IBM Ran a 15-Minute Quantum Computation — and for the First Time Proved the Answer Was Right

Researchers at IBM and the University of Chicago say they have run a computation on a quantum computer that classical machines cannot practically reproduce — and, for the first time in such a demonstration, produced mathematical proof that the quantum machine's answer was correct.

That second half is the part that has been missing from every previous claim. Quantum advantage demonstrations since 2019 have almost all used random circuit sampling, a task chosen precisely because it is hard for classical computers. The catch is circular: if a classical computer cannot reproduce the result, it also cannot check it. Several high-profile advantage claims have been deflated over the following years by cleverer classical algorithms, including one earlier this year that a laptop reproduced.

"Verification remains one of the biggest challenges in firmly establishing experimental quantum advantage," said Bill Fefferman, an associate professor at the University of Chicago who worked on the project with graduate student Soumik Ghosh. The team's answer was to design a structured alternative to random circuit sampling — a problem still hard for classical simulation, but built so that the fidelity of the quantum output can be independently certified with statistical confidence.

The run itself is one of the largest error-corrected computations yet performed. The machine executed 70 logical qubits — qubits assembled from many physical qubits with error-correction machinery layered on top, so that individual hardware faults do not corrupt the result — through 2,415 logical two-qubit operations and 468 logical T gates. T gates are the expensive ones: they are the operations that make a quantum circuit genuinely hard to simulate classically, and each one costs substantially more error-correction overhead than the simpler gates around it. Total runtime was about 15 minutes.

Against that, the team showed that the leading classical simulation methods would face what they describe as prohibitive runtimes on the same problem. The result was posted as "Sampling hard circuits with verifiably high fidelity," by Simon Martiel and colleagues, on the arXiv preprint server, and has not yet completed peer review.

The practical significance is narrower than the headline suggests — the sampling problem has no direct commercial use, and 70 logical qubits is still far short of what factoring or chemistry simulation would demand. What changes is the standard of evidence. A quantum advantage claim that comes with its own certificate of correctness is much harder to overturn with a better classical algorithm, because the argument no longer depends on nobody having found one.