What is a logical qubit?

A quantum computer stores information in qubits, the quantum version of the bit. Qubits are fragile. Heat, electrical noise and stray interactions with their surroundings can flip or blur the information they hold, and one disturbed qubit can spoil a whole calculation. A logical qubit is a unit of information that is protected from these errors. It is built from many physical qubits that work together, and it is the building block that fault-tolerant quantum computers are expected to need. This explainer separates what has been shown in the laboratory, what has only been reported by companies and press, and what remains a projection.

Why one qubit is not enough

A physical qubit on its own has no way to tell that it has been disturbed. Quantum error correction gets around this by spreading one piece of information across several qubits, and then checking the group for signs of error without reading the stored answer directly. Reading the answer would collapse the quantum state, so the checks must be designed to reveal only whether something has gone wrong.

An idea from 1995

In 1995 the physicist Peter Shor published a scheme for reducing decoherence in quantum computer memory in Physical Review A. The abstract says the method assumes that decoherence acts independently on each stored bit, an assumption that real hardware may not fully meet. A feature in Physics magazine describes how Shor encoded one qubit in nine entangled physical qubits, arranged in three groups of three. If one qubit is disturbed, its group stops matching the other two, but the two intact groups still hold an accurate record of the original state. Shor's paper was a theoretical proposal. It was not a working device.

What has been demonstrated

These results come from laboratory experiments, and each one has specific conditions.

Google Quantum AI, 2024. Google's team ran two quantum memories on its superconducting Willow processor, using a type of code called the surface code. The surface code arranges qubits on a grid and checks neighbouring groups repeatedly. The experiment used a distance-5 memory and a distance-7 memory. The larger one used 101 physical qubits, and it was published in Nature in December 2024, after a preprint appeared in August of that year.

The key finding was that making the code larger made the stored information more reliable, not less. The paper reports that each increase of two in code distance reduced the logical error rate by a factor of 2.14, with an uncertainty of 0.02. The distance-7 memory lost information at about 0.143 per cent per correction cycle (plus or minus 0.003), and it kept its information about 2.4 times longer than its best single physical qubit. A real-time decoder, the classical software that reads error signals and decides which corrections to apply, averaged 63 microseconds of latency at distance 5, with each cycle lasting 1.1 microseconds.

These are memory experiments. The team stored information and corrected errors, but they did not run a useful calculation on the logical qubits.

What is only reported

The following claims come from companies and have not been checked in the way a peer-reviewed paper is checked, so they should be read with care.

• Quantinuum, October 2026. In a press release, Quantinuum reports a logical fidelity of 99.9995 per cent on its Helios computer, using a colour code. It says each tested increase in code distance cut the remaining logical errors by a factor of 12. The release says the colour code used up to 37 physical qubits per logical qubit, but it does not give the code distances tested, and it does not say whether the work has been peer reviewed.
• QuEra and Harvard, 2025. QuEra's December 2025 press release describes a Harvard-led team executing algorithms with up to 96 logical qubits. The company says the work was published in Nature. The release does not state the number of physical qubits used for that experiment.
• Google's own account of Willow. Google's announcement describes Willow as having 105 qubits and calls it the first system to operate below threshold. The peer-reviewed Nature paper supports the below-threshold result, but the wider claims about Willow are Google's own.

What is speculative

Some statements look ahead rather than report results.

• Quantinuum's release quotes Robin Blume-Kohout of Sandia National Laboratories, who was not involved in the study, saying a scaled-up system could work at a feasible size of "less than a million physical qubits". That is a projection, not an experimental result.
• Quantinuum says its planned Apollo machine, a fault-tolerant computer, is expected to be available by 2029. This is a company target, not a demonstrated result. The company also warns that it may not achieve its roadmap on that timeline, or at all.

Limits and open problems

Overhead. A logical qubit needs many physical qubits, and the size of that overhead decides whether large machines are practical. The Google memory used 101 physical qubits for one logical memory, and the sources reviewed here do not give a verified overhead figure for a useful machine.

Correlated errors. In the Google work, repetition codes tested up to distance 29 were limited by rare correlated error events, roughly once an hour. The paper reports this for repetition codes. The sources reviewed here do not show whether the surface code avoids the same limit.

Decoding at scale. The decoder must keep pace with the chip. The Google paper shows real-time decoding at distance 5; keeping pace at much larger sizes has not been shown in the sources reviewed here.

One platform, small sizes. The most detailed result comes from a single superconducting platform and a small range of code sizes. Results on one type of qubit do not automatically carry over to others.

Hardware assumptions. Shor's scheme assumed that errors act independently on each bit. Real devices may have errors that are linked to one another, which makes correction harder.

What to watch next

• Whether independent groups reproduce below-threshold results on other hardware, including trapped-ion and neutral-atom platforms.
• Whether companies publish code sizes, total physical qubit counts and peer-reviewed papers behind their press-release figures.
• Whether logical error rates keep falling as codes grow, and whether the decoders can keep pace at larger sizes.
• Whether company roadmap targets for fault-tolerant machines, such as Quantinuum's 2029 target for Apollo, are met.