IBM × University of Chicago: 70 Logical Qubits Deliver Verifiable Quantum Advantage

> This article is part of the column "From Coins to Error Correction" (Q-057) · Column index: https://zhichai.net/topic/178636834

On July 30, 2026, IBM and the University of Chicago jointly announced a quantum computing demonstration that, for the first time, satisfied the two metrics that past quantum advantage demonstrations over the last decade found hardest to achieve simultaneously:

1. Computation results beyond the practical capability of the strongest classical simulation methods 2. A statistical confidence proof that "this result is trustworthy, not because we got lucky"

They used 70 logical qubits on an IBM Heron processor, with 2,415 logical two-qubit gates and 468 logical T gates, producing a result of "fidelity lower bound 0.284 + 95% confidence + 10× error-rate suppression." The full computation took 15 minutes — an equivalent classical simulation with the best known algorithms would require an "impractical" runtime.

The paper was posted to arXiv on July 30 as a preprint titled "Sampling hard circuits with verifiably high fidelity" (arXiv:2607.25941). IEEE Spectrum followed with coverage, noting that "the 95% confidence promise is something no previous method could do."

Why This Is a Watershed Moment

Since Google first proposed quantum advantage in 2019 using Random Circuit Sampling (RCS), the concept has been stuck on one question: how do you know what you're seeing isn't noise?

The core of RCS: run a quantum processor through random gate sequences to produce extremely complex outputs that classical computers cannot reproduce. If classical reproduction fails, quantum is declared the winner.

But if even classical machines can't simulate it, how do you prove the quantum result isn't "coincidentally spitting out strings that look complex but are actually noise"? Over the past 7 years, classical algorithms gradually caught up with Google's 2019 demonstration, fueling the narrative that "quantum advantage isn't that big after all."

IBM + Chicago's solution this time was not "write even more complex circuits," but making the circuit structure itself verifiable:

• Build circuits using an approach called a spacetime code
• Use mainly Clifford gates (which classical computers can simulate efficiently)
• Add a portion of non-Clifford gates (which make classical simulation complexity grow exponentially)
• Crucially, this structure allows error detection and discarding of failed runs during the computation

The result: classical simulation difficulty rises exponentially, while the quantum side can continuously detect failures, discard them, and retry — with retained runs carrying a 95%-confidence fidelity lower bound.

Jay Gambetta, Director of IBM Research and IBM Fellow, made a widely quoted statement:

> "We are already in the era of quantum advantage. We have demonstrated a quantum computation beyond the practical capability of classical computing, and given a lower bound on execution fidelity with statistical confidence. This provides a foundation of trust for researchers, developers, and enterprises when quantum systems are used for problems classical computing cannot reach."

The Data Foundation

• Processor: IBM Heron (tunable-coupler architecture first released in 2023)
• Qubits: 70 logical qubits (not physical — the difference is that logical qubits carry error correction)
• Operations: 2,415 logical two-qubit gates + 468 logical T gates
• Runtime: ~15 minutes
• Fidelity lower bound: 0.284 (95% confidence)
• Error-rate suppression: 10× via syndrome post-selection
• Classical simulation equivalent time: "impractical" (exact multiple undisclosed, but IEEE Spectrum reported that the 95% confidence came at the cost of running 860× more runs to offset the discard rate)
• Paper: arXiv:2607.25941, 9 authors including Simon Martiel
• Publication: posted to arXiv July 30, 2026, journal submission to follow

Why the "Verification Problem" Went Unsolved for a Decade

Quantum gates fall into two categories:

• Clifford gates: efficiently simulable on classical computers
• Non-Clifford gates: classical simulation complexity grows exponentially with their number

Google's 2019 RCS demonstration used many Clifford gates plus a small number of non-Clifford gates — pushing classical simulation cost to exponential levels, but simultaneously making verification impossible. If you can't compute it yourself, on what basis do you claim the result is correct?

The key to IBM + Chicago's approach: many Clifford gates, few non-Clifford gates, plus the spacetime-code structure. This means:

• Classical simulation remains hard (non-Clifford gates are still present)
• The computation can self-check (many Clifford gates allow the spacetime code to detect errors)
• Failed runs are discarded; only retained runs count toward the result

The cost: the error-detection overhead means 860× more runs than an unverified version. In exchange: a "trustworthy result" at 95% confidence.

The Key Questions Ahead

The "verification problem" has been the most criticized dilemma in quantum computing for the past decade — the inability to prove you're right. After Google's 2019 demonstration, classical algorithms kept closing in, and "the wall of quantum advantage" was repeatedly pushed lower.

IBM + Chicago's "trusted version of quantum advantage" means:

• Classical challengers can no longer casually claim to have caught up — with 95% confidence, catching up becomes difficult
• Quantum advantage begins to have quotable, reproducible, contestable engineering properties
• In the subsequent phase of finding applications, writing papers, and negotiating commercial contracts, quantum computing results now have verification backing

In the short term, IBM's own roadmap will port this "logical qubit + spacetime code" design to the next-generation Kookaburra processor; in the mid term, partners Algorithmiq and Qedma will push the "verifiable" approach to mid-scale NISQ systems; in the long term, US Department of Energy national laboratories will begin integrating this method into quantum + AI + classical hybrid computing workflows.

Why This Matters Beyond "IBM Showed Off Quantum Advantage Again"

For the past decade, "quantum advantage" has been a somewhat self-affirming, demonstration-style proposition — you couldn't prove your demo was trustworthy, so classical algorithms kept catching up and you kept "redefining the boundary."

What IBM + Chicago delivered is not "quantum advantage with bigger numbers," but "quantum advantage with statistical confidence." It advances quantum computing from "demonstration" to "verifiable task."

Concretely for the industry chain:

• Classical supercomputing center operators can use this paper as the answer to "why the quantum part isn't just for show"
• R&D leaders in pharma, logistics, batteries, and AI can start seriously evaluating "should we try doing something real with quantum"
• Hybrid "quantum + AI + classical" computing centers such as the US Department of Energy, Argonne National Laboratory, and Oak Ridge National Laboratory can consider integrating this verification methodology into their workflows

One final cold fact: simulating the equivalent problem classically with "the strongest practical classical simulation algorithms" requires an "impractical runtime." That "impractical" figure was not disclosed, because even on IBM's best classical simulation hardware it is an astronomical number.

But IEEE Spectrum made a key point: "A single full run of this experiment takes 15 minutes, which strips classical simulation of any commercial relevance."

In other words: this is the first quantum advantage demonstration in a decade where the classical side has no room to say "just wait, I'll write another algorithm and catch up." The only remaining task is "proving the result is correct" — and the 95% confidence solves exactly that.

---

• Sources: IBM Research press release (July 30), IEEE Spectrum detailed report (early August), phys.org technical review (August 1), Machine Herald retrospective commentary (August 5)
• First principles: 70 logical qubits + 2,415 two-qubit gates + 468 T gates + spacetime code + syndrome post-selection
• Key figures: fidelity lower bound 0.284 / 95% confidence / 860× run-count cost
• Historic significance: the first time quantum computing was simultaneously established on both "trustworthiness" and "beyond-classical" dimensions
• International coverage: independently cross-confirmed by IEEE Spectrum / phys.org / Machine Herald / IBM press release