IBM and University of Chicago Demonstrate Verifiable Quantum Advantage on 70 Logical Qubits

Overview

IBM and researchers from the University of Chicago announced on July 30 a quantum computing demonstration that the two organizations say meets what has long been the field’s hardest combined bar: a computation beyond the practical reach of leading classical simulation methods, paired with statistical confidence that the result is actually correct, according to IBM’s newsroom announcement. The claim, drawn from a paper titled “Sampling hard circuits with verifiably high fidelity,” was independently described the same way by Phys.org, which reported it as “the point where quantum computers can be confirmed to have outperformed classical computers on trusted computations.”

What We Know

The team ran a 70-qubit circuit on an IBM Heron processor, executing 2,415 logical two-qubit operations and 468 logical “T gates” — metrics used to quantify circuit complexity — while shielding the qubits from errors, according to both IBM and Phys.org, which called it “one of the world’s largest error-correction demonstrations to date.” The computation took approximately 15 minutes, a runtime both outlets say was prohibitive for many leading classical simulation approaches.

The underlying paper, posted to arXiv by lead author Simon Martiel and eight co-authors including University of Chicago’s Bill Fefferman and Soumik Ghosh and IBM’s Jay Gambetta, reports that the 70-qubit circuit achieved “a fidelity lower bound of 0.284 with 95% confidence while suppressing gate errors by 10× after syndrome post-selection,” according to the arXiv preprint. That fidelity figure — roughly 28 percent of retained runs executing faithfully — was also reported by IEEE Spectrum, which noted the 95 percent confidence guarantee is something “no previous approach has been able to provide.” IEEE Spectrum reported that reaching that confidence level came at a cost: the error-detection scheme’s overhead of discarded runs meant the team had to perform 860 times more runs than they would have needed without it.

The technique builds on random circuit sampling (RCS), the approach that underpinned Google’s contested 2019 quantum advantage claim, according to IEEE Spectrum. Rather than generating fully random circuits, the IBM-UChicago team built circuits out of “Clifford gates” — which classical computers can simulate efficiently — using a method they call a “spacetime code” that can detect when errors occur and discard failed runs. They then added a limited number of “non-Clifford gates,” which make classical simulation exponentially harder, in places chosen not to disturb the error-detection process, IEEE Spectrum reported. Jay Gambetta, director of IBM Research and an IBM Fellow, said in a press briefing covered by IEEE Spectrum that the 468 non-Clifford gates used were “more than double what is classically simulatable.”

“Verification remains one of the biggest challenges in firmly establishing experimental quantum advantage,” said Bill Fefferman, an associate professor at the University of Chicago, in the IBM announcement. “This experiment develops techniques to better characterize the fidelity of hard quantum states under noise, increasing confidence that the quantum computer is solving a computationally hard problem.” Soumik Ghosh, a graduate student in Fefferman’s group, added: “Beyond strengthening experimental validation, advances in verification have the potential to unlock practical applications for the next generation of quantum computers,” according to Phys.org.

“We are now firmly in the quantum advantage era,” Gambetta said in the IBM announcement. “We have demonstrated a quantum computation beyond the practical reach of classical computers that establishes, with statistical confidence, a lower bound on how faithfully it was executed. This milestone gives scientists, developers, and businesses a new foundation for trusting quantum computers as they scale to problems far beyond what we can achieve classically.”

Two Companion Papers, Two Different Approaches

The UChicago collaboration was one of three papers IBM and partners released the same day, according to IEEE Spectrum, which described the trio as tackling verification “from a different angle.” In a paper with startup Qedma, researchers used an IBM Heron processor to simulate oscillatory behavior in a theoretical model of a magnetic material and compared results against classical techniques running on Japan’s Fugaku supercomputer and an Nvidia H100 GPU server. All three methods produced matching oscillation patterns up to 35 qubits, IEEE Spectrum reported, but classical methods held only briefly at 51 qubits and could not handle 74 qubits at all. To build confidence in the result, the Qedma team ran the same simulation on Quantinuum’s H2 and Helios trapped-ion machines and observed matching behavior. “The results we get with Quantinuum…are in perfect match with what we got from IBM, which really gives us very strong confidence in the accuracy of these results,” said Netanel Lindner, Qedma’s chief technology officer, per IEEE Spectrum.

A third paper, with startup Algorithmiq, ran a 56-qubit circuit on an IBM Heron processor and compared results against several classical approaches, finding that in the most difficult regimes the classical methods disagreed with both the quantum results and each other, IEEE Spectrum reported. “This consistency is not coincidence,” Algorithmiq CEO Sabrina Maniscalco said in the briefing. “It’s evidence.”

What We Don’t Know

All three papers have yet to undergo peer review, according to IEEE Spectrum. Outside physicists quoted by the outlet offered a mixed assessment. Dominik Hangleiter, a postdoctoral researcher at ETH Zurich, said “all three papers are good science,” according to IEEE Spectrum, which reported that Hangleiter viewed the UChicago team’s shift from fully random circuits to more structured ones enabling verification as a promising direction he had proposed himself in a paper the previous year. IEEE Spectrum reported he was more skeptical of whether the Algorithmiq or Qedma work demonstrates a clear advantage, since neither paper makes that claim outright. “Both of them seem to say, ‘Oh yeah, this seems to be hard to simulate.’ Which I think is good. They shouldn’t make much stronger claims than that,” Hangleiter said.

Jens Eisert, a professor at the Free University of Berlin, cautioned in written comments to IEEE Spectrum that establishing quantum advantage is not a single milestone to be crossed but an ongoing process, writing: “Verification of quantum simulations in regimes beyond the reach of straightforward classical simulation is not a single procedure, but rather a process of building confidence through a portfolio of complementary validation methods.”

Analysis

The distinction the three papers are drawing is a narrow but consequential one for the field: it is one thing to run a computation no classical machine can replicate, and another to prove that computation’s output can be trusted when there is no classical answer to check it against. The UChicago collaboration’s own results illustrate the trade-off starkly — reaching 95 percent statistical confidence in the outcome required, per IEEE Spectrum’s reporting, running the circuit 860 times more often than a version without error detection would have needed, and even then only 28 percent of retained runs executed faithfully. That the Qedma and Algorithmiq teams stopped short of claiming outright “advantage” in their own papers, as Hangleiter noted, suggests the industry itself is still calibrating how much evidence is enough before the term applies without qualification.