Quantum advantage, quantum supremacy and quantum utility: what the words mean

Three phrases appear over and over in quantum computing coverage: quantum supremacy, quantum advantage, and quantum utility. They are sometimes treated as synonyms and sometimes as a ladder of progress. They are neither. They come from different papers and talks, they make different comparisons, and they can all be used carefully or carelessly.

This guide explains where the words came from, what a responsible claim needs to specify, and how to read them next to the more basic ideas of qubits and error correction. It pairs with our checklist on how to read a quantum computing announcement.

Quantum supremacy

John Preskill introduced “quantum supremacy” in a 2012 rapporteur talk, later posted as “Quantum computing and the entanglement frontier.” He used it for the hoped-for moment when a well-controlled quantum system could perform a task that goes beyond what ordinary digital computers can achieve. In that talk, a convincing supremacy demonstration did not have to solve a useful industrial problem. It could be a contrived task chosen because it is hard to simulate classically and possible to run on near-term hardware.

That meaning — a quantum device doing something that stretches or exceeds plausible classical simulation, not necessarily something anyone would pay to compute — is the meaning that stuck in the research literature.

The best-known claim arrived in October 2019, when Google and collaborators published “Quantum supremacy using a programmable superconducting processor” in Nature. They reported a random-circuit sampling experiment on the 53-qubit Sycamore processor and estimated that a task taking their quantum device about 200 seconds would take a state-of-the-art supercomputer about 10,000 years. Those numbers are the authors’ estimates, not a universal constant of nature: they depend on how one models the classical simulation.

IBM researchers responded publicly, including in a blog post titled “On ‘quantum supremacy’,” arguing that a more careful classical simulation — in particular one that made better use of secondary storage — could finish the same sampling task in a matter of days rather than millennia. The disagreement was not about whether Sycamore ran a hard circuit. It was about what “beyond the reach of classical computers” meant once the classical algorithm and the machine model were specified more generously.

That exchange is the template for reading every later supremacy-style claim: name the task, name the quantum device, name the classical baseline, and expect the baseline to move as classical algorithms improve.

Quantum advantage

“Quantum advantage” is used in two overlapping ways, which is why it causes trouble.

In one usage it is a milder synonym for supremacy: a quantum computer outperforms a classical computer on a well-defined task. Some researchers prefer “advantage” because “supremacy” has political overtones in English that have nothing to do with physics. When a paper says it is avoiding the word supremacy for that reason, it is usually still talking about a computational comparison, not about a product being generally better than classical computers.

In the other usage, “advantage” is reserved for a speedup (or an improvement in cost or energy) on a problem people already care about — chemistry, materials, optimization, cryptography — rather than on a sampling task invented for the comparison. Under that stricter usage, a random-circuit sampling experiment can be supremacy without being advantage.

You cannot tell which meaning an announcement intends unless it says what was compared with what. A headline that says only “quantum advantage achieved” is unfinished.

Quantum utility

“Utility” entered wide circulation after a 2023 Nature paper by Youngseok Kim, Andrew Eddins, Abhinav Kandala and colleagues at IBM and academic partners, “Evidence for the utility of quantum computing before fault tolerance.” The experiment ran circuits on a noisy 127-qubit superconducting processor, including circuits with up to 60 layers of two-qubit gates (2,880 CNOT gates in the largest cases they highlight). The authors argued they could measure accurate expectation values at a scale beyond brute-force classical computation, and that in a strongly entangled regime their results remained reliable where some leading approximate classical methods, including certain tensor-network approaches, broke down.

Two cautions follow directly from the paper itself. First, the authors present this as evidence for utility in a pre-fault-tolerant era — a foundation for exploring near-term methods — not as a claim that a commercial application now beats every classical solver. Second, they compare against particular classical techniques (brute-force simulation where it still applies, and named approximate methods where it does not). Later classical algorithms can narrow or close that gap, just as classical simulation work answered parts of the 2019 supremacy comparison.

In looser marketing language, “utility” is sometimes stretched to mean “customers are getting value from our quantum system.” That may be a reasonable business statement. It is not the same statement as the 2023 paper.

How these words sit next to NISQ and fault tolerance

Preskill’s later NISQ essay (2018) described an era of devices with tens to hundreds of noisy qubits, interesting enough to run experiments that are hard to simulate, and not yet protected by full fault-tolerant error correction. Supremacy-style sampling experiments live naturally in that era: they ask whether a noisy device can outrun classical simulation on a tailored task.

Useful, large-scale algorithms of the kind discussed in textbooks generally assume a fault-tolerant machine, in which logical qubits and logical gates are reliable enough that deep circuits can run. That is a different milestone from supremacy, advantage, or utility. A group can report progress on error correction without claiming advantage, and can report a supremacy-style experiment without having a logical qubit.

Mixing these layers — “we demonstrated supremacy, therefore useful algorithms are here” — is the most common leap in popular write-ups. The experimental papers themselves rarely make that leap.

A reading test for any of the three claims

Whatever word an announcement chooses, the same questions apply:

• Task. What input is given, what output is produced, and why is that output hard classically?
• Quantum device. How many qubits, which modality, what circuit depth, and what error mitigation or error correction was used?
• Classical baseline. Which algorithm, which machine, how much time and memory, and who chose the implementation — the quantum team or an independent classical group?
• Verification. How does anyone know the quantum output is correct once the problem is too large to simulate exactly?
• Stability of the claim. Has a later classical method already reduced the gap? Supremacy and advantage claims can shrink without any error in the original experiment.
• Usefulness. Is the task something a user outside the experiment wanted solved, or is it a benchmark constructed for the comparison?

If those answers are in the paper or the technical blog post, the word in the headline matters less. If they are not, the word in the headline cannot supply them.

How DesignQuantum^™ uses the terms

In our news stories we prefer to repeat the comparison an announcing group actually made, in their own attributed words, and then link to the paper or post. We use “supremacy,” “advantage,” or “utility” in our own voice only when it is clear which comparison is meant. When a release uses one of these words without a comparison, we treat that as a claim by the speaker, not as a result we have independently rated.

The vocabulary will keep shifting. What should not shift is the demand that a computational claim name its task and its baseline.