What does a GHZ record actually measure?

What is a GHZ state, and why is it the benchmark for “entangling the whole chip”?

A GHZ state is the simplest state in which every qubit on the device participates in one single entangled whole: an equal superposition of “all qubits 0” and “all qubits 1”, written (|0…0⟩ + |1…1⟩)/√2. It is the standard full-device benchmark because it is maximally unforgiving: preparing one across a chip forces every qubit, every coupler and the calibration of all of them to work correctly inside the same circuit — one bad gate or one decohered qubit anywhere degrades the whole state. IQM's CTO put the appeal plainly in the October 29 release: “Entangling the entire chip means every qubit and every connection between them has to work correctly in the same circuit” (IQM press release, 2026-10-29). For what entanglement itself is — and what it does not allow — see Rosetta's dictionary entry.

What does a fidelity above 0.5 certify?

A measured GHZ fidelity above 0.5 is a sufficient witness of genuine multipartite entanglement (GME): the measured state cannot be reproduced by any mixture in which some subset of qubits is unentangled from the rest. The criterion goes back to Leibfried et al. (Nature, 2005) and is the standard cited both by the academic record-holders (arXiv:2101.08946) and by IQM's own benchmark documentation (docs.iqm.tech, consulted 2026-11-05).

What the witness does not certify is just as important. It is a statement about one state, prepared by one shallow, heavily optimized circuit, at one calibration moment. It says nothing about running deep algorithms on those 54 qubits, nothing about how long the entanglement survives (GHZ states decohere fast by design), and nothing about repeatability next week. IQM's own release notes that users previously ran comparable demonstrations on roughly 30 qubits of the same chip, and that large-chip studies commonly use 10 to 20 qubits (IQM, 2026-10-29) — full-chip GHZ is a calibration milestone, not a computation milestone.

What exactly did IQM report on October 29, 2026?

Two numbers, each in a different accounting regime, with no uncertainty attached to either:

The release attributes the result to roughly a year of calibration work — an updated calibration methodology, an improved gate set, dynamical decoupling, automated qubit and gate-order selection — and claims error rates for quantum operations were reduced by up to 50% (IQM, 2026-10-29). It names no measurement method for the fidelity itself; IQM's public benchmark suite implements two estimators, multiple quantum coherences (per Mooney et al., arXiv:2101.08946) and randomized measurements (per Elben et al., 2019), per its documentation (docs.iqm.tech, consulted 2026-11-05). No preprint is referenced in the release, and we found none on arXiv as of November 5, 2026.

One more claim in the release deserves flagging: that the results “rank among the strongest reported for any commercially available quantum computer.” No benchmark data or competing system is cited for it. Quantinuum's public system documentation lists a 56-qubit GHZ fidelity knowledge article for its commercial H2 machine (docs.quantinuum.com, consulted 2026-11-05; the fidelity value did not render in the page text we could retrieve), so the “strongest commercial” framing will remain unadjudicated until someone puts the numbers side by side — in the same accounting regime.

How does 0.516 compare with other entanglement records?

It is neither the largest GHZ state nor the highest GHZ fidelity on record — the claim that survives is narrower and still real: all qubits of one commercially cloud-accessible chip, entangled at once. The dated record ladder:

Two patterns in the table are worth more than any single row. Trapped ions deliver much higher GHZ fidelity at smaller qubit counts — Quantinuum's 32-qubit 0.82 was measured raw, with no mitigation at all (arXiv:2305.03828). Superconducting chips push the qubit count up while the fidelity hugs the 0.5 threshold — IBM's 120-qubit state, the largest GHZ reported to date, sits at 0.56 only after discarding 72% of shots via error detection (arXiv:2510.09520). And one row is deliberately not a GHZ state: USTC's 51-qubit result was a 1D cluster state, a different (and more error-tolerant) entanglement structure — headlines routinely compress all of these into “biggest entangled state,” and the compression is where the meaning leaks out.

Why are these numbers not directly comparable?

Because each sits in a different accounting regime, and the regime can be worth more than the margin. IQM's 54-qubit figure uses readout error mitigation; its raw figure is only published at 40 qubits (0.515), so the raw-to-mitigated gain at 54 qubits is not computable from the release (IQM, 2026-10-29). Mooney's 0.546 is also a mitigated number — QREM — with the raw value lower (arXiv:2101.08946). IBM's 0.56 keeps only the 28% of shots that pass error detection (arXiv:2510.09520). Quantinuum's 0.82 is raw (arXiv:2305.03828). Lining these up in one column without the accounting column is how a record table turns into marketing. Rosetta Quantum applies the same reading to every scoreboard: the number and its accounting regime travel together, or the number travels alone and means less — the same discipline we applied to logical-qubit demos and to Quantinuum's «10×» record.

The measurement method matters less than the mitigation column — the IBM 120-qubit team certified fidelity with multiple methods and reports they agree within practical considerations (arXiv:2510.09520) — but it still belongs in the disclosure: multiple quantum coherences, parity oscillations, randomized measurements and direct fidelity estimation have different statistical behavior at large qubit counts, and a release that names no method at all, like IQM's, leaves the reader unable to reproduce the bar it claims to have cleared.

How should you read the next GHZ headline?

Run it against seven fields; every one of them is checkable from the source, or its absence is the finding:

1. Qubit count and state type. GHZ, cluster, graph — they are not interchangeable; “biggest entangled state” without the state type is already a red flag.
2. Fidelity with its error bar. A margin of 0.016 over the threshold means one thing with ±0.005 and another with ±0.02. No error bar, no adjudication.
3. Raw or mitigated — and which mitigation. REM, QREM, post-selection, error detection: each moves the number differently.
4. Post-selection fraction. Keeping 28% of shots is a legitimate, declared technique (arXiv:2510.09520) — undeclared, it is a silent subsidy.
5. Measurement method. MQC, parity oscillations, randomized measurements, DFE — named, or not reproducible.
6. Preprint or peer review. A press release is a claim; a paper is a checkable claim.
7. Independent replication. For every row in today's table: zero that we know of, as of 2026-11-05.

This is the same checklist logic Rosetta maintains for quantum advantage claims — applied to entanglement records, where the threshold is sharper (0.5 is a bright line) but the accounting games are identical.

What we don't know

• The uncertainty on IQM's 0.516. No error bars were published as of 2026-11-05. Whether the 0.016 margin above the GME threshold survives one standard deviation is not assessable from public information; comparable experiments report uncertainties between ±0.01 and ±0.03.
• The raw 54-qubit fidelity. Only the 40-qubit raw figure (0.515) was released; the unmitigated full-chip number is unknown.
• The measurement method behind the release figure. Not named; IQM's benchmark suite offers MQC and randomized measurements, but which produced 0.516 is unstated.
• Whether a preprint will follow. None is referenced in the release and we found none on arXiv as of 2026-11-05; if one appears with error bars, this post will be updated — same slug, new date.
• How the «strongest commercial» claim adjudicates. No comparison data was cited; Quantinuum's documentation lists a 56-qubit GHZ benchmark for H2 whose value we could not retrieve from the page text (docs.quantinuum.com, consulted 2026-11-05).
• Independent replication. For IQM's result and for every record in our table: none that we know of at this date.

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