70 Logical Qubits and a 15-Minute Miracle: IBM Claims Quantum Advantage

On July 30, 2026 — yesterday, as I write this — IBM and researchers at the University of Chicago announced something that quantum computing scientists have been chasing for nearly four decades. They demonstrated what they call quantum advantage: a computation performed on a quantum computer that surpasses the practical capabilities of the best classical supercomputers, with statistical proof that the result is correct. The machine used 70 logical qubits, encoded with error correction, and finished the task in approximately 15 minutes. Leading classical simulation methods, by IBM's analysis, would require infeasible amounts of time.

This is not another benchmark paper. This is not a carefully chosen problem designed to make a noisy quantum chip look good. This is a structured quantum computation with verifiable fidelity, encoded in a quantum error-correcting code, achieving effective error rates ten times lower than the physical hardware would permit without encoding. The result is now openly tracked on the Quantum Advantage Tracker, and the paper, "Sampling hard circuits with verifiably high fidelity," is on arXiv.[1]

The Verification Problem

For years, the gold standard for claiming quantum advantage has been random circuit sampling (RCS). The idea is elegant: build a quantum circuit so complex that its output distribution is infeasible to simulate classically, then sample from that distribution. If the samples pass certain statistical tests, you have evidence — circumstantial, but strong — that a quantum computer did something no classical machine could replicate.

The trouble is verification. As the circuits grow harder, checking that the quantum computer's answer is correct becomes as difficult as simulating the circuit itself. At some point, you are reduced to trusting the device. You cannot prove it. You can only argue that cheating would be harder than doing the computation honestly. This is not a satisfying foundation for a technology that aspires to solve problems in cryptography, drug discovery, and materials science.

IBM and UChicago's new approach solves this by trading pure randomness for structure. They construct circuits that retain the computational hardness of RCS — proved rigorously in the paper — but embed that hardness within a framework that admits error detection. The circuit is encoded in a quantum error-correcting code, specifically a spacetime code, which allows the team to measure syndromes and certify the fidelity of the computation without classically simulating the full output.

The result is a statistical certificate: a lower bound on the fidelity of the quantum state, established with 95% confidence, that requires substantially weaker assumptions about the hardware than existing benchmarks. The device is not a black box whose inner workings must be trusted. It is a code whose syndromes can be checked.

Seventy Logical Qubits

The numbers deserve attention. The experiment used 97 physical qubits to encode 70 logical qubits — a ratio that will improve as codes become more efficient, but that already represents one of the largest logical quantum computations ever performed. The logical circuit executed 2,415 two-qubit gates and 468 T gates, a measure of non-Clifford operations that are essential for universal quantum computation and notoriously difficult to implement fault-tolerantly.

Because the computation was encoded, the effective logical error rates were approximately ten times lower than the physical error rates of the underlying hardware. This is the promise of quantum error correction realized: not just detecting errors, but suppressing them enough that a complex computation survives to the end with its quantum coherence intact. The paper reports a fidelity lower bound of 0.284 with 95% confidence — meaning the quantum state produced by the circuit is at least 28.4% faithful to the ideal, noise-free state. That may sound modest, but for a 70-qubit, depth-70 circuit with nearly 2,500 entangling operations, it is extraordinary.

The classical verification, meanwhile, hits a wall. The IBM team tested multiple leading classical simulation methods — tensor network contractions, stabilizer decompositions, and others — and found that all faced prohibitive runtimes for circuits of this size. The quantum computer's 15-minute execution time was not merely faster; it was in a different computational regime entirely.

Three Demonstrations in One Day

The IBM-UChicago result was not announced in isolation. On the same day, IBM published two companion demonstrations of quantum advantage with trusted computation. In collaboration with Qedma Quantum Computing, IBM used Qedma's quantum error mitigation software (QESEM) to model quantum dynamics in a 74-qubit system — observing long-lasting quantum oscillations that state-of-the-art classical simulations could not reproduce. In a third experiment with Algorithmiq, IBM established a framework for trusted quantum computation beyond classical verification.

The coordination is strategic. IBM is making a statement: quantum advantage is not a single experiment, not a one-off headline, but an emerging capability that multiple groups are now achieving independently and simultaneously. Jay Gambetta, Director of IBM Research, put it directly: "We are now firmly in the quantum advantage era."[2]

Whether that claim holds depends on what you mean by "quantum advantage." The problem solved here — sampling from a structured quantum circuit — is not a practical application. It is a benchmark, designed to be hard for classical computers and feasible for quantum ones. No one will use this to design a drug or break a cryptographic code. But it is a benchmark that meets the formal criteria for quantum advantage: classically intractable, verifiably correct, and achieved with encoded logical qubits rather than raw physical ones.

What "Advantage" Means

The term "quantum advantage" has been contested since Google claimed it in 2019 with their 53-qubit Sycamore processor. Critics argued that the classical simulation could be improved. IBM itself disputed Google's claim, publishing a preprint showing that a classical supercomputer could perform the same computation in a few days rather than the 10,000 years Google had claimed. The advantage, if it existed, was narrow and possibly temporary.

This time, IBM has been more careful. They explicitly benchmarked against the best known classical methods and showed that all face exponential scaling. They released their circuits and data openly. They encoded their computation in an error-correcting code, addressing the criticism that noisy physical qubits cannot be trusted. And they provided a statistical certificate of fidelity, not just a benchmark score.

Bill Fefferman, the UChicago computer scientist who co-led the research, framed the advance as methodological: "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."[2] The goal is not merely to claim advantage but to make the claim falsifiable.

The Road Ahead

Where does this lead? The immediate next step is scale. Seventy logical qubits is a milestone, but useful quantum computations — simulating catalytic reactions, factoring large integers, optimizing logistics networks — will require thousands or millions of logical qubits. The ratio of physical to logical qubits must improve. The error rates must drop further. The classical verification methods must be pushed back even farther.

The spacetime code used in this experiment is a promising direction. Unlike surface codes, which encode logical qubits in a two-dimensional lattice of physical qubits, spacetime codes exploit the temporal dimension of circuit execution to detect errors more efficiently. The IBM team showed that syndrome post-selection — discarding runs where error syndromes indicate too many faults — can suppress effective error rates dramatically. This is a pragmatic approach: rather than correcting errors in real time, detect them and retry.

But the deeper question is whether quantum advantage on benchmarks translates to quantum advantage on problems that matter. History is littered with technologies that excelled at test problems and failed at real ones. The IBM result is a genuine advance in the foundations of quantum computing — error correction, verification, and the rigorous demonstration of classical intractability. Whether those foundations will support a skyscraper of practical applications remains the open question that has defined the field since Richard Feynman first proposed quantum computers in 1981.

For now, the achievement stands on its own. A quantum computer, running a computation encoded in 70 logical qubits, produced a result that the best classical methods cannot verify in any reasonable time — and the result itself carries a statistical certificate of its own correctness. Fifteen minutes. Ten times lower error rates. A proof, not a promise. The quantum advantage era, if it has arrived, looks different from the hype: quieter, more careful, and grounded in error correction rather than raw qubit count.