CLOPS
System-levelIBM's speed benchmark: the sustained number of circuit layers per second a quantum system and its classical stack execute while running a batch of parameterized model circuits.
CLOPS (Circuit Layer Operations Per Second) is IBM’s speed benchmark, proposed by Wack and colleagues in 2021 as one of the three headline attributes (quality, speed, and scale) for measuring near-term quantum computers. It deliberately measures the whole stack, not just the QPU: compilation, parameter binding, data movement, and execution all count against the clock, mirroring how iterative workloads such as variational algorithms run.
How it works
The protocol executes a large batch of parameterized model circuits: M circuit templates, each updated with K sets of parameters, run for S shots at D layers deep, giving CLOPS = M × K × S × D / (total wall-clock time). In the original 2021 definition the templates were Quantum Volume circuits with D = log2(QV), retroactively called CLOPS_v. In 2023 IBM redefined the benchmark around hardware-aware circuits (CLOPS_h): 100 qubits × 100 layers, each layer a round of two-qubit gates parallelizable on the device’s actual connectivity plus parameterized single-qubit rotations, adopting the layer notion of layer fidelity. IBM now reports CLOPS_h simply as “CLOPS”.
Strengths and limitations
CLOPS captures a dimension quality benchmarks ignore, throughput of the full hybrid loop, and leaves execution optimization to the vendor by design: it scores the system as delivered. The costs: two incompatible definitions share one name (about 2.4K CLOPS_v versus about 200K CLOPS_h on the same IBM Heron device), so cross-definition comparisons are meaningless. Because the classical stack is included, scores shift with software updates and are hard to compare across vendors, and the hardware-aware layer definition can flatter sparsely connected architectures. The defining paper is an arXiv draft explicitly requesting feedback, never journal-published; the redefinition lives only in an IBM blog post and documentation.
Notable results
As of July 2026 the IBM Quantum Platform lists a CLOPS value for every deployed QPU (roughly 24K to 340K across Heron and Nighthawk systems), shown alongside layered two-qubit error (derived from layer fidelity) and a newer measurement-cycles-per-second (MCPS) metric. IQM’s iqm-benchmarks suite also lists CLOPS among its holistic benchmarks.