A quantum computer processes information through controlled quantum states. Its usefulness depends on the algorithm, the quality of the operations, and the resources needed to obtain a reliable answer. To interpret an announcement, ask what task was demonstrated, how errors were measured, and what classical comparison was used.
This explanation builds the distinctions needed to read quantum-computing news: physical and logical qubits, error correction, memory versus computation, and a benchmark versus an application. It uses dated research examples to explain evidence, rather than presenting a changing hardware leaderboard.
What a qubit contributes
A qubit is a two-level quantum system whose state can be described using amplitudes associated with two basis states, conventionally labeled 0 and 1. Measurement in that basis gives a classical outcome with probabilities determined by the state. Operations change amplitudes and their relative phases; interference is part of how a quantum algorithm shapes the distribution of answers.
Several qubits can have joint states with correlations that require a quantum description, including entanglement. Reading the qubits produces a finite set of classical outcomes. An algorithm must arrange those outcomes so that repeated measurements answer a useful question.
IBM's introduction to quantum information develops the mathematical model from states and measurements. It is a useful next step if you want to understand why an amplitude differs from an ordinary probability and why phase matters.
A practical workflow also includes classical computing: preparing inputs, compiling operations, controlling equipment, collecting measurements, and interpreting results. Evaluate the whole workflow when comparing time or cost. A fast quantum subroutine may sit inside a much longer process.
Separate physical and logical qubits
On a narrow screen, scroll the table sideways. Keyboard: focus the table and use the arrow keys.
| Term | What it describes | Evidence to look for |
|---|---|---|
| Physical qubit | A controllable quantum degree of freedom in hardware | Operation and measurement quality, connectivity, and relevant stability measures |
| Logical qubit | Quantum information encoded across physical resources using a code | Logical error behavior, code and decoding method, and resource overhead |
| Quantum memory | Preserving a state through a period or repeated correction cycles | Survival/error measures and a clearly defined physical-qubit comparison |
| Logical operation | A computation performed on encoded information | Gate implementation, error rate, and how operations compose into a circuit |
| Algorithm benchmark | A specified computational task run under defined conditions | Dataset/problem instance, success criterion, resources, and classical baseline |
Physical-qubit counts become meaningful alongside what those qubits are doing. Some support encoded data; others support checks, measurement, or other operations. Two systems with the same count can have different connectivity, error behavior, and algorithmic capability.
How error correction helps
Quantum error correction encodes information across multiple physical qubits and measures carefully chosen checks that reveal error information while preserving the encoded data. A classical decoder interprets those checks and determines how to handle the inferred errors. The details depend on the code, noise, hardware, and decoding method. IBM's lesson on correcting quantum errors explains these checks through explicit code examples.
Adding physical resources helps when the relevant errors are sufficiently controlled for the chosen scheme. Researchers therefore examine how logical error changes as code size increases. They also test whether the necessary decoding can keep up, whether rare correlated events matter, and whether logical operations retain useful quality.
The original page at this URL reported Emanuel Knill's 2005 work. His NIST-listed paper on realistically noisy devices investigated error thresholds and resource demands under specified assumptions, including independent errors. It provided theoretical evidence about fault tolerance. Its threshold estimates belong to that model and should be read together with the overhead required.
That distinction remains valuable: a theoretical construction establishes what a model permits; an experiment tests behavior in a particular physical system. Connecting the two requires examining whether the experiment's errors and operations fit the assumptions.
Read a demonstrated result within its scope
Google Quantum AI and collaborators' surface-code memory paper, published online in December 2024 and in a 2025 Nature issue, reported decreasing logical error as code distance increased in the tested Willow devices. The paper also studied real-time decoding and rare correlated errors. The current article page notes an April 2026 author correction.
The result provides evidence for error suppression in the reported memory experiments. To assess progress toward a full application, also look for logical gate operations, sustained multi-qubit computation, input/output handling, and the resources needed for the target algorithm. These are additional capabilities to demonstrate and measure.
Keeping the task name in the sentence helps: “preserved an encoded state through repeated cycles” says what happened. A broader phrase such as “solved practical problems” needs a separate application result and its comparison evidence.
A simple error-budget example
Illustrative model, not a device forecast: Suppose a sequence contains 1,000 operations, each with an independent failure probability of 0.001, and any failure makes the sequence unacceptable. The probability of no failures is (1 − 0.001)^1000, approximately 36.8%. Reducing the assumed failure probability to 0.00001 raises that value to about 99.0%.
The example explains why a small error probability can matter in a long calculation. Real quantum experiments involve structured and potentially correlated noise, correction cycles, measurements, and different definitions of success. A measured gate error should therefore not be inserted into this simple expression and presented as an actual algorithm's success rate.
For an error-corrected result, ask whether the reported number is per gate, per cycle, per unit time, or per entire experiment. Keep the unit and denominator with the value. A large improvement factor can refer to a narrowly defined comparison.
Evaluate a claim in five questions
- What was computed or preserved? Identify the task, input size, output, and success criterion. Distinguish a hardware demonstration from an application result.
- Which resources were counted? Look for physical and logical qubits, circuit depth, repetitions, decoding, classical preparation, and data transfer.
- How was reliability established? Inspect uncertainties, repeated trials, error definitions, and whether unsuccessful runs were excluded.
- What was the classical baseline? Ask which algorithm, hardware, accuracy target, and end-to-end timing were compared. Record whether the comparison is measured or estimated.
- What remains to scale? Identify the unperformed operations, assumptions, and resource growth between the demonstration and the proposed application.
Use the quantum-claim reading worksheet to record the paper, publication and correction dates, task, evidence, and remaining questions. It is a structured reading aid rather than a score that declares a technology useful or useless.
As new results arrive, compare them on the same task and success criteria wherever possible. The most informative progress is evidence that a previously limiting part of the complete computation now works reliably at a relevant scale.