There is no single benchmark score that ranks quantum computers for every purpose. A useful comparison considers quality, scale, and speed alongside results on workloads resembling the task you care about—and checks how each score was produced.
What a benchmark score can—and cannot—tell you
A benchmark measures performance under a defined workload and protocol. It does not, by itself, establish how well a device will run every useful algorithm or which system is best overall. IBM describes quantum-computer performance in terms of quality, scale, and speed; application-oriented benchmarks add evidence about relevance to particular workloads.
As an Amazon Associate I earn from qualifying purchases.
Treat a result as conditional on its benchmark version, circuits, selected device or qubits, compilation and mapping, optimizations, error mitigation, sampling, and timing boundaries. If those conditions differ, two headline numbers may not be a direct comparison.
What the main benchmark families measure
Gate quality: randomized benchmarking
Randomized benchmarking estimates the performance of tested gate operations under a specified protocol. NIST describes it as a platform-independent method for assessing one-qubit rotations and reports an extension for multiqubit Clifford unitaries. A result characterizes the operations and protocol tested; it is not a measure of all application behavior.
System-level circuit capability: Quantum Volume
Quantum Volume tests whether a device can produce correct outputs for square random circuits, with circuit width and depth scaling together. It offers a composite view of performance on that workload family, not a direct measure of execution speed or every useful algorithm. Report the tested score together with its protocol and device configuration.
Execution speed: CLOPS
CLOPS is a workload-specific indicator of circuit-execution speed. IBM describes it as including quantum execution and the classical computation needed to process and update circuits. The hardware-aware CLOPSh definition reflects how circuits run after compilation. State which definition a result uses and what its timing includes; a speed score is not meaningful without those boundaries.
Rank #2
Application relevance: QED-C benchmarks
The QED-C application-oriented suite tests algorithms and small applications over varying problem sizes, examining output fidelity across circuit width and depth. Its measures are intended to reflect quality and time-to-solution aspects. The repository describes its examples as reference prototypes, not performance standards, so identify the exact benchmark and suite revision rather than treating a suite result as a universal ranking.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Compare results on the dimensions that matter
| Dimension | What to compare | Why it matters |
|---|---|---|
| Quality | Gate error or circuit output fidelity, benchmark protocol, and mitigation | Results depend on which operations are measured and whether mitigation is used. |
| Scale | Usable qubits and reliable circuit width and depth under the same test | Nominal qubit count alone does not establish reliable computation. |
| Speed | Circuit-layer throughput and the execution or classical steps included in timing | Different CLOPS definitions and timing boundaries can represent different work. |
| Workload relevance | Algorithm or application family, problem sizes, and output measure | A representative task connects benchmark performance to the intended use. |
| Reproducibility and fairness | Protocol version, compiler, mapping, device, sampling, optimizations, and reporting | Defined methods and disclosed overheads make results easier to interpret and reproduce. |
A practical workflow for comparing quantum computers
- Define the question. Decide whether you need evidence about gate quality, maximum reliable circuit size, execution throughput, or performance on an algorithm resembling your application. A headline score cannot choose the workload for you.
- Match the protocols. Before placing scores side by side, check that the benchmark and version, circuit family, success statistic, device selection, and measurement conditions match. If they do not, describe the differences instead of presenting the numbers as a direct race. IBM Research authors argue that benchmarks should be randomized, well-defined, holistic, and device independent, and distinguish benchmarks from diagnostics.
- Use complementary measures. Pair a relevant quality or scale measure with a speed measure when throughput matters. Add an application-oriented result when task relevance matters. Do not substitute one metric for another: Quantum Volume is not a speed test, and a gate-quality result does not establish application performance.
- Inspect the full measurement path. Check whether circuit transpilation and mapping, device topology and selected qubits, parameter tuning, dynamical decoupling or other optimizations, error mitigation, classical computation, queueing, and calibration state are included. Record where timing starts and stops. IBM’s benchmark-design paper specifically warns that optimization differences and unreported error-mitigation overhead can materially affect results; its CLOPS description includes both quantum and classical work.
- State only what the result supports. If the evidence is a benchmark score, say that one system scored higher on that benchmark under that protocol. To claim that it is better for a task, compare the systems on a representative task and report success quality and time-to-solution with relevant overheads.
Why optimization and protocol details change interpretation
Scores are not isolated properties of a chip: choices in the benchmark run can change the observed outcome. In a 2023 IBM Research example using a 26-qubit mirror randomized-benchmarking circuit, the authors observed error per qubit change from 2% to 1% when dynamic decoupling was included. This is an example of sensitivity to an optimization, not a general expected improvement for other devices or workloads.
Historical experimental figures also need their original context. A NIST-listed 2012 study by Gaebler and colleagues reported 0.162 ± 0.008 error per random two-qubit Clifford unitary for the trapped-ion experiment described in its publication record. That result concerns a particular experiment and protocol; it is not a current cross-vendor comparison.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to answer “Which quantum computer is best?”
First specify the workload and the criterion: for example, tested gate quality, reliable circuit width and depth, throughput under a named CLOPS definition, or output quality and time-to-solution on an application-oriented benchmark. Then compare results only where protocols and conditions are sufficiently aligned. Without that context, “best” is not a conclusion a benchmark score can support.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches
