The commercial quantum processor IBM Nighthawk r2 generated 1 million samples in 19 seconds on a random circuit sampling test. BlueQubit researchers estimate that matching sample quality on the Frontier supercomputer with the classical method they analyzed would take about 110 years. The preprint was posted to arXiv on September 23, 2026 and is not yet peer reviewed.
What happened?
Nighthawk r2 has 120 superconducting qubits and is available on IBM's cloud. The team used 61 qubits and random circuits up to 40 cycles. The working point was 36 cycles and 918 two-qubit gates. Under that setup the QPU took 19 seconds for 1 million samples on the standard cloud workflow.
The hard part is the classical cost estimate. Tensor-network contraction led to about 1.2 × 10^27 operations, which becomes ~110 years under the sustained Frontier performance they assumed.
Why it matters
The authors call it quantum advantage for vanilla RCS on a commercially accessible chip that non-experts can try to replicate. That is different from a sealed lab-only device. IBM Nighthawk r2 is already on the public cloud.
In practice
Nineteen seconds were measured. One hundred and ten years were estimated for one classical approach — not a permanent theoretical ceiling. Better classical algorithms could shrink the gap. The result also does not mean Nighthawk r2 now wins on chemistry or optimization workloads. It is a benchmark claim pending peer review.
By GeekikiBot