Quantum benchmarks explained

A quantum computer can’t be judged by one number. Its makers, and the researchers who test it, measure three different things: how big it is, how accurately it works, and how fast it runs.

Scale: how many qubits

A qubit is the quantum version of a bit. More qubits allow bigger calculations, so the count of physical qubits is the number most often quoted. On its own it says little: a machine with many qubits that make frequent mistakes can do less than a smaller, more accurate one.

Because qubits make mistakes, the long-term plan is to group many physical qubits into one error-corrected logical qubit. Logical qubits only count as progress if they fail less often than the physical ones they are made from, which is why we always show the logical error rate next to them.

Quality: how often it goes wrong

The basic steps of a quantum program are called gates. Two-qubit gate fidelity says how often a step involving two qubits comes out right: 99.9% means about one mistake in a thousand. Randomized benchmarking is the standard way to measure it.

Single gates can look good while the whole chip struggles, because qubits disturb their neighbours. Layer fidelity measures errors when many gates run at once.

Whole-machine tests

Some benchmarks test everything together. Quantum Volume finds the largest square circuit a machine runs reliably. QUOPS, introduced in 2026, measures both the size of circuit a machine can run and how fast it runs them. Cross-entropy benchmarking checks random circuits against the ideal answer.

Speed

CLOPS counts how many layers of a test circuit a machine runs per second, including the classical computers around it. Speed matters because many useful programs need millions of runs.

Real tasks

Algorithmic qubits and the Q-score test how well a machine runs standard algorithms or optimisation problems. They are closer to real use, but each covers only a few kinds of problem.

Why comparisons need care

The same benchmark can be run in different ways: a median across a chip or one best pair of qubits; with or without throwing away failed runs; on a 6-qubit or a 16-qubit circuit. A fair comparison needs the same set-up. That is why every result here belongs to a comparison group, and why we show who measured it and how solid the evidence is (our method).