IBM’s Nighthawk r2 produced 1 million random-circuit samples in 19 seconds — an operation estimated to take a top supercomputer about 110 years

IBM's Nighthawk r2, a commercially available 120-qubit superconducting processor, was used to perform a random-circuit sampling (RCS) experiment on 61 qubits. Researchers ran circuits up to 40 cycles using the cloud platform's standard workflow and found a performance sweet spot at 36 cycles (918 two-qubit gates). At that point the device produced 1 million samples in 19 seconds.

How the classical comparison was estimated

The team used tensor-network contraction methods to estimate how hard it would be for a classical supercomputer to reproduce the million samples. Their calculation produced an estimate of 1.2 × 10^27 computational operations. Using a conservative sustained-performance figure for Frontier, a former fastest exascale supercomputer, they translated that cost into roughly 110 years of classical computation.

Sycamore experiment, which aimed to demonstrate quantum advantage. Classical researchers responded with improved simulation techniques, and quantum teams continued scaling systems. This new result differs because it used a commercially accessible processor via a public cloud interface, rather than a specialized laboratory device or bespoke calibration.

The 110-year figure is an estimate tied to a particular classical simulation approach. It is not a fundamental bound on what classical machines might achieve. The history of quantum-versus-classical benchmarking shows classical algorithms and simulation strategies can improve and reduce estimated times. The authors acknowledge the possibility that more efficient classical methods could change the comparison.

Led by Tigran Sedrakyan, the paper (available as a preprint on arXiv) describes the work as the first demonstration of quantum advantage for a vanilla RCS task on a commercially and broadly accessible quantum processor that non-expert users can replicate using standard cloud tools.

Follow-up items include attempts by classical-computing groups to find more efficient simulation shortcuts, independent replication of the sampling on similar machines, and experiments that vary qubit selection, circuit depth, and error mitigation. Those developments will affect whether this performance remains out of reach for classical systems or whether new classical techniques narrow the gap.