The Efficiency Revolution
Fewer evaluations means fewer circuit executions to pay for, so the evaluation count of an optimiser is a cost figure, not only a technical one. This article sets out the cost model. It cites no benchmark figure: our GHZ-6 result was withdrawn in August 2026 because its setup favoured the optimiser.
The Cost Challenge
Cloud-based quantum computing operates on a pay-per-use model where every circuit execution costs money. For enterprises exploring quantum solutions, the mathematics is straightforward:
Illustrative formula, not a measured result
cost per optimization = evaluations × price per evaluation
Use your own backend price and your own measured evaluation counts. No price on this page comes from a customer or a provider quote.
At a fixed price, halving the evaluations halves the cost of an optimisation. What a given optimiser saves you depends on your circuit, your backend and your price per evaluation, so measure it on your own workload.
Key Findings
Cost Reduction
Fewer evaluations means fewer circuit executions to pay for on a pay-per-use backend. The saving depends on your workload and your price.
Energy Use
Less compute time can mean less energy used per experiment.
Increased Throughput
Fewer evaluations per experiment leaves room for more experiments in the same time window.
No Benchmark Figure Today
We withdrew our GHZ-6 optimiser benchmark in August 2026 because its setup favoured the optimiser. Measure evaluation counts on your own workload before budgeting.
European Market Context
Fewer evaluations per experiment matter for two things that European teams care about:
Lower Energy Use
Less compute time can mean less energy used per experiment.
Cost Control
Fewer evaluations per experiment means fewer executions to pay for on a pay-per-use backend. The saving depends on your workload and your price.
How to Compare Optimisers Fairly
Start every run from random parameters, vary the order in which parameters are tuned, and count failed runs against the method. Our GHZ-6 benchmark started near the answer in a fixed order, so we withdrew it on 25 August 2026. The techniques inside the Autopilot are not described on this page.
Practical Takeaways
For CFOs
View sample efficiency as a lever for ROI, not just a technical benchmark. Fewer evaluations means fewer billed executions; measure the saving on your own workload before budgeting.
For Developers
Use QCOS to develop and tune circuits on a CPU state-vector simulator (up to 24 qubits) before spending time on a real backend. Efficiency leaves room for more experiments.
References
- 1.SoftQuantus. Withdrawn: the QCOS Autopilot GHZ-6 Benchmark (withdrawn 25 August 2026; do not cite its figures).