Quantum Litmus — The Daily Reality Check for Quantum Computing
Independent, evidence-first analysis of what changed in quantum computing today, why it matters, and what the evidence does not yet show.
Sandia-led benchmark puts today’s quantum circuit capability about five orders below utility targets
Direct tests on Google, IBM and Quantinuum systems introduce one comparable measure. The result is a preprint, and no useful workload was demonstrated.
Today’s top signal
A Sandia-led team has measured Google, IBM and Quantinuum processors with one cross-platform circuit-capability benchmark.
In plain English
A laptop is not judged only by the number of transistors inside it. Quantum computers have the same problem: qubit counts, gate accuracy and speed each describe only part of what a complete machine can do.
A team led by Sandia National Laboratories, with researchers from Quantinuum, NVIDIA and the University of New Mexico, proposes a benchmark called QUOPS. It asks how large a representative quantum circuit a machine can execute successfully and how quickly it can do so. The team ran the same benchmark on Google’s 105-qubit Willow processor, IBM’s 156-qubit ibm_boston processor and Quantinuum’s 56-qubit H2-1 and 98-qubit Helios-1 systems.
The best direct physical-qubit score was 1,824 QUOPS on Helios-1 when runs showing leaked ions were discarded. Willow and IBM scored 216 and 204; H2-1 reached 1,392 with the same kind of postselection. The team also ran a fault-tolerant version on as many as eight small encoded qubits on Helios-1, reaching 40 QUOPS. These numbers are not qubit counts or useful calculations. They summarize the size of benchmark circuits that crossed a defined success threshold.
Why this matters: competing machines use different qubit technologies, connectivity and control systems, so headline specifications are difficult to compare. One architecture-neutral test can expose trade-offs. The superconducting machines ran much faster, while the trapped-ion machines completed larger benchmark circuits. Error mitigation enlarged the apparent circuit capability but reduced throughput by three to four orders of magnitude.
The boundary is equally important. The paper converts published resource estimates for factoring RSA-2048 and modelling the FeMoco molecule into targets of roughly 250 million and 340 million QUOPS. Current direct scores are about five orders of magnitude smaller. But those targets are estimates tied to particular algorithms, error models and a new metric. The work is a preprint, has not been independently reproduced, and several authors work for one tested vendor. No useful problem, customer workload, practical classical comparison or commercial economics were demonstrated.
This is a potentially useful ruler for measuring progress. It is not evidence that useful quantum computing has arrived, so Quantum Litmus keeps MONITOR and the 2030–2033 window unchanged.
A team led by Sandia National Laboratories, with researchers from Quantinuum, NVIDIA and the University of New Mexico, proposes a benchmark called QUOPS. It asks how large a representative quantum circuit a machine can execute successfully and how quickly it can do so. The team ran the same benchmark on Google’s 105-qubit Willow processor, IBM’s 156-qubit ibm_boston processor and Quantinuum’s 56-qubit H2-1 and 98-qubit Helios-1 systems.
The best direct physical-qubit score was 1,824 QUOPS on Helios-1 when runs showing leaked ions were discarded. Willow and IBM scored 216 and 204; H2-1 reached 1,392 with the same kind of postselection. The team also ran a fault-tolerant version on as many as eight small encoded qubits on Helios-1, reaching 40 QUOPS. These numbers are not qubit counts or useful calculations. They summarize the size of benchmark circuits that crossed a defined success threshold.
Why this matters: competing machines use different qubit technologies, connectivity and control systems, so headline specifications are difficult to compare. One architecture-neutral test can expose trade-offs. The superconducting machines ran much faster, while the trapped-ion machines completed larger benchmark circuits. Error mitigation enlarged the apparent circuit capability but reduced throughput by three to four orders of magnitude.
The boundary is equally important. The paper converts published resource estimates for factoring RSA-2048 and modelling the FeMoco molecule into targets of roughly 250 million and 340 million QUOPS. Current direct scores are about five orders of magnitude smaller. But those targets are estimates tied to particular algorithms, error models and a new metric. The work is a preprint, has not been independently reproduced, and several authors work for one tested vendor. No useful problem, customer workload, practical classical comparison or commercial economics were demonstrated.
This is a potentially useful ruler for measuring progress. It is not evidence that useful quantum computing has arrived, so Quantum Litmus keeps MONITOR and the 2030–2033 window unchanged.
Quantum Litmus assessmentMONITOR
Commercial Readiness Outlook — Industry
Early
Estimated broad enterprise window: 2030–2033
Today: What moved: a Sandia-led team directly applied one cross-platform benchmark to Google Willow, IBM ibm_boston and Quantinuum H2-1 and Helios-1, including a small logical-qubit implementation. What did not move: the benchmark did not run RSA-2048, FeMoco or another useful workload, and its utility targets are resource-model estimates, so the 2030–2033 readiness window is unchanged.