Quantum advantage vs. quantum supremacy: what's the difference?

Status as of: July 2026. These are the terms of art and the public record of who claimed what; the dates and rebuttals are sourced below.

The two words are not synonyms, and the gap between them is where most of the confusion in quantum computing lives. Quantum supremacy means a quantum computer performs some task that a classical computer cannot do in any reasonable time — the task does not have to be useful. Quantum advantage, used strictly, means the quantum computer beats the best available classical method on a task someone actually wants solved. Supremacy is a physics milestone; advantage is an economic one. Almost every headline you've read conflates them, and that conflation is not harmless: it is exactly how a lab demonstration becomes, in a press cycle, "quantum is here."

What each word actually means

Supremacy was coined by John Preskill in 2012 to name a threshold: the point where a quantum device does something no classical computer can feasibly reproduce. The definition is deliberately silent on usefulness — random-circuit sampling, the task used for the first claims, produces numbers nobody needs. Preskill himself later wrote that he'd picked the word partly to be provocative, and distanced himself from the hype it attracted.

Advantage raises the bar to something a customer would pay for: the same problem, solved better or faster than the strongest classical solver available today. This is a moving target, because classical algorithms keep improving — and that is the whole difficulty. A quantum result is only an advantage until someone writes a better classical algorithm; then it isn't.

"Utility" is a third, softer word IBM introduced in 2023 for running a useful-scale computation whose accuracy is worth checking, without claiming to beat classical. It's an honest hedge: because it doesn't assert supremacy, it can't be refuted by a faster classical method. Reading these three words as interchangeable is how a careful, hedged claim and an unhedged one end up in the same sentence.

The scoreboard: every headline claim, and what survived

The pattern is consistent. In 2019 Google reported that its Sycamore chip did in 200 seconds what would take a supercomputer "10,000 years"; within days IBM argued a better classical method would need roughly 2.5 days, not millennia. In 2023 IBM ran a 127-qubit "utility" experiment — carefully not a supremacy claim — and within weeks several groups reproduced the result classically with tensor-network methods. In October 2025 Google's "Quantum Echoes" algorithm on the Willow chip reported a ~13,000× speedup that is, importantly, verifiable — a genuine step — but the task is a physics measurement (out-of-time-order correlators), not an industrial problem a company would buy.

Sources linked at the foot of this page. None of these is fraud; each is a real result read honestly. The recurring lesson is that the classical baseline moves, so "beats classical" has to be re-earned against the best method known today, not the one cited in the press release.

Where "advantaged solve" fits

Rosetta uses a stricter unit than any of these words: an advantaged solve is a single instance of a useful problem where a quantum recipe beat the user's own tuned classical baseline, same instance, same budget, measured and dated. It sits at the bottom rung of the ladder above — the hardest to claim and the only one a buyer can act on. By that bar, the honest state of the field on useful, industry-relevant problems is not yet: our own public verdict RQ-0012 is a measured "not yet" across 20 sealed optimization runs, with the classical solver at the proven optimum every time. Supremacy on a contrived task and advantage on your task are separated by years, not months.

What we know / What we don't know

What we know (sourced, dated). The definitions and their origins. That each headline supremacy/utility claim to date was either narrowed by a better classical algorithm or was explicitly not a usefulness claim. That Google's 2025 Willow result is verifiable — a real advance in checkability — on a physics task.

What we don't know. Whether the Willow result survives future classical algorithms (the 2019 and 2023 claims did not fully). When, or whether, any of these techniques crosses into a durable advantage on an industry problem — that is the crossover we measure and cannot yet call. And how much of any given speedup is the hardware versus a classical baseline that simply hadn't been optimized yet; distinguishing the two is the entire job of an honest referee.

Rosetta Q publishes verdicts with raw, reproducible data. This is educational content, not a product claim, and it takes no position for or against any vendor — every claim above is a sourced public result read at face value.