Quantum error correction has crossed its most important threshold

The problem in one line

A qubit is fragile. Heat, stray fields and the act of reading it all nudge its state, and a calculation that takes millions of steps cannot survive an error every thousand steps.

The idea

Ordinary computers fix errors by repetition: store a bit three times and take the majority. Quantum information cannot be copied, but it can be spread across many physical qubits in a way that lets you detect a disturbance without looking at the value itself. The spread-out object is called a logical qubit.

The theory promises something remarkable. If each physical qubit is good enough, then making the logical qubit bigger reduces its error rate exponentially. If the physical qubits are worse than that threshold, adding more only adds more ways to fail.

What was shown

For years the second case was reality. Bigger codes performed about the same or worse. In a paper published in December 2024, a superconducting processor of just over one hundred qubits ran a surface code at three sizes. Each step up in size roughly halved the logical error rate. The largest logical qubit stayed coherent more than twice as long as the best physical qubit inside it.

That is the first clear demonstration of operating below threshold on a real device.

What it does not mean

It does not mean a useful quantum computer exists. The logical error rate shown, about one in a thousand per cycle, needs to fall by roughly a thousandfold more, and one logical qubit is not a computer. Useful problems need hundreds to thousands of them, so hundreds of thousands to millions of physical qubits.

There are also unsolved engineering issues in the paper itself, including rare bursts of correlated errors, possibly from cosmic rays, that overwhelm the code.

Why it is still the milestone

Before this result the open question was whether the exponential improvement predicted on paper would appear in hardware at all. Now it has. The road ahead is long, but it is downhill in the sense that matters: more qubits now help.

Other approaches to watch

Trapped-ion and neutral-atom machines have shown logical qubits with different codes and fewer physical qubits. Which platform wins is open. What is no longer open is whether error correction works.