The Quantum Investor's Notebook logo The Quantum Investor's Notebook
← Back

Wednesday, June 3, 2026

How Should Investors Read a Gate-Fidelity Claim?

Gate fidelity matters. It matters a great deal. But it is not a complete score for a quantum computer.

Not a single number

There is no single gate-fidelity number for a whole machine. A company may report the fidelity of a single-qubit gateAn operation that acts on just one qubit. Generally easier to perform accurately than a two-qubit gate., a two-qubit gateA gate that acts on two qubits at once, such as the CNOT. Much harder to perform accurately than a single-qubit gate, and the real test of a machine., its best-performing qubitThe basic unit of a quantum computer. Like a 'bit' in a normal computer, but instead of being only 0 or 1 it can be 0, 1, or a blend of both at once. pair, an average taken across a processor, or an operation tested under carefully controlled laboratory conditions. These are not the same claim, and they should not be read as if they were.

The way the number was measured matters too.

How the number gets made

One common approach is called randomized benchmarkingA standard way to measure gate error by running random gate sequences and watching how fast the results decay.. Researchers apply carefully chosen random sequences of gates that should, in principle, return the qubit to a known final state. They then watch how quickly the results fall apart as the sequences grow longer. The faster the decay, the higher the inferred error rate.

A related method, interleaved randomized benchmarkingA version of randomized benchmarking that isolates the error of one specific gate by repeatedly inserting it into random sequences., estimates the error of one particular gate by inserting that gate into the sequence over and over and seeing what it costs.

A fidelity figure is not just a property of the hardware. It is the product of a measurement protocol, run under specific conditions.

So the details matter. A fidelity figure is not simply a property of the hardware. It is also the product of a specific measurementReading a quantum system, which forces it out of its blend of possibilities into a single definite 0 or 1 and ends its quantum behaviour. protocol, run under specific conditions.

A mind-map titled Reading a gate-fidelity claim. A central 99.99% gate fidelity is surrounded by six numbered questions: 1. Which gate (single-qubit versus two-qubit); 2. Measured how (a randomized sequence whose success decays with length); 3. Best or average (best pair versus full-chip average); 4. Alone or parallel (one gate alone versus gates run in parallel, where crosstalk appears); 5. Native or compiled (a target operation versus one compiled into native gates); 6. Fast and stable (how fast, how stable against drift over time). A panel below labelled More than one number lists qubit count, connectivity, readout, coherence, calibration, and error correction.
Figure 1: One headline number, many hidden questions. Before trusting a gate-fidelity claim, work around the dial: which gate, measured how, best case or average, alone or in parallel, native or compiled, and remember a machine is more than one number.

Questions I bring to a claim

When I read a fidelity claim, here is what I want to know.

Question Why it matters
Is this a single-qubit or two-qubit gate? Two-qubit gates are usually the harder and more consequential test.
Is this a physical gate or a logical gate? A logical operation may sit on top of error correction and many physical qubits underneath.
Is this an average, a median, or the best-performing result? One excellent qubit pair does not prove the whole processor performs as well.
Was the gate tested alone, or while other gates ran in parallel? Crosstalk may appear, or become more consequential, when several gates run in parallel.
Is this a native operation or a compiled circuit? A useful algorithm may need several physical gates to carry out one abstract instruction.
How fast and stable is the gate? A slow gate, or one that needs constant recalibration, can become its own bottleneck.
How was the fidelity measured? Different benchmarking protocols may summarize different aspects of performance.

A useful machine needs more than a good gate. It needs enough qubits, the right connectivityWhich qubits in a machine can directly interact with each other. More connectivity makes more algorithms possible., accurate readoutThe step of measuring a qubit to extract its final 0-or-1 answer., coherence timesHow long a qubit holds its delicate quantum state before noise scrambles it. Longer coherence means more time to compute. long enough to matter, calibrationTuning the control signals so gates stay accurate. It drifts over time and must be redone. that holds steady, fast operations, and a credible path toward error correctionTechniques that combine many shaky physical qubits into fewer reliable ones, so a long calculation stays correct.. You cannot reduce quantum hardware to one number.

So the useful question is never simply, “What is the fidelity?” It is this:

Which fidelity, measured how, on what system, and at what scale?