· Benchmarks
Nineteen seconds. That's how long it took a quantum computer you can rent by the hour to do something a top supercomputer would need about a century for.
No special lab tuning. No bespoke hardware. Just the standard cloud workflow, and a task designed from the start to be brutal for classical machines.
IBM's Nighthawk r2, a 120- superconducting processor accessible through its standard cloud platform, generated one million samples from a random quantum circuit in 19 seconds. Using tensor-network contraction to estimate the classical cost, researchers led by Tigran Sedrakyan of BlueQubit calculated that reproducing those samples would take a top supercomputer roughly 110 years. The experiment used 61 qubits at a circuit depth of 36 cycles, with no special calibration beyond the platform's standard workflow, and the team has released the circuits, samples, and analysis code publicly. The result, described in an unreviewed preprint, is framed as a benchmark in an ongoing contest between quantum and classical computing, not a final verdict — classical researchers have found shortcuts against similar claims before.
A commercially accessible 120-qubit chip, run through IBM's standard cloud workflow, produced the sampling result.
A Quantum Computer Just Beat a Century of Supercomputing — In 19 Seconds
In just 19 seconds, an IBM quantum processor generated one million samples from a random quantum circuit — a task the researchers estimate would take a top supercomputer around 110 years.
The experiment ran on Nighthawk r2, a 120-qubit superconducting processor that anyone can access through IBM's cloud platform, using its standard workflow rather than special lab tuning.
The work, led by theoretical condensed matter physicist Tigran Sedrakyan of BlueQubit, is described in a preprint on arXiv and has not yet been peer-reviewed.
The researchers describe it as the first demonstration of quantum advantage on a vanilla random-circuit sampling task, run on hardware that non-expert users can access and replicate themselves.
That accessibility is the detail that separates this from many earlier quantum advantage claims, which typically ran on bespoke, specialized laboratory equipment.
What "Quantum Advantage" Actually Means
Quantum advantage is not a vague marketing phrase. It describes a specific, falsifiable claim: that there exists a computation a quantum computer can do that a classical computer effectively cannot.
That race has been running since 2018, when researchers at UC Berkeley provided strong theoretical evidence for a leading candidate task capable of clearing that bar.
In 2019, Google announced that its 53-qubit Sycamore processor had performed that task beyond the practical reach of a classical supercomputer — a widely reported milestone.
It did not settle the argument. Classical researchers kept finding creative ways to claw back ground, reproducing similar results with smarter algorithms, while quantum researchers kept pushing their systems further.
This new IBM result is the latest move in that ongoing contest, not a final answer to it.
Random Circuit Sampling: The Task Built to Break Classical Computers
The task at the center of this contest is called random circuit sampling, or RCS, and the name describes exactly what happens.
A quantum circuit is a sequence of operations performed on qubits, roughly analogous to the operations a conventional computer performs on ordinary bits.
In RCS, those operations are chosen largely at random, which builds up an increasingly complex quantum state as the circuit runs.
The processor then measures that state over and over, producing long strings of ones and zeroes — the "samples" that give the technique its name.
Producing the samples is the easy part for a quantum computer. The hard part, for a classical computer, is calculating the probability distribution needed to reproduce those same results.
As the number of qubits and operations grows, that calculation becomes Sisyphean for classical hardware — which is exactly why RCS was chosen as a candidate benchmark in the first place.
Nighthawk r2: An Off-the-Shelf Chip, Not a Lab Special
What makes this result stand out is not just the number 19 seconds. It is what kind of machine produced it.
Sedrakyan and colleagues did not use a specialized experimental quantum processor built for this one demonstration. They used Nighthawk r2, a 120-qubit superconducting processor commercially accessible to ordinary users through IBM's cloud platform.
From that chip, they selected 61 qubits and subjected them to random circuits of increasing complexity, running up to 40 cycles — repeated rounds in which the qubits are manipulated and entangled.
Significantly, they ran the experiment using the platform's standard cloud workflow, with no special calibration tailored to squeeze out this specific result.
The researchers have also released the circuits, samples, and analysis code publicly, so other groups can attempt to reproduce or challenge the result themselves.
Finding the Goldilocks Zone: 61 Qubits, 36 Cycles, 19 Seconds
Not every circuit depth produces an interesting result. Too few cycles, and a classical computer can keep up easily. Too many, and noise in the hardware can start to degrade the signal.
| Parameter | Value |
|---|---|
| Qubits used | 61 of 120 available |
| Circuit depth (Goldilocks zone) | 36 cycles |
| Two-qubit gates in that circuit | 918 |
| Samples generated | 1 million |
| Time taken | 19 seconds |
The team found their Goldilocks zone at 36 cycles, where the circuit involved 918 two-qubit gates — complex enough to be genuinely hard classically, but still within the hardware's reliable operating range.
Under those conditions, Nighthawk r2 generated one million samples in 19 seconds. That speed is the part that is straightforward to verify: the samples exist, and the time is measured directly.
What is much harder to pin down is the other side of the comparison — exactly how long that same task would take a classical machine to reproduce.
How Do You Know It Would Take 110 Years? The Tensor-Network Trick
Because the task is supposed to be effectively impossible for a supercomputer, and because the world's top supercomputers are in constant high demand, researchers can't simply run the program and time it.
Instead, they rely on a technique called tensor-network contraction, which estimates the computational cost of simulating a given quantum circuit on classical hardware without actually running the full simulation.
Using that approach, Sedrakyan and colleagues calculated that reproducing their one million samples would require roughly 1.2 × 10²⁷ computational operations.
To translate that into a human timescale, they compared it against Frontier, the world's first exascale supercomputer and the fastest machine until 2024, capable of over a quintillion operations per second at peak.
Using a more conservative estimate of Frontier's sustained, real-world performance rather than its theoretical peak, they arrived at roughly 110 years of computing time.
That figure is an estimate built on a specific method, not a measurement taken from an actual 110-year supercomputer run — a distinction the next section takes seriously.
The Caveats: Why 110 Years Isn't a Law of Physics
The researchers themselves are careful to flag the limits of their own claim, and it is worth taking that caution at face value.
History gives good reason for that caution. Classical computing researchers have repeatedly found creative shortcuts after earlier quantum advantage claims, and more efficient algorithms could substantially cut down the estimated 110 years.
Sedrakyan and colleagues acknowledge this explicitly. Their claim is not that no classical computer will ever match Nighthawk r2's performance on this task — it is that, by this particular benchmark, the quantum processor has pulled substantially ahead for now.
The findings themselves are also not yet peer-reviewed. They currently exist as a preprint on arXiv, open for scrutiny but not yet formally vetted by independent reviewers.
None of that erases the result. It just places it correctly: a strong, checkable claim in an ongoing scientific contest, not a settled final verdict.
Manohar Insight
What stands out to me most is not the 110-year figure — headline numbers like that always deserve a second look. It is the word "commercially accessible."
Earlier milestones in this race were often demonstrated on specialized hardware built and tuned for one specific experiment. This one ran through the same cloud pipeline any paying customer can use, with no special calibration.
That is a meaningfully different kind of claim. It says something about where the technology actually stands today, not just about what is possible in a hand-built laboratory setup.
I would also give real credit to the transparency here. Releasing the circuits, samples, and analysis code invites the classical computing community to try to break the claim, which is exactly how this kind of contest should be settled.
The honest framing matters too. The researchers describe this as a challenge thrown down in a collegial contest, not a declaration of quantum supremacy — and given how many times classical algorithms have clawed back ground before, that humility looks well earned.
The detail I would watch closest is whether classical researchers find a smarter simulation method that meaningfully shrinks that 110-year estimate. That back-and-forth, more than any single number, is what will actually tell us how real this advantage is.
The real story here is not that a quantum computer won a race. It is that the race is now being run on hardware ordinary users can rent by the hour — and that changes who gets to keep score.
Join the Signal
If classical researchers eventually find a shortcut that shrinks that 110-year estimate, would that undercut this result — or is that back-and-forth exactly how quantum advantage is supposed to be proven? Share your take in the comments.
Discussion