Nearly 5,000 mid-circuit checks on up to 100 qubits. A step toward error correction, not error correction

On an IBM superconducting processor, a team from IBM, Rutgers and six other institutions let two steps compete on a chain of up to 100 qubits: one scrambles information between neighbours, the other measures single qubits mid-circuit and resets them. Changing how often each step runs produced a sharp transition between chaotic quantum dynamics and dynamics steered to one fixed state.

Research desk IBM Quantum and Rutgers, Nature Physics Read on 2 November 2026

Schematic of a chain of qubits shown twice: in the top row neighbouring qubits are joined by arcs for the scrambling step, in the bottom row four qubits are highlighted and boxed for the measure-and-reset step. Below, a dial from rarely to often, violet on the left for chaos and teal on the right for a steered fixed state, with a dashed marker for the transition. Three panels give up to 100 qubits, nearly 5,000 two-qubit gates and nearly 5,000 mid-circuit measure-and-reset steps. — Our schematic. The three numbers are the abstract's; the dial is a picture, and the transition point is not drawn to scale.

Error correction needs a quantum computer that can look and react

Every error-correction scheme works the same way underneath: in the middle of a computation, measure some qubits, decide what the result means, and act on it before carrying on. Each of those mid-circuit measurements disturbs the machine, and doing thousands of them in one run, quickly and cleanly, is a hardware skill in its own right. This experiment tests that skill at a large scale, using a physics question as the test.

What IBM and Rutgers report

The team, with Bibek Pokharel and Haining Pan as first authors and Jedediah Pixley and Maika Takita as senior authors, ran a chain of up to 100 qubits on a 156-qubit IBM Heron superconducting processor. Two steps took turns. One scrambles information between neighbouring qubits, a quantum version of a classically chaotic map. The other measures single qubits mid-circuit and resets them, steering the chain toward one fixed state. In the abstract's words, the circuits applied up to nearly 5,000 entangling gates and 5,000 non-unitary mid-circuit operations.

When the steering step runs rarely, scrambling wins and the dynamics stay chaotic and quantum. When it runs often, the chain is pulled to the fixed state. In between the team saw a sharp transition between the two, one that became more clearly defined in larger systems, and matched it with noisy simulations, matrix product states and statistical-mechanics models. The paper appeared in Nature Physics in October 2026, after an arXiv preprint in September 2025.

What this does not show

It is not error correction. No logical qubit is encoded and no error rate is pushed down. The paper uses the experiment as a signpost toward fault tolerance, which is fair; headlines that say it “corrects its own mistakes” go further than the work.
“Up to” is doing work. 100 qubits and nearly 5,000 operations of each kind are the largest runs, not every run.
What we read. The published paper sits behind a login from where we work. Our numbers come from the arXiv abstract and the university's release. Where the two frame the transition differently, we give neither number and point to the paper.
A beyond-simulation result about physics, not about usefulness. The release says 100 qubits is far past the roughly two dozen the team could simulate exactly. That is a statement about this model, not a claim of a useful quantum advantage.

Why it matters

Fast measure-and-react cycles are the plumbing of every error-corrected machine on the roadmaps, including IBM's own, tracked on the Ledger. Showing thousands of them in one run on 100 qubits, with the output still making physical sense, is the kind of quiet progress that rarely makes the front page and matters more than most results that do.

Sources

We read the sources above ourselves. Where a figure or number is ours, the article says so. If you find a mistake, tell us: a wrong sentence here gets logged like any other.