet on a molecule or material someone needed measured. Google's companion work points the method at NMR spectroscopy of molecules, so far at sizes classical methods still handle[23].

Claims from 2026

The Quantum Advantage Tracker. In late 2025 IBM, Algorithmiq, the Flatiron Institute and BlueQubit opened a public tracker where candidate experiments are posted with their circuits and data, and anyone can submit a classical method that matches them[26]. Candidates are listed as active, superseded when a classical method closes the gap, or kept as baselines. Its first test was BlueQubit's peaked circuits, random-looking circuits built so that one output string is far more likely than the others, which makes the answer easy to check[28]. Quantinuum's H2 found the peak in under 2 hours[29] where the best classical method was extrapolated to years, and IBM hardware later ran larger instances in under 12 minutes[27]. By February 2026 new classical methods solved the instances in an hour down to seconds[27], faster than both machines. The tracker did its job.

Q-CTRL's materials simulation. In May 2026 Q-CTRL reported what it called practical quantum advantage: a Fermi-Hubbard simulation on 120 qubits[30] of an IBM processor, 3,000 times[31] faster in wall-clock time than a tensor-network solver with bond dimension 4,096[30]. Within weeks, Rausch and colleagues used the model's symmetry and four GPUs[32] to cut the gap to about 36 times[32] and converged the regime the original classical run had not. Ouyang, Chi and Chan then reproduced all 7,260[33] measured observables with a different contraction scheme at bond dimension 32[33], faster than the quantum run. One-dimensional dynamics is where tensor networks are strongest, and the baseline chosen was not the best available.

Quantinuum's complement sampling. A game in which a player receives a quantum state spread over a hidden half of all bit strings and must return a string from the other half. No classical strategy can do much better than guess, and that ceiling is proven rather than assumed. On H2 machines with up to 55 qubits[34] and strings of up to 37 bits[34], the scores beat the classical ceiling by a margin that grew exponentially with size[34]. The proof makes this the cleanest separation yet. It assumes a trusted referee, and it shows that the device behaves quantum mechanically, not that it computed something useful.

IBM and partners, July 2026. IBM presented three papers as demonstrations of quantum advantage, each with a way to trust the output when no classical answer exists[38]. IBM and University of Chicago researchers ran a hard 64-qubit, depth-73[35] circuit encoded in an error-detecting code on 76 physical qubits[35] and certified a fidelity of at least 0.349[35] with 95%[35] confidence. Qedma and partners tracked the dynamics of a periodically driven magnet on up to 74 qubits[36], where tensor-network simulations failed to converge and Pauli-path simulations depended on their truncation; selected runs were repeated on Quantinuum hardware. Algorithmiq measured a new observable on 56 qubits[38] where classical methods disagreed with each other, and argued its result is the most credible of the methods considered[37]. IBM itself says the announcement opens the results to scrutiny rather than closing the case. They are on the tracker, and too new to judge.

How to judge the next quantum advantage claim

Six questions, each answered by something in the paper or its replies. The examples are from the table above.

  1. What exactly was computed, and does anyone need it?

    Sampling random circuits and photons, certified bits and game scores are tests of the machine. A magnetisation, a correlation function or an energy of a named model is closer to use. A claim of practical advantage should name the user who needed the answer.

  2. Can the output be checked?

    The strongest claims have an output that can be verified: a known answer, as in peaked circuits[28], a proven bound[34], or a value another quantum computer can reproduce[23]. RCS can only be checked on smaller versions of itself.

  3. Was the best classical method tried, and by whom?

    The authors' own baseline is the weakest test. Look for independent classical groups and for the tracker. IBM's 2023 experiment and Q-CTRL's 2026 one both fell to methods their authors had not used[14][32].

  4. Is the comparison fair?

    Check what is timed on each side: bare QPU time or the whole job, one CPU node or a GPU cluster, a generic solver or one built for the problem's structure. In the Q-CTRL case the choice of classical solver alone moved the gap from 3,000 times[31] to about 36 times[32].

  5. Does the gap grow with size?

    A real advantage widens as the problem grows. If the quantum result degrades with noise faster than the classical cost rises, the window closes. Look for results at several sizes, not one.

  6. Has it lasted?

    Most refutations so far came within weeks or months. A claim that has stood a year against open attempts, with its data and circuits public, is worth more than a new one with a large ratio.

What the evidence adds up to

Quantum computers have done things no classical computer can match in reasonable time. That is no longer in serious doubt for random circuit sampling at the sizes of Willow and Zuchongzhi 3.0, for the newest boson sampling, and for the Quantum Echoes measurement. The gap has also been moving the right way: the 2019 result fell to classical methods, but the hardware moved far enough ahead that the experiments of 2024 and 2025 have not.

Useful advantage is a different matter. Every claim on a problem with a use, from IBM's utility experiment to Q-CTRL's materials simulation, has been matched or narrowed by a classical method, and the July 2026 candidates have not yet been tested by outsiders for long. The pattern is not that quantum results are wrong: they are usually accurate. It is that classical simulation of quantum systems keeps getting better, especially with tensor networks, and today's noisy devices only stay ahead on problems with entanglement that spreads in ways those networks cannot follow.

Where the first lasting useful advantage is expected is clearer than when. The leading candidates are simulations of quantum dynamics and of materials and molecules, the same territory as Quantum Echoes, D-Wave's spin glasses and the IBM partner experiments; the quantum simulation guide goes through each simulation claim and its classical reply. Claims that need deep circuits, such as factoring or chemistry at industrial accuracy, wait for error-corrected machines, explained on the quantum error correction page. The where quantum computing will help first guide goes through the candidates by field. Machine learning on classical data is not among the early ones, for reasons the quantum machine learning guide sets out.

Questions

Has quantum advantage been achieved?

Yes, for tasks built to test it. Random circuit sampling on Google's Willow and USTC's Zuchongzhi 3.0, the latest boson sampling on Jiuzhang 4.0 and Google's Quantum Echoes measurement have no published classical match. Advantage on a problem someone needed solved has not been shown beyond dispute: each claim so far was matched by a better classical method, is contested, or is still under review.

What is the difference between quantum supremacy and quantum advantage?

Quantum supremacy, the term John Preskill introduced, means a quantum device doing any task a classical computer cannot do in reasonable time, useful or not. Quantum advantage is the term most groups now use for the same idea. Practical or useful advantage is the stronger claim: beating the best classical tool on a problem with real value.

Did classical computers catch up with Google's quantum supremacy experiment?

Yes. The circuits of Google's 2019 Sycamore experiment were later sampled classically with tensor networks on a GPU cluster, and Google's own later estimate put the task at seconds on the Frontier supercomputer. Google's newer random circuit sampling on Willow is much larger and has no classical match.

What is verifiable quantum advantage?

A beyond-classical result whose output can be checked. Some tasks have a known or efficiently checkable answer, such as peaked circuits and Quantinuum's complement sampling game. Google's Quantum Echoes measures an expectation value that another quantum computer of similar quality should reproduce, which is how Google uses the term.

What are examples of quantum advantage?

Random circuit sampling by Google and USTC, Gaussian boson sampling on USTC's Jiuzhang processors, certified random numbers on a Quantinuum trapped-ion computer, Google's Quantum Echoes on Willow, and Quantinuum's complement sampling game. D-Wave's spin-glass simulation and IBM's 2026 partner experiments are claims still under debate.

When will quantum computers be useful?

The first lasting useful advantage is expected in simulating quantum systems, such as magnets, materials and molecules, where today's devices are closest. Most other applications, including factoring and chemistry at industrial accuracy, need error-corrected quantum computers that are still being built.

Sources

  1. J. Preskill. Quantum computing and the entanglement frontier. arXiv preprint (rapporteur talk at the 25th Solvay Conference on Physics), 2012. arXiv:1203.5813.
  2. F. Arute, K. Arya, R. Babbush et al. Quantum supremacy using a programmable superconducting processor. Nature 574, 505–510, 2019. arXiv:1910.11333.
  3. E. Pednault, J. A. Gunnels, G. Nannicini, L. Horesh, R. Wisnieff. Leveraging secondary storage to simulate deep 54-qubit Sycamore circuits. arXiv preprint, 2019. arXiv:1910.09534.
  4. F. Pan, K. Chen, P. Zhang. Solving the sampling problem of the Sycamore quantum circuits. Physical Review Letters 129, 090502, 2022. arXiv:2111.03011.
  5. X.-H. Zhao, H.-S. Zhong, F. Pan, Z.-H. Chen, R. Fu, Z. Su, X. Xie, C. Zhao, P. Zhang, W. Ouyang, C.-Y. Lu, J.-W. Pan, M.-C. Chen. Leapfrogging Sycamore: harnessing 1432 GPUs for 7x faster quantum random circuit sampling. arXiv preprint, 2024. arXiv:2406.18889.
  6. A. Morvan, B. Villalonga, X. Mi, S. Mandrà, A. Bengtsson, P. V. Klimov et al. Phase transitions in random circuit sampling. Nature 634, 328–333, 2024. arXiv:2304.11119.
  7. H. Neven, Google Quantum AI. Meet Willow, our state-of-the-art quantum chip. Google blog, 2024.
  8. H.-S. Zhong, H. Wang, Y.-H. Deng, M.-C. Chen et al. Quantum computational advantage using photons. Science 370, 1460, 2020. arXiv:2012.01625.
  9. C. Oh, M. Liu, Y. Alexeev, B. Fefferman, L. Jiang. Classical algorithm for simulating experimental Gaussian boson sampling. Nature Physics 20, 1461–1468, 2024. arXiv:2306.03709.
  10. N. Goodman, A. S. Dellios, M. D. Reid, P. D. Drummond. Gaussian boson sampling: benchmarking quantum advantage. arXiv preprint, 2026. arXiv:2604.12330.
  11. H.-L. Liu, H. Su, Y.-H. Deng et al. Gaussian boson sampling with 1,024 squeezed states in 8,176 modes. Nature 653, 687–692 (preprint: Robust quantum computational advantage with programmable 3050-photon Gaussian boson sampling), 2026. arXiv:2508.09092.
  12. D. Gao, D. Fan, C. Zha, J. Bei, G. Cai et al. Establishing a new benchmark in quantum computational advantage with 105-qubit Zuchongzhi 3.0 processor. Physical Review Letters 134, 090601, 2025. arXiv:2412.11924.
  13. Y. Kim, A. Eddins, S. Anand, K. X. Wei, E. van den Berg, S. Rosenblatt et al. Evidence for the utility of quantum computing before fault tolerance. Nature 618, 2023.
  14. J. Tindall, M. Fishman, E. M. Stoudenmire, D. Sels. Efficient tensor network simulation of IBM's Eagle kicked Ising experiment. PRX Quantum 5, 010308, 2024. arXiv:2306.14887.
  15. T. Begušić, J. Gray, G. K.-L. Chan. Fast and converged classical simulations of evidence for the utility of quantum computing before fault tolerance. Science Advances 10, eadk4321, 2024. arXiv:2308.05077.
  16. A. D. King, A. Nocera, M. M. Rams, J. Dziarmaga, R. Wiersema et al. Beyond-classical computation in quantum simulation. Science 388, 199–204, 2025. arXiv:2403.00910.
  17. D-Wave Quantum Inc. D-Wave's quantum supremacy result stands (press release, filed with the SEC as exhibit 99.1). D-Wave press release, 2026.
  18. J. Tindall, A. Mello, M. Fishman, E. M. Stoudenmire, D. Sels. Dynamics of disordered quantum systems with two- and three-dimensional tensor networks. Science 392, 868–872, 2026. arXiv:2503.05693.
  19. L. Mauron, G. Carleo. Challenging the quantum advantage frontier with large-scale classical simulations of annealing dynamics. arXiv preprint, 2025. arXiv:2503.08247.
  20. A. D. King, A. Nocera, M. M. Rams, J. Dziarmaga, J. Raymond, N. Kaushal, A. W. Sandvik, G. Alvarez, J. Carrasquilla, M. Franz, M. H. Amin. Comment on: Dynamics of disordered quantum systems with two- and three-dimensional tensor networks. arXiv preprint, 2025. arXiv:2504.06283.
  21. M. Liu, R. Shaydulin, P. Niroula, M. DeCross et al. Certified randomness using a trapped-ion quantum processor. Nature 640, 343–348, 2025. arXiv:2503.20498.
  22. Google Quantum AI and collaborators (D. A. Abanin et al.). Observation of constructive interference at the edge of quantum ergodicity. Nature 646, 825–830, 2025. arXiv:2506.10191.
  23. X. Mi, K. Kechedzhi. A verifiable quantum advantage. Google Research blog, 2025.
  24. P. Bermejo, B. Villalonga, B. Ware, G. Vidal, A. Szasz. Tensor networks with belief propagation cannot feasibly simulate Google's quantum echoes experiment. arXiv preprint, 2026. arXiv:2604.15427.
  25. O. Lanes, M. Beji, A. D. Córcoles, C. Dalyac, J. M. Gambetta, L. Henriet, A. Javadi-Abhari, A. Kandala, A. Mezzacapo, C. Porter, S. Sheldon, J. Watrous, C. Zoufal, A. Dauphin, B. Peropadre. A framework for quantum advantage. arXiv preprint, 2025. arXiv:2506.20658.
  26. Algorithmiq, BlueQubit, Flatiron Institute, IBM and contributors. Quantum Advantage Tracker. Community website, 2025.
  27. J. Gambetta, R. Davis. Quantum Advantage Tracker: the race to advantage. IBM Quantum blog, 2026.
  28. S. Aaronson, Y. Zhang. On verifiable quantum advantage with peaked circuit sampling. arXiv preprint, 2024. arXiv:2404.14493.
  29. H. Gharibyan, M. Z. Mullath, N. E. Sherman, V. P. Su, H. Tepanyan, Y. Zhang. Heuristic quantum advantage with peaked circuits. arXiv preprint, 2025. arXiv:2510.25838.
  30. G. S. Hartnett, K. S. Najafi, A. Khindanov, H. Liao, M. Schutzman, M. R. Hush, M. J. Biercuk, Y. Baum. Fast, accurate, high-resolution simulation of large-scale Fermi-Hubbard models on a digital quantum processor. arXiv preprint, 2026. arXiv:2605.04025.
  31. Q-CTRL. Practical quantum advantage signals a new commercial era for quantum computing. Q-CTRL blog, 2026.
  32. R. Rausch, S. Singh, S. S. Jahromi, A. Kshetrimayum, R. Orús. Pushing the classical frontier of 1D Fermi-Hubbard quench dynamics beyond current quantum simulations. arXiv preprint, 2026. arXiv:2606.04771.
  33. X.-Y. Ouyang, R. Chi, G. K.-L. Chan. Fast classical simulation of "Fast, accurate, high-resolution simulation of large-scale Fermi-Hubbard models on a digital quantum processor". arXiv preprint, 2026. arXiv:2608.13805.
  34. M. Benedetti, H. Buhrman et al. Unconditional and exponentially large violation of classicality. Nature Communications 17, 10557, 2026.
  35. S. Martiel, J.-U. Chung, A. Seif, S. Ghosh, I. Hincks, A. Deshpande, H. Manabe, H. Gu, F. Pan, B. Fefferman, J. M. Gambetta, A. Javadi-Abhari. Sampling hard circuits with verifiably high fidelity. arXiv preprint, 2026. arXiv:2607.25941.
  36. E. Leviatan, T. Watad, R. Perry, L. Broers, M. Z. Mullath, O. Alberton et al. Resolving structure in prethermal Floquet dynamics with precision quantum computation. arXiv preprint, 2026. arXiv:2607.24937.
  37. S. V. Barron, B. Mitchell, V. Tripathi, F. Grieco, I. Rosen, F. Pietracaprina et al. Observable estimation in the absence of classical verification. arXiv preprint, 2026. arXiv:2607.25998.
  38. A. Kandala, A. Javadi-Abhari, J. Gambetta. Researchers demonstrate quantum advantage through trusted quantum computation. IBM Quantum blog, 2026.

Run it yourself with eigenkit

eigenkit is EigenOrbit's open-source Python library: write a circuit once, run it on any simulator or QPU, and test it against a classical baseline on your own data before you invest in it or rule it out.

$ pip install eigenkit