computer science//quantum computing//quantum advantage

Quantum advantage is the demonstration that a quantum computer performs some task that is impractical, or far more expensive, for the best classical methods available, and it is the milestone the field uses to measure whether quantum hardware has crossed from physics experiment to computing tool. Under the older name **quantum supremacy** it described any such task, useful or not; the distinction that matters for an engineer is between the two meanings the word now carries.

Quantum advantage is the demonstration that a quantum computer performs some task that is impractical, or far more expensive, for the best classical methods available, and it is the milestone the field uses to measure whether quantum hardware has crossed from physics experiment to computing tool. Under the older name quantum supremacy it described any such task, useful or not; the distinction that matters for an engineer is between the two meanings the word now carries.

Advantage on a benchmark. A task designed to be hard to simulate classically and easy to run on the quantum chip, usually sampling the output of a random circuit. Google's Sycamore (53 qubits, 2019) claimed 200 seconds against an estimated 10,000 years on a supercomputer; IBM answered that a better classical method would take days, and later tensor-network simulations closed most of the gap. Willow (2024) repeated the claim with a margin of about 102510^{25}1025 years. These results show the hardware works at a scale classical machines cannot track; the output has no use.

Useful, fault-tolerant advantage. Solving a problem someone would pay for (a catalyst's energy levels, a material's behaviour, factoring a key) faster or cheaper than classically, with results reliable enough to trust. That needs logical qubits and long error-corrected computations (quantum error correction), and it has not been shown.

An advantage is always relative to the best classical method, and that method keeps improving.

Several claimed advantages evaporated when someone found a cleverer classical algorithm; a claim is only as strong as the effort spent trying to beat it classically.

The speedup depends on the problem and the algorithm, never on the machine alone. Shor's factoring is exponential, Grover's search only quadratic, and for most workloads (databases, training a network, running a PLC) no quantum speedup is known at all.

Cost counts too. A quadratic speedup on a machine whose operations are a million times slower and need error correction may never break even, which is why the serious candidates are problems with exponential gaps, chiefly simulating quantum systems themselves.

A useful habit when reading an announcement: ask which task, against which classical baseline, how many physical versus logical qubits, and whether the output has any value outside the demonstration.