How well does a real quantum computer work?

Quantum computers are real machines that anyone can sign up to use, so how well do they actually work? For small jobs, almost perfectly. For the famous job, using Shor's algorithm to factor a number the way it would to break encryption, the machine did no better than guessing.

To find this out, on October 11, 2026 we ran eight small programs from the lessons on a real quantum computer with 156 qubits. We ran each program 4,000 times and compared the results with what a perfect quantum computer would give. Each line below is one program, and its name links to the lesson that explains it.

Programs that use one or two qubits come out right almost every time

The simplest programs set, mix and entangle one or two qubits, and they gave the right answer 98.8% to 99.7% of the time. That includes interference. Mixing a qubit twice makes the two halves of its blend cancel, so a perfect machine always lands back on 0, and this one landed on 0 99.7% of the time.

The most striking result comes from Bell's game, from the entanglement lesson. Two players are each asked a random question and have to answer without talking to each other. Any plan they agree on in advance wins at most 75% of the time. Two entangled qubits on this machine won 83.9% of the time. Experiments showing that edge earned the 2022 Nobel Prize in Physics, and the same test now takes a few seconds on a machine anyone can sign up to use.

Each extra step gives the machine another chance to make a mistake

Things get harder as programs get longer. Each step a quantum computer takes has a small chance of going wrong, and later steps build on the mistakes of earlier ones. It works like a game of telephone: whispered from one person to the next, a message arrives intact, but whispered along a line of 200 people it comes out garbled. The search programs show this happening. Finding one code out of 4 took 2 steps that act on two qubits at once, and the machine found it 97.6% of the time. Finding one code out of 16 took 161 of those steps, and it found the code 42.6% of the time. A perfect quantum computer finds it 96% of the time, and a random guess finds it 6% of the time. So the search still worked, but more than half of its tries were wasted.

Factoring 15 the long way failed

Shor's algorithm is the reason quantum computers threaten encryption, because it finds which two numbers were multiplied to make a large number. We ran it on the number 15 the long way, doing the arithmetic step by step, the way it would have to be done for a large number. That took 1,266 two-qubit steps. On a perfect machine every reading points toward the answer, 3 and 5, because 3 × 5 = 15. On the real one, 27.1% of readings did. Picking readings at random would give 25%, so the machine learned nothing about 15.

You might wonder why we didn't use a simpler version. 15 happens to have unusually easy arithmetic, so a program written for 15 alone needs only about 20 of those steps. A machine that handles 161 steps should manage that, and from next month we run the short version too. But it says little about breaking encryption, because no large number has shortcuts like these. The next number up, 21, needs about 2,405 steps, more than 100 times as many, and no quantum computer has factored 21 without a shortcut.

Some reports of factoring 15 or larger numbers went further and built the answer into the program itself, a bit like a magician who picks your card for you and then guesses it. In 2013, three IBM researchers showed that a shortcut like that can "factor" any number with a circuit that amounts to flipping coins.

Breaking real encryption takes billions of steps in a row without an uncorrected mistake. The machine we used loses the thread within a few hundred. That's why the number that says most about progress is how many steps a machine can take before its answers turn into noise. Qubit counts say less, though the Q-Day tracker follows those too.

We rerun these programs at the start of every month.

Sources

• Smolin, Smith and Vargo, on shortened factoring circuits, Nature, 2013
• Craig Gidney, "Why haven't quantum computers factored 21 yet?", 2025, for the step counts for 15 and 21
• The 2022 Nobel Prize in Physics, for experiments with entangled particles

Each program ran 4,000 times on IBM's free plan, on whichever machine was least busy (Kingston for the latest run). The factoring program uses a multiply-by-7 circuit built for 15, with four qubits for the reading. Programs were written with the Qiskit software and checked on a perfect simulator first.