Researchers introduced a benchmark that estimates the fidelity of Greenberger–Horne–Zeilinger (GHZ) states using Direct Fidelity Estimation (DFE). The benchmark searches for the largest GHZ state whose estimated fidelity is compatible, within statistical error, with the genuine multipartite entanglement (GME) threshold $F=1/2$ – the point at which a system can be said to share entanglement among all its parts.

The test was run on nine commercially accessible quantum processors accessed over the cloud, without any error‑mitigation tricks. The results reveal a large gap between the total number of physical qubits a device offers and the size of the largest GHZ state that actually passes the benchmark. On the best superconducting processors – IBM Phoenix (Nighthawk r2) and IBM Basquecountry (Heron r2) – GHZ states up to 27 and 24 qubits, respectively, survived the test. These numbers are less than a quarter of those devices’ full register sizes. In contrast, the 12‑qubit trapped‑ion AQT IBEX Q1 device succeeded with its full 12‑qubit register.

A key finding is that the benchmark outcome depends strongly on how strict the decision rule is. If the rule requires a certified GME threshold (largest GHZ size for which even the lower bound of the statistical error exceeds ½), only four devices succeed at any tested size, down from nine when a looser criterion is used. Moreover, the usually reported median calibration data provide poor predictions of the benchmark result.

Overall, the study illustrates that current quantum processors can produce large clusters of entangled qubits, but real‑world performance – measured by a statistical fidelity test – lags far behind the nominal qubit counts. This benchmark provides a concrete baseline for future cross‑platform comparisons and highlights the importance of rigorous statistical validation.