iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
There is no single benchmark score that ranks quantum computers for every purpose. A sound comparison uses complementary measures of quality, scale and speed, then checks whether each test resembles the workload you care about. Treat any result as conditional on its benchmark protocol, compilation, hardware configuration, error mitigation and timing boundaries.
Start with the question you want a benchmark to answer
Before comparing scores, decide what performance means for your use case. You might want to know whether a system can run a deeper circuit reliably, how quickly it can process a particular circuit workload, or how well it performs on an algorithm similar to your application. Those are different questions, and a headline score cannot answer all of them.
- Gate quality: How accurately does the system perform the tested operations?
- Scale: How wide and deep a circuit can it run with useful output quality?
- Speed: How much circuit work can it complete under the metric’s timing rules?
- Workload relevance: How well does it perform on a representative algorithm or application?
IBM describes quantum-computing performance in terms of quality, scale and speed. These dimensions complement one another; none alone establishes that a system is best for a particular task.
What the main benchmark families tell you
Randomized benchmarking estimates gate performance
Randomized benchmarking uses sequences of operations to estimate how accurately a device performs the operations tested by a defined protocol. NIST describes it as a platform-independent way to assess one-qubit rotations and lists a multiqubit extension for Clifford unitaries. The result concerns those gates and that protocol; it is not a direct measure of all application behavior.
For context, a NIST-listed 2012 trapped-ion experiment by Gaebler and colleagues reported an error of 0.162 ± 0.008 per random two-qubit Clifford unitary. That historical result describes the experiment in that publication record, not a current cross-vendor comparison.
Quantum Volume tests square random circuits
Quantum Volume (QV) is a composite measure based on whether a device can produce correct outputs for square random circuits, with circuit width and depth increasing together. It can summarize performance on that workload family, but it does not directly measure execution speed or the performance of every useful algorithm. When quoting a QV result, identify the tested score and protocol rather than presenting the number as a universal ranking.
Rank #2
CLOPS measures a particular kind of execution throughput
Circuit Layer Operations Per Second (CLOPS) is a workload-specific speed indicator. IBM describes it as including quantum execution as well as classical computation used to process and update circuits. IBM also introduced a hardware-aware definition, CLOPSh, which reflects how circuits run after compilation. Because definitions and timing boundaries matter, name the version used and explain what work is included in the timed interval.
Application-oriented tests connect results to workloads
The QED-C application-oriented suite evaluates algorithms and small applications across varying problem sizes. It examines output fidelity over circuit width and depth, with measures intended to address both quality and time-to-solution. Its repository describes the programs as reference prototypes, not performance standards. Check the exact benchmark and suite revision before quoting a result; results from different applications or revisions should not be treated as a direct race without explaining the differences.
Compare results only when the protocols are comparable
Two scores are meaningful side by side only when they measure the same benchmark and version under sufficiently similar conditions. IBM Research’s 2023 benchmark-design paper argues for benchmarks that are randomized, well-defined, holistic and device-independent, and distinguishes benchmarks from diagnostics. A diagnostic can help explain or troubleshoot a system; it should not automatically be treated as a fair performance comparison.
Before placing scores next to each other, check the following:
- Benchmark version, circuit family and success statistic.
- Which device and qubits were tested, including any selected subset.
- Compilation, qubit mapping and optimization settings.
- Sampling or shot conditions and how output quality was calculated.
- Error mitigation and other circuit optimizations, including whether their overhead is counted.
- Calibration state and when measurements were taken.
- Timing start and stop points, including any classical computation included in the result.
These details can change what a number represents. IBM Research’s 2023 paper highlights that optimization differences and omitted error-mitigation overhead can materially affect a benchmark result. For example, its authors reported observed error per qubit changing from 2% to 1% in a 26-qubit mirror randomized-benchmarking circuit when dynamic decoupling was included. That is an example of sensitivity to an optimization choice, not a general expected improvement for other devices or workloads.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Use complementary measures instead of forcing one ranking
| Comparison axis | What to examine | What it helps establish |
|---|---|---|
| Quality | Gate error or circuit output fidelity, protocol and mitigation | How accurately the tested operations or circuits performed |
| Scale | Usable qubits and reliable circuit width and depth under the same test | How large a circuit the tested system handled at the reported quality |
| Speed | Circuit-layer throughput, metric definition and timed operations | How quickly the specified workload ran under the stated timing boundaries |
| Workload relevance | Algorithm or application family, problem sizes and output measure | How closely the test reflects the intended task |
| Reproducibility and fairness | Protocol version, compiler, mapping, device, sampling and optimization disclosures | Whether another reader can interpret and fairly reproduce the comparison |
Nominal qubit count alone does not establish reliable computation. Likewise, a high QV score does not show that a system is faster, and a CLOPS result does not by itself show that it produces better answers for an application. Use the axis that corresponds to the question, then consult the others for context.
Best Value
A practical workflow for comparing quantum computers
- Define the target. Write down whether you care about gate quality, reliable circuit size, execution throughput or an application-like task.
- Select a fitting benchmark. Use randomized benchmarking for the tested gate operations, QV for square random circuits, CLOPS for its circuit-throughput workload, or an application-oriented benchmark for a representative algorithm.
- Align the test conditions. Match the benchmark and version, circuit family, success statistic, device selection and measurement conditions. If they differ, describe the differences rather than calling the scores a direct comparison.
- Record overhead and configuration. Note compilation and mapping, device topology and selected qubits, tuning and optimization, mitigation, classical processing, sampling, calibration and the timing interval.
- State only what the result supports. A higher score supports the statement that a system scored higher on that benchmark under that protocol. A claim that it is better for a specific task needs a suitably representative test and relevant success-quality and time-to-solution results, with overheads reported.
How to read a claim that one quantum computer is “best”
Ask: best at what workload, under which benchmark version and protocol, and with which parts of the classical and quantum process counted? A defensible comparison names the workload, reports complementary quality, scale and speed measures where relevant, and discloses the configuration and overheads. If those conditions are missing, the score may still describe a result, but it cannot support a broad ranking.
Quick Recap
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.
