A 98-qubit computer fixes its own mistakes, and beats its raw hardware

What happened

A quantum computer is fast, but it is also fragile. Its qubits, the tiny parts that hold information, make mistakes all the time. Heat, stray light and even the act of reading them can flip an answer. How do you compute with parts that keep making mistakes? You spread one piece of information across many qubits, so that when one slips, the others still hold the answer. This is called quantum error correction.

For years this was mostly theory. In September 2026, a team at Quantinuum, a quantum computing company based in the United States and the United Kingdom, showed the full method working on a real machine. The machine is called Helios. It has 98 qubits made from trapped ions, which are charged atoms held in place by electric fields.

The test

The team used a code called C4-Helix. In plain words, it stores 2 protected qubits inside 20 real ones. They ran the machine in cycles. In each cycle the computer checked itself for mistakes and repaired them. Then they did three things.

First, they kept information stored, cycle after cycle, and counted the mistakes. Second, they ran a full set of basic logic operations on the protected qubits while the repairs kept going. Third, they connected two different kinds of protected qubits together, which is a step future machines need.

The result

The protected qubits made a mistake about 4.6 times in every 100,000 cycles. That is roughly one slip in every 22,000 cycles. A smaller version of the same code, with 10 real qubits for 2 protected ones, made 21 mistakes in 100,000 cycles. Doubling the size cut the errors by about 4.5 times. This is the sign researchers look for. It means the bigger you build the code, the fewer mistakes you get.

The logic operations told the same story. A protected two-qubit operation failed about 2.8 times in 10,000 tries. The same operation on bare, unprotected qubits failed about 12 times in 10,000 tries. Protected was four times better. The linked state across two code types came out at least 99.9 percent correct.

Why does this matter more than earlier results? This is because the team did not throw away bad runs. In many quantum papers, the good numbers come after “postselection”, which means deleting the runs that went wrong. Here, every run counted, and the protected version still won.

Think of a choir. Twenty singers hold two notes. If one singer wavers, the other nineteen keep the note true, and the listener never hears the slip. That is what the code does. Therefore, a wobbly machine can still give a steady answer.

What it means

A quantum computer that fixes its own mistakes is the door to useful quantum computing. Computer simulations in the same paper say that with the better physical parts already seen in test labs, the same design could reach one mistake in a million cycles, or better. For now, nobody’s bank password is at risk. However, the direction is clear. The machines that could break today’s encryption are being built one working piece at a time. THE FIX IS REAL.

Business ideas from this paper

1. 1. Quantum-safe migration for banks and mobile-money companies

   When error correction works, the codes that protect money transfers today will not last. Governments already publish “post-quantum” encryption standards. Most companies in Africa and Europe have not started moving. Sell the audit and the migration plan.

   Buyer Banks, mobile-money operators, telecoms. First test Ask ten security chiefs for a 30-minute call about their post-quantum plan. Five who say they have none, and want one, is a business.

2. 2. A scanner that finds old encryption in software

   Before you can fix encryption you have to find it. Build a tool that reads a company’s code and lists every place that uses the old methods, with a risk score.

   Buyer Software teams at banks, hospitals and governments. First test A free scan for three companies in exchange for a quote you can publish.

3. 3. A short course for engineers who will never touch a quantum computer

   Most engineers do not need to build one. They need to know what changes for them. Teach that in four evenings.

   Buyer Employers and universities in Nairobi and Budapest. First test Pre-sell twenty seats before you record a single lesson.

How sure can you be?

This is a preprint, so other scientists have not yet reviewed it. It was written by the company that built the machine, which means the same people measured their own product. In its favor, the paper reports every result with error bars and used no postselection. The main mistake rate has wide error bars, because rare events are hard to count. Wait for an outside group to repeat the measurement before treating the exact numbers as settled.

Source: Experimental validation of a compact fault-tolerant architecture for trapped ions, arXiv preprint, September 2026. arXiv:2609.03194 · arxiv.org (preprint · not yet peer reviewed).

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.