Quantum Utility vs Quantum Advantage
Two precise, easy-to-conflate claims about where the field actually stands.
- Quantum Utility (achieved, 2023) — a demonstration that a quantum computer can run quantum circuits beyond the ability of a classical computer simulating a quantum computer. IBM’s claim rests on the same Nature 618 paper (Kim et al., 2023) already cited in this vault’s own 1D Ising Chain and the Mirror Trick note — the 127-qubit utility demonstration that motivated the Trotter-circuit benchmark used throughout the Error Mitigation section.
- Quantum Advantage (not yet achieved) — a demonstration that a quantum computer can solve a problem more accurately, cheaper, or more efficiently than classical computing alone. This is a strictly higher bar: utility only requires being hard to simulate classically, not being useful or better than the best classical alternative for the same task.
Key insight: these are commonly conflated in pop-science coverage. “We ran a circuit no classical computer could simulate” (utility) is not the same claim as “we solved a real problem better than the best classical method” (advantage) — see the honest verdict on a real 56-qubit benchmark that walks through exactly this distinction.
Why the gap is hard to close
Individual gate error rates are low (~0.1%), but failure probability compounds with every additional operation — a circuit with thousands of two-qubit gates accumulates meaningful aggregate error even at excellent per-gate fidelity (see Backend Properties). IBM has deployed 60 devices under 100 qubits and 29 devices over 100 qubits capable of over 5,000 two-qubit gates since 2016 — utility-scale hardware is real and growing, but closing the remaining gap to advantage is an active research problem, not a solved one.
The three pillars of advantage
Claiming quantum advantage for real requires all three at once:
- Complexity — the problem must be one where classical heuristics genuinely become computationally unviable, either through a proven complexity separation or practical time/memory exhaustion.
- Trust — the output must be trustworthy even when it can’t be classically double-checked, built up step-by-step through calibration and mitigation — the same way physicists came to trust telescopes and particle colliders before either could be “checked” against a known answer.
- Utility — it has to drive real impact (materials science, drug discovery, fundamental physics), not just demonstrate hardware speed on an abstract puzzle.
The three classical competitors
Whenever a quantum result claims to beat classical methods, it’s being compared against the best available classical simulation — and each has a different limiting resource:
| Method | Limited by |
|---|---|
| State vector | Qubit count — exact, but memory is exponential (see Simulators — Statevector vs Shot-Based) |
| Tensor networks | Entanglement / bond dimension (see Hamiltonian Simulation — Why It’s Hard) |
| Pauli path methods | Operator spreading — how fast Heisenberg-picture operators grow under evolution |
A quantum result only counts as advantage over all three — beating one classical method while another would have solved it faster isn’t advantage, it’s picking a weak opponent.
The Quantum Advantage Tracker
A community-run platform (quantum-advantage-tracker.github.io) that catalogs real experimental attempts at the gap above, organized into three pathways that map directly onto the three problem families poised for advantage:
| Pathway | Example on the tracker |
|---|---|
| Observable Estimation | Ising-model simulation — see The Operator Loschmidt Echo (OLE) Benchmark |
| Variational Problems | Sample-Based Quantum Diagonalization (SQD) for molecular ground states |
| Classically Verifiable Problems | QAOA for optimization problems |
Two concrete 2025-baseline entries worth knowing: the ground-state energy of Fe₄S₄ (an iron-sulfur cluster, biologically important and classically hard due to strong electron correlation) computed via TrimCI (Trimmed Configuration Interaction — builds ground states from random Slater determinants without a pre-selected reference, then trims unimportant components; recovers >99% of the 8×8 Hubbard model’s ground-state energy using just of the Hilbert space, beating AFQMC); and the expectation value of the Loschmidt Echo operator via Single-Path Monte Carlo at 49 qubits / 648 gates — the same quantity, and largely the same mitigation story, as The Operator Loschmidt Echo (OLE) Benchmark’s 56-qubit run, just a different baseline entry on the tracker rather than the same experiment.
Key insight: the tracker doesn’t declare winners — it’s a running scoreboard of attempts, most of which (like the OLE benchmark in this vault) land on “more trustworthy than one classical method” rather than “advantage over the field.”
Related
- What Quantum Computers Are Good For
- 1D Ising Chain and the Mirror Trick — the same 2023 Nature paper, from the mitigation-benchmark side
- Backend Properties
- Error Correction (EC) — the long-term path to closing this gap for good
- Peaked Circuits and Verifiable Quantum Advantage
- The Operator Loschmidt Echo (OLE) Benchmark — a real worked example of exactly this gap
- Quantum Advantage — Definition and Criteria — the formal two-criterion framework this whole distinction is built on
- Sample-Based Quantum Diagonalization (SQD) — the Variational Problems pathway’s flagship example
Self-Check
- Could you explain the difference between quantum utility and quantum advantage in your own words?
- Why is “no classical computer could simulate this circuit” a weaker claim than “this beat the best classical method”?
- Why does a low individual gate error rate (~0.1%) not guarantee a low overall circuit error rate?
- Why does claiming advantage require beating all three classical competitors, not just one?
- How do the Quantum Advantage Tracker’s three pathways map onto the three problem families in Quantum Advantage — Definition and Criteria?