Where to run a circuit

Read live from each provider's own API when this page loaded. These are the vendor's claims about its own hardware, not measurements Qly made. Medians, never means — a handful of dead sites drags a mean orders of magnitude away from what a circuit sees — with the census behind each median shown underneath, because the count of dead sites is exactly what a median hides.

Every run of one circuit on real hardware, as individual submissions. Nothing is averaged across days: a machine in June is not the same machine in September, and a mean over both describes neither. The shot count sits beside every number, and a run too small to read prints that instead of a score.

OPENQASM 2.0;
include "qelib1.inc";
qreg q[2];
creg meas[2];
h q[0];
cx q[0],q[1];
barrier q[0],q[1];
measure q[0] -> meas[0];
measure q[1] -> meas[1];

Ordered by how many readable runs each machine has, not by score — ordering by score would put a machine that ran once above one that ran twice, on a difference of a tenth of a point. A run under 32 shots is listed but has no reading and is left out of the median.

This is the question worth coming back for. Comparing a June run against today's calibration at the top of this page would compare it against a different machine wearing the same name — IQM recalibrates daily, and Rigetti's Cepheus published six dead qubits and thirty-one dead pairs on one day and a different set on another. Qly did not record the calibration a circuit was submitted under until 2026-09-20, so for every earlier run in section 02 this comparison is not available and never will be. It now fills one row at a time, as runs carrying the snapshot complete.

One row is publishable, not conclusive. It shows whether this run and the vendor's claim sit in the same neighbourhood; it cannot show whether that holds on another day or another machine. Three runs on one machine, on different days, would start to say whether the claim tracks the outcome or drifts.

Stated here rather than discovered later.

• One two-qubit circuit is not a benchmark. It says nothing about a machine at width eight or depth thirty, where the interesting failures are.
• The measured runs were not designed as an experiment. They are what people ran, which is why they are all the editor's default example, and why the shot counts and dates are uneven.
• Which physical qubits a circuit landed on is recorded for almost none of these runs. A two-qubit circuit on a good pair and on a bad pair of the same chip are different experiments, and we mostly cannot tell which happened.
• Section 01 and section 02 were not taken at the same time and are not two views of one thing. That is what section 03 is for.
• Some machines above have vendor calibration but no measured run here yet, and some have runs but no live calibration. Neither absence is a verdict on the machine.
• Emerald and Cepheus-1-108Q ran this circuit (4 runs and 5 runs respectively) but are not in section 02 — one run belongs to another account and the rest cannot be separated from it yet. They are listed as withheld at the bottom of that section rather than silently absent.