Data/Quantum computing/China · Singapore
537 million is a model, not a measurement
A review prices quantum error cleanup at 537 million times more runs for 1,000 noisy places in a circuit. No machine was run to get that figure, and it is not a count of qubits.
537 million is how many times more runs a common quantum cleanup method would need to fix errors at 1,000 places in a circuit, at a half percent error rate. It comes from a table in an August 2026 review, which we score 3 out of 10. It is arithmetic from a simple model, not a hardware measurement.
The number
537 million. That is how many times more runs you would need to clean up errors at 1,000 places in a quantum circuit, when each place goes wrong half a percent of the time.
Cleaning up here means error mitigation. You run the same noisy program over and over, then do maths on the pile of answers to guess what a perfect machine would have said. The maths is cheap, but the extra runs are not.
Our full write-up is here: Erasing errors at 1,000 spots needs 537 million times more runs.
Where it comes from
The paper is “Practical Error Suppression and Mitigation for Reliable Quantum Computing”, from the National University of Singapore, the Singapore Institute of Technology and Shanghai University. It is a preprint from August 2026.
It is a review, and it runs no experiment of its own. The 537 million comes out of one table, worked out from a formula, using a simple model where the only thing that goes wrong is a bit flipping. In that same table, 100 places cost 7.46 times more runs, and 500 places cost about 23,200 times more. The error rate never changed. Only the number of places did.
We score it 3 out of 10. That is low. It has three named institutions and a clear argument. However, it has no peer review, no new experiment, no code and no outside attention yet.
What it does not mean
It does not mean anybody measured this. No machine was run for it. The figure falls out of a formula, using a model where the only error is a bit flipping from 0 to 1. Real hardware noise is messier than that. The authors put the table in to show the shape of the growth, not to price a machine.
It does not mean 1,000 qubits. A place is a spot in the circuit, not a qubit. The same review records a real run on a 127-qubit machine with 2,880 two-qubit operations. This is because the places where errors creep in run far ahead of the count of qubits. Therefore, a modest machine can pass 1,000 places.
It does not mean quantum computers are 537 million times too slow. The factor is the extra runs for one cleanup method, at one error rate, at one circuit size. It is not a speed, and it is not a gap between quantum and ordinary computers. Use it that way and someone will correct you.
It does not mean error mitigation failed. It works today. The review records cleanup running on a circuit of 26 qubits, 120 layers deep, with about 1,080 two-qubit gates. It was later pushed onto that 127-qubit machine, so the method is real. The trouble starts when you ask for more.
It does not mean the answer is a faster machine. Look at the other row of the table. Hold the size at 100 places and double the error rate from 0.5 to 1 percent. The cost goes from 7.46 to 56.9. This is because the cost multiplies at every place rather than adding up. Therefore, a machine that makes fewer errors buys you far more than a machine that fires more runs per second.
And it does not mean the figure is exact. The paper says about 537 million, and two of the three numbers in that row are given as approximations. Read it as an order of size. The lesson is in how fast it climbs, not in the digits.
What it does mean
Error cleanup is not free, and its price is machine time, because cloud quantum time is sold by the second. A method that needs a million times more runs is a method nobody can pay for, however good the maths.
The growth is the whole story. The cost multiplies at every place you try to fix, so it does not creep upward. It jumps. Go from 100 places to 1,000, at the same error rate, and the bill goes from about seven times to hundreds of millions of times.
The review makes one more point worth taking. The old story had two chapters, noisy machines now and perfect machines later. Real machines are arriving in the middle, because they correct some errors and still leak others. Cleanup is being rewritten to sit on top of correction rather than replace it.
Why it matters to you
You may never touch a quantum computer, but you will still be sold one.
The pitch will show a right answer to a hard problem, and the missing line will be the run count. A result that took a hundred million runs and a result that took a thousand look the same on a slide. They are not the same claim.
If you do buy cloud quantum time, this table is your budget, and runs are the thing you pay for. Doubling the size of your circuit does not double your bill.
Do this today
Find the last quantum result you read about, and search the page for the words “shots”, “runs” or “samples”. If the number is not there, write to whoever published it and ask two questions. How many runs did this take? How does that number grow when the problem gets bigger? Those two answers will tell you more than the headline did.
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.
- The 537 million figure is arithmetic from a simple bit-flip model inside a review paper, and no quantum machine was run to produce it.
- The cost multiplies at every place you try to fix, so at the same 0.5 percent error rate the overhead climbs from 7.46 at 100 places to about 23,200 at 500 and about 537 million at 1,000.
- A place is a spot in the circuit rather than a qubit, and the same review records a real run on a 127-qubit machine using 2,880 two-qubit operations.
Questions people ask
Was 537 million measured on a real quantum computer?
No. The paper is a review and reports no new experiment. The figure comes from a formula applied to a simple model in which the only error is a bit flipping. The authors present the table to show how fast the cost grows, not to price any particular machine.
Does 1,000 places mean 1,000 qubits?
No. A place is a location in the circuit where you try to cancel an error, and one circuit can hold far more locations than qubits. The same review records a real run on a 127-qubit machine using 2,880 two-qubit operations, so a machine of that size can easily pass 1,000 locations.
Is error mitigation the same as error correction?
No. Error correction encodes information across many physical qubits so errors can be spotted and undone during the run. Mitigation works on the results afterwards and does not protect the state. The review argues they are layers of one strategy, and that mitigation is now being adapted to run on top of correction.
Does this mean error mitigation is a dead end?
No. The review records it working on real hardware, including a circuit of 26 qubits and 120 layers with about 1,080 two-qubit gates. The review's own conclusion is that these methods extend the reach of noisy hardware without turning it into a fault-tolerant machine. The limit is cost, not whether the idea works.
Mark Alex
Mark Alex is the founder and Managing Director of Real Biz Digital, a technology company operating out of Nairobi since 2018. He works in agentic AI and the Model Context Protocol, AI governance, enterprise software architecture and cybersecurity. He holds an MSc in Mechatronical Engineering from Obuda University in Budapest and a BSc in IT, Forensic Technology and Cybercrime, from USIU-Africa in Nairobi, and has published IEEE conference research on an AI-powered digital twin for greenhouse systems. He is the author of seven books. Between 2020 and 2024 he mentored more than 200 university students and interns in Nairobi. He writes every Just Out Tech article from the original research paper.