What is “quantum utility” and how is it different from quantum advantage?

TL;DR (status as of 2026-10-26): “Quantum utility” is a weaker claim than quantum advantage, by design: IBM introduced the term on 2023-06-14 for results that are “accurate enough to be useful” at “a scale beyond brute-force classical computation” — while stating, in the same launch post, that “this isn’t a claim that today’s quantum computers exceed the abilities of classical computers” (IBM, 2023-06-14). The gap between the two words is measurable: the experiment that named the utility era was matched classically in 12 days (Tindall et al., arXiv v1 2023-06-26) and later reproduced at ≈10 s per data point on a single laptop core (Begušić, Gray & Chan, Sci. Adv. 2024), while quantum advantage — being “faster, more accurate, or more cost-effective” than classical-only methods “in a way that can be rigorously validated” (IBM, 2026-02-23) — remains contested as of today. Rosetta Quantum pins all three words of the quantum vocabulary to their sources and dates below.

What does “quantum utility” actually mean?

There is no formal definition — and that is the first thing to know. IBM introduced the term in June 2023 around one experiment: a 127-qubit Eagle processor simulating a kicked Ising model with zero-noise extrapolation, published as “Evidence for the utility of quantum computing before fault tolerance” (Nature 618, 500-505, 2023-06-14). The paper’s operative claim was “accurate expectation values for circuit volumes at a scale beyond brute-force classical computation”. IBM’s launch post framed utility as quantum computers serving as “a scientific tool to explore a new scale of problems”, delivering results “accurate enough to be useful” — and explicitly disclaimed the stronger reading: “this isn’t a claim that today’s quantum computers exceed the abilities of classical computers” (IBM blog, 2023-06-14).

The companion term stuck as a size label: IBM’s own course material says that after the 2023 paper, “from then on, we called such experiments and circuits ‘utility-scale’” — circuits “at a scale beyond statevector simulation” (IBM Quantum Learning, 2024-07-05). Note what that measures: circuit size relative to one classical technique (brute-force statevector), not victory over the best classical technique.

How is utility different from quantum advantage?

Utility is a claim about scale and usefulness; advantage is a claim about beating every classical alternative. IBM’s current definition of advantage is the strictest it has published: the point at which quantum computation is shown to be “faster, more accurate, or more cost-effective” than classical-only methods for some problem, “in a way that can be rigorously validated” (IBM, 2026-02-23). Utility, in IBM’s own 2023 language, requires neither beating the best classical methods nor rigorous validation against them — only useful accuracy at a scale where brute force gives out. The ladder of words, pinned to sources:

The practical consequence: a vendor sentence like “we reached utility scale” is compatible with a laptop outperforming the same computation. That is not an accusation — it is the literal content of the definitions above, including IBM’s own disclaimer.

What happened to the 2023 utility experiment?

It was matched classically in 12 days. The Nature paper went online on 2023-06-14; on 2023-06-26, Tindall, Fishman, Stoudenmire and Sels posted a tensor-network simulation of the same kicked-Ising system that they reported as “significantly more accurate and precise than the results obtained from the quantum processor” (arXiv v1 2023-06-26; PRX Quantum 5, 010308, 2024-01; dates per Quantum Timeline, accessed 2026-10-26). Two days later, on 2023-06-28, Begušić and Chan posted a second, independent classical match; the journal version reports the sparse-Pauli-dynamics data “simulated within on average 10 s per point on a single core of a laptop computer” — “three orders of magnitude faster than the reported quantum wall-clock run time” — and reaching absolute accuracy below 0.01 without extrapolation, while also identifying a point (angle 6π/32) where “the experimental extrapolation … has a significant error outside of its error bars” (Science Advances 10, eadk4321, 2024-01-17).

Two honest footnotes to that scoreboard. First, IBM anticipated classical verification: the launch post described the experiment as a challenge to classical methods and expected them to respond — which they did. Second, the 2019 row is even stranger than it looks: IBM’s 2.5-day estimate (arXiv:1910.09534, 2019-10-21) landed two days before Google’s Nature paper was formally published (Nature 574, 2019-10-23), because the manuscript had leaked in September. By 2021-11, Pan, Chen and Zhang had generated one million uncorrelated samples from the supremacy circuits — fidelity ≈0.0037, matching the experiment’s own — in 15 hours on 512 GPUs (PRL 129, 090502); the full seven-year erosion is traced in our 2019-supremacy retrospective (2026-09-24).

Is the same pattern repeating with “advantage” in 2026?

The clock ran again, and the reply took 14 days. On 2026-07-30, IBM and the University of Chicago (with Qedma and Algorithmiq) announced three preprints that, per IBM’s newsroom, “achieve[…] the fundamental criteria for quantum advantage”, the flagship being a 70-qubit encoded circuit with built-in validation (IBM Newsroom, 2026-07-30). On 2026-08-13, a preprint by Manabe, Gu and Pan reported computing the exact outputs of that flagship circuit in 37.3 minutes on 256 H100 GPUs. On 2026-09-02, IBM’s revised preprint acknowledged that subsequent work “computed the exact ideal probabilities of all 2,051 reported output samples” (≈159 H100-GPU-hours) and changed the circuit layout to block that shortcut; the claim, amended, remains open, and the Qedma and Algorithmiq results were not covered by the classical challenge (all per Quantum Timeline, accessed 2026-10-26). Both sides are linked; we are not scoring the 2026 episode here — the claims and replies will still be moving when we re-examine them.

Why does the vocabulary keep shifting?

Because the classical frontier moves, and each word pins the bar at a different height. There are two readings of the supremacy → advantage → utility drift, and we print both. The charitable, methodological reading: the field is learning to claim less and specify more — “utility” was accompanied by an explicit disclaimer, and IBM’s 2026 Advantage Tracker, an open effort where classical researchers contest quantum submissions, states that advantage “will not come from a single announcement” (IBM, 2026-02-23). That is movement toward exactly the kind of adversarial, dated verification Rosetta exists to do. The skeptical, marketing reading: when a strong word gets refuted, a softer word keeps the headline — “utility-scale” in particular now functions as a size badge that sounds like usefulness while requiring, by its own definition, no comparison against the best classical methods at all. The dated record above is compatible with both readings; what it does not support is treating “utility” as a synonym for “advantage”. Our own unit — the advantaged solve (2026-08-26) — exists to avoid the ambiguity: name your baseline, measure, publish the raw data. A checklist for reading any such claim is in our credibility guide (2026-08-25).

What we don’t know

We don’t know whether any utility-scale computation running today lacks a classical match: IBM’s Advantage Tracker lists 30+ submissions under open review (2026-02-23), and verdicts there are provisional by construction. We don’t know whether IBM’s revised 70-qubit layout (2026-09-02) will withstand the next classical reply — that episode is open as of 2026-10-26. We don’t know what “utility” formally requires, because no formal definition has been published; whether a given result “counts” is editorial judgment, including ours. And we can’t isolate how much of the classical progress above was caused by the quantum claims themselves: tensor-network and GPU methods improved against these exact targets, so the 12-day clock measures a contest, not a constant.

Rosetta Quantum publishes verdicts with raw, reproducible data and dated sources. This is educational content, not a product claim.