Skip to content
Dossier · ExplainerComputingUpdated Q4 2026 ·

Which Quantum Advantage Claims Have Been Made, And Which Held Up?

First version 5 min11/11 facts confirmed

Several “quantum advantage” claims have been made since Google’s original Sycamore result in 2019. Classical computers have closed the gap on most of them, sometimes within months. The newest is different. Google’s “Quantum Echoes” result on its Willow chip, published in Nature on 22 October 2025, is the first the company designed so that other quantum computers can reproduce and check it. Earlier claims only raced a classical machine.

The first claim, and the fastest rebuttal

Google published in Nature on 23 October 2019 that its 53-qubit Sycamore processor had sampled a random quantum circuit’s output in about 200 seconds. It said the fastest classical supercomputer of the time would need around 10,000 years for that task. That was the first claim of “quantum supremacy”: a case where a quantum computer does something no classical computer can realistically match.

IBM’s rebuttal appeared a day before Google’s paper came out in Nature. According to IBM, researchers Edwin Pednault, John Gunnels, Dmitri Maslov and Jay Gambetta argued in an October 2019 blog post that an ideal classical simulation of the same circuit could run in about 2.5 days. That simulation would deliver far greater accuracy, using a supercomputer’s full RAM and hard-disk storage, they said. The original 10,000-year threshold, in their view, had not actually been crossed.

China’s larger claim, and the algorithms that caught up

USTC reported its own quantum-advantage claim in June 2021, later published in Physical Review Letters. Its 56-qubit Zuchongzhi processor sampled random circuits in about 1.2 hours. The team estimated the most powerful classical supercomputer would need at least 8 years for that task, 2 to 3 orders of magnitude harder than Sycamore’s result.

Classical methods then went after both claims, with uneven results. In November 2021, physicists Feng Pan, Keyang Chen and Pan Zhang reported that a tensor-network algorithm run on a 512-GPU cluster solved the Sycamore sampling problem in about 15 hours. They estimated an exascale supercomputer could do the same task in a few dozen seconds, far below Google’s original 10,000-year estimate. A separate team went further in a June 2024 paper posted on arXiv. Using 1,432 Nvidia A100 GPUs, it generated 3 million uncorrelated samples of the Sycamore circuit in about 86 seconds, seven times faster than Google’s original experiment, while using about 13.7 kilowatt-hours of electricity. Between them, those two results closed most of the gap the 2019 Sycamore claim had opened. Zuchongzhi’s larger claim proved harder to match. In a paper posted on arXiv on 30 December 2021, Gleb Kalachev, Pavel Panteleev, PengFei Zhou and Man-Hong Yung described a bounded-fidelity classical sampling method. It produced 1 million low-fidelity samples of Zuchongzhi’s 56-qubit circuit on the Selene supercomputer, using 4,480 Nvidia A100 GPUs, in about 4 days. That narrowed the classical cost of USTC’s claim without closing it the way Sycamore’s claim was closed.

A logical-qubit claim, undone by a faster algorithm

A team at Harvard, MIT and QuEra took a different route, with error correction rather than raw speed. According to Harvard and QuEra, it reported in Nature in December 2023 that its 48-logical-qubit neutral-atom processor ran instantaneous quantum polynomial-time circuits. Those circuits, the team said, would become rapidly harder to simulate classically as more circuit layers were added, part of a wider demonstration of error-corrected logical-qubit operations.

That claim did not last long. According to IBM and IonQ, a classical algorithm built by IBM’s Dmitri Maslov and Sergey Bravyi with IonQ researchers computed an amplitude for the same circuit in about 0.0026 seconds in a February 2024 paper. That was roughly 1,000 times faster than the simulation time the Harvard/QuEra team had itself reported, challenging the idea that those circuits were classically intractable.

An annealer’s claim, and a quick classical answer

D-Wave took the next run at advantage with a different kind of hardware. According to D-Wave, it said in a paper published in Science on 12 March 2025 that its Advantage2 annealing prototype simulated the dynamics of programmable spin-glass magnetic materials in minutes. It said the Frontier supercomputer at Oak Ridge would need almost one million years for that task, and more electricity than the world uses in a year.

Joseph Tindall and colleagues at the Flatiron Institute answered five days earlier than D-Wave’s own publication. According to the Flatiron Institute, they posted a paper on arXiv on 7 March 2025 describing a classical tensor-network method. It matched or beat D-Wave’s Advantage2 results on several of the same spin-glass lattices, verified on systems involving hundreds of qubits, disputing the quantum-supremacy claim before the Science paper was even out.

What makes the newest claim different

Every earlier claim rested on the same bet: that no classical computer, however large, could keep up. In most cases, a better algorithm eventually closed most of that gap, or came close enough to make the claim debatable; Zuchongzhi’s claim only narrowed this way, without closing. Google’s Quantum Echoes result rests on a different kind of evidence. Running on its 105-qubit Willow chip, the algorithm measures out-of-time-order correlators about 13,000 times faster than the Frontier supercomputer could, according to Google. The company calls it the first “verifiable” quantum advantage for that reason: other quantum computers of similar quality can reproduce and cross-check the result themselves.

Claim Hardware Classical answer
Sycamore, Oct 2019. 53-qubit superconducting chip. Disputed within a day by IBM; largely closed by a 2024 GPU result.
Zuchongzhi, Jun 2021. 56-qubit superconducting chip. Narrowed, not closed, by a December 2021 bounded-fidelity algorithm.
Harvard/QuEra IQP, Dec 2023. 48 logical qubits, neutral atoms. Classical algorithm ran about 1,000 times faster than the team’s own reported simulation time.
D-Wave Advantage2, Mar 2025. Quantum annealer. Matched by a classical tensor-network method days earlier.
Google Quantum Echoes, Oct 2025. 105-qubit Willow chip. Designed to be cross-checked by other quantum hardware rather than raced against a classical one.

Reproducing the Quantum Echoes result, Google says, takes a quantum computer of comparable quality to Willow, not a classical challenger.

Sources

  1. [1]A verifiable quantum advantage Google Quantum AI
  2. [2]Observation of constructive interference at the edge of quantum ergodicity Nature en Nature-tijdschriften
  3. [3]Quantum Supremacy Using a Programmable Superconducting Processor Google Quantum AI
  4. [4]Quantum supremacy using a programmable superconducting processor Nature en Nature-tijdschriften
  5. [5]On 'Quantum Supremacy' IBM Quantum en IBM Research
  6. [6]Strong Quantum Computational Advantage Using a Superconducting Quantum Processor arXiv (quant-ph)
  7. [7]Solving the Sampling Problem of the Sycamore Quantum Circuits arXiv (quant-ph)
  8. [8]Leapfrogging Sycamore: Harnessing 1432 GPUs for 7x Faster Quantum Random Circuit Sampling arXiv (quant-ph)
  9. [9]Logical quantum processor based on reconfigurable atom arrays Nature en Nature-tijdschriften
  10. [10]Fast Classical Simulation of Harvard/QuEra IQP Circuits arXiv (quant-ph)
  11. [11]Beyond Classical: D-Wave First to Demonstrate Quantum Supremacy on Useful, Real-World Problem D-Wave Quantum
  12. [12]Beyond-classical computation in quantum simulation Science (AAAS)
  13. [13]Dynamics of Disordered Quantum Systems with Two- and Three-Dimensional Tensor Networks arXiv (quant-ph)
  14. [14]Classical Sampling of Random Quantum Circuits with Bounded Fidelity arXiv (quant-ph)
MethodologyHow this article was made
Facts confirmed
11 / 11
Sources
14
Last checked
Nov 12, 2026
Image
Photo

This article was written by an AI system from facts in international trade media, publications by companies and research institutes and public sources, with the source for every fact, and checked under our methodology. The byline is an editorial pseudonym. Found an error? Report it through the contact page.

Facts used

  1. Google said in Nature on 22 October 2025 that its 105-qubit Willow chip ran a new "Quantum Echoes" algorithm, measuring out-of-time-order correlators, about 13,000 times faster than the Frontier supercomputer could, and called the result the first "verifiable" quantum advantage because other quantum computers of similar quality can reproduce and cross-check it. Google Quantum AI · Nov 12, 2026
  2. On 23 October 2019 Google published in Nature that its 53-qubit Sycamore processor sampled the output of a random quantum circuit in about 200 seconds, a task it said would take the fastest classical supercomputer of the time around 10,000 years, the first claim of "quantum supremacy". Google Quantum AI · Nov 12, 2026
  3. In an October 2019 blog post, IBM researchers Edwin Pednault, John Gunnels, Dmitri Maslov and Jay Gambetta said an ideal classical simulation of Google's Sycamore circuit could run in about 2.5 days with far greater accuracy, using a supercomputer's full RAM and hard-disk storage, arguing the original "supremacy" threshold had not been crossed. IBM Quantum · Nov 12, 2026
  4. USTC reported, in work posted in June 2021 and published in Physical Review Letters, that its 56-qubit Zuchongzhi superconducting processor sampled random circuits in about 1.2 hours, a task it estimated would take the most powerful classical supercomputer at least 8 years, 2-3 orders of magnitude harder than Sycamore's 53-qubit result. arXiv (USTC) · Nov 12, 2026
  5. Physicists Feng Pan, Keyang Chen and Pan Zhang reported that a tensor-network algorithm run on a 512-GPU cluster solved the Sycamore sampling problem in about 15 hours, and estimated an exascale supercomputer could do the same task in a few dozen seconds, far below Google's 10,000-year classical estimate for that circuit. arXiv · Nov 12, 2026
  6. A team posted a paper on arXiv in June 2024 saying it used 1,432 Nvidia A100 GPUs to generate 3 million uncorrelated samples of the Sycamore circuit in about 86 seconds, a rate it said was seven times faster than Google's original 53-qubit experiment, using about 13.7 kWh of electricity. arXiv · Nov 12, 2026
  7. Harvard, MIT and QuEra reported in Nature in December 2023 that their 48-logical-qubit neutral-atom processor ran instantaneous quantum polynomial-time (IQP) circuits the team said would become rapidly harder to simulate classically as more circuit layers were added, part of a wider demonstration of error-corrected logical-qubit operations. Nature · Nov 12, 2026
  8. In a February 2024 paper, IBM's Dmitri Maslov and Sergey Bravyi with IonQ researchers said a classical algorithm they built computed an amplitude for the Harvard/QuEra 48-qubit IQP circuit in about 0.0026 seconds, roughly 1,000 times faster than the simulation time the Harvard/QuEra team had reported, challenging the claim that those circuits were classically intractable. arXiv · Nov 12, 2026
  9. D-Wave said in a paper published in Science on 12 March 2025 that its Advantage2 annealing prototype simulated the dynamics of programmable spin-glass magnetic materials in minutes, a task it said would take the Frontier supercomputer at Oak Ridge almost one million years and more electricity than the world uses in a year. D-Wave Quantum · Nov 12, 2026
  10. Joseph Tindall and colleagues at the Flatiron Institute posted a paper on arXiv on 7 March 2025 describing a classical tensor-network method that matched or beat D-Wave's Advantage2 results on several of the same spin-glass lattices, verified on systems involving hundreds of qubits, disputing D-Wave's quantum-supremacy claim. arXiv (Flatiron Institute) · Nov 12, 2026
  11. Gleb Kalachev, Pavel Panteleev, PengFei Zhou and Man-Hong Yung posted a paper on arXiv on 30 December 2021 describing a bounded-fidelity classical sampling algorithm, and estimated that for Zuchongzhi's 56-qubit, 20-cycle circuit it could produce 1 million samples at 0.066% fidelity on the Selene supercomputer, using 4,480 Nvidia A100 GPUs, in about 4 days, narrowing the classical cost of that claim. arXiv · Nov 12, 2026