A study reports that IBM’s 120-qubit Nighthawk r2 generated 1 million random-circuit-sampling (RCS) outputs in 19 seconds using 61 qubits. It estimates that simulating the same ensemble on Frontier would take more than a century under its stated model and assumptions.
IBM Nighthawk r2’s 19-second sampling result
The runtime figures describe the same one-million-sample task, but they have different bases: one is the reported quantum run, while the other is a classical simulation estimate.
| System and role | Time for 1 million samples | Basis and conditions |
| IBM Nighthawk r2, quantum processor | 19 seconds | Reported run using 61 of the processor’s 120 qubits |
| Frontier, classical supercomputer | More than 100 years | Modeled estimate under the study’s simulation assumptions and favorable-memory conditions |
What random-circuit sampling tests
RCS is a benchmark in which a quantum processor generates samples from randomly constructed circuits. The result concerns that defined task, not a routine application such as running ordinary software.
The study reports that the experiment used 61 qubits on a square-lattice configuration with native CZ gates. It ran through IBM’s standard cloud execution stack without benchmark-specific calibration. At the 36-cycle comparison point, the circuits contained 918 two-qubit gates.
How the study assessed its samples
The study reports agreement between mirror benchmarking and three- and four-patch cross-entropy benchmarking (XEB) estimators at their respective measured depths. Mirror benchmarking covered 4 to 40 cycles; the patch-based XEB estimators covered 20 to 40 cycles.
At 36 cycles, the reported XEB fidelity was 2.3 × 10⁻³. For the modeled classical task at that point, the study estimated 1.2 × 10²⁷ machine operations using a bounded-fidelity rejection-sampling model and favorable memory assumptions.
What this quantum advantage shows
The more-than-century figure is an estimate derived from a classical simulation model, not a completed run on Frontier. Its scale depends on the study’s tensor-network and sampling assumptions, including the favorable-memory conditions.
The result is specific to one RCS benchmark and its modeled classical comparison. It does not establish that quantum computers are generally faster across different workloads.
