Ask how good a quantum computer is and the easy answer is a qubit count. Fifty, a hundred, a thousand. It sounds like progress you can rank. But a machine with a thousand shaky qubits that lose their state in a microsecond may be useless, while a smaller, cleaner machine can run circuits the big one cannot finish. Qubit count is the horsepower rating on a car that might not have wheels.
Quantum volume was invented to fix that. Proposed by researchers at IBM in 2018 and later adopted by others, it is a single number meant to capture how much useful quantum computation a machine can actually do, not just how many qubits it nominally holds.
What the number actually measures
The test is built around a specific kind of circuit called a square circuit. You take some number of qubits, call it n, and run a circuit that is exactly n layers deep. Each layer scrambles the qubits with random two-qubit operations, pairing them up and applying gates drawn at random. Because the circuit is as deep as it is wide, both dimensions grow together. Adding a qubit means adding a layer, so errors have more chances to accumulate.
After running the circuit many times, you look at which output bit strings appear most often. A perfect quantum computer produces a lopsided distribution where certain outcomes, the so-called heavy outputs, show up more than others. A noisy machine smears that structure into randomness. The test asks whether your hardware still produces the heavy outputs often enough, above a fixed statistical threshold, to prove it ran the circuit faithfully.
The largest square circuit a machine can pass, using n qubits and n layers, sets the score. Quantum volume is two raised to that n. Pass at 5 qubits and you get 32. Pass at 6 and you get 64. Each extra qubit doubles the number, which is why quantum volume figures come in powers of two: 128, 512, 1024, and so on.
Why doubling matters
The exponential scale is deliberate. It rewards the hard thing, which is running a wider and deeper circuit at once, rather than the easy thing of stapling more idle qubits onto a chip. To raise your score you need every ingredient to improve together: gate fidelity, qubit lifetimes, how well qubits connect to their neighbors, how accurately you read them out, and how cleanly you can compile a random circuit onto the physical layout. A single weak link caps the whole number.
That is the appeal. Connectivity is a good example. On a machine where every qubit can talk to every other one, a random two-qubit gate is cheap. On a sparsely wired chip, the compiler has to insert extra swap operations to shuttle information across the grid, and each swap is another chance to fail. Quantum volume folds that penalty in automatically, which is one reason trapped-ion machines with all-to-all connectivity have posted strong scores relative to their modest qubit counts.
The limits of a single figure
Quantum volume has real weaknesses, and its own inventors have said as much. The most obvious is that it saturates. By definition it stops at the number of qubits in the device, so a 433-qubit processor cannot report a quantum volume beyond 2 to the 433, and in practice the achievable number is far smaller. More importantly, the test is expensive to run and grinds to a halt well before today's largest chips are fully characterized. Classically simulating the ideal answer, which you need to check the result, becomes intractable past roughly 50 qubits.
It also says nothing about speed. A machine could pass a high volume by running very slowly, which is why IBM introduced a separate throughput measure, CLOPS, to count how many circuit layers it executes per second. And because the benchmark uses random circuits rather than real algorithms, a good score does not guarantee your chemistry or optimization problem will run well. Other vendors have leaned on their own yardsticks, such as counting algorithmic qubits, partly because no single number satisfies everyone.
Why it still gets quoted
For all that, quantum volume did something valuable. It shifted the conversation away from raw qubit tallies and toward quality, and it gave engineers one honest, holistic target to chase. Watching a company's quantum volume climb over several years tells you the whole stack is getting better, not just that the marketing brochure has a bigger headline number. As the field moves toward error-corrected logical qubits, the exact metric will keep evolving. The lesson underneath it will not: the size of a quantum computer means little until you know how well it holds together.
