Home » Quantum Computing » Nighthawk r2 Quantum Advantage Claim: 19 Seconds vs. 110 Years (Est.)

Nighthawk r2 Quantum Advantage Claim: 19 Seconds vs. 110 Years (Est.)

A preprint claims IBM’s Nighthawk r2 chip sampled in 19 seconds what a supercomputer might need over a century to match. The estimate is unverified, and here is why.

Abstract glowing circuit lattice illustrating the Nighthawk r2 quantum advantage breakthrough in quantum computing

What would you do with 110 years of computing time? That’s roughly how long the paper’s authors estimate it would take the Frontier supercomputer to match something IBM’s quantum chip did in 19 seconds. It’s the headline behind the new Nighthawk r2 quantum advantage claim, and it is a claim in a preprint, not a verified result. The twist worth watching: the run used IBM’s standard cloud stack.

The authors, led by Tigran Sedrakyan of the quantum software company BlueQubit according to ZME Science, ran the experiment on IBM’s 120-qubit Nighthawk r2 processor. They report using the standard cloud stack with no benchmark-specific calibration. They used 61 qubits, ran 36 cycles of operations, and pulled one million samples in 19 seconds. They estimate a classical method would need more than a century on Frontier, one of the fastest supercomputers on Earth, under favorable memory assumptions. The paper posted September 23, 2026. You can read the full paper on arXiv for the math behind that estimate.

How the run compares with Sycamore

Runtimes in seconds, quantum and classical, for the two experiments. The classical bar is the 2021 Sunway rerun of the Sycamore task.

Run Seconds
BlueQubit on IBM Nighthawk r2 (2026), quantum

19 seconds

Google Sycamore (2019), quantum

About 200 seconds

Sycamore task rerun on China’s Sunway supercomputer (2021), classical

About 304 seconds

Source: ZME Science, Sep 29, 2026; arXiv:2609.28657. Sycamore used 53 qubits and was originally estimated to need about 10,000 years classically. Nighthawk r2 used 61 of 120 qubits, with an estimate of more than a century on Frontier (about 110 years) that has not yet been tested. The preprint is not peer reviewed.

That Sycamore history is the point. Google’s 2019 claim got challenged once classical researchers found smarter shortcuts, and the 10,000-year estimate collapsed to about 304 seconds. The same could happen here, since the 110-year estimate rests on one specific classical method, not on every method that may come. ZME Science’s coverage raises that pushback too.

The task is random-circuit sampling. Qubits get hit with random operations until the output becomes a probability pattern that’s brutally expensive for an ordinary computer to fake. It’s a stress test, not a product demo. The authors used two separate fidelity estimators, mirror benchmarking and cross-entropy benchmarking, and report that they agree across depths. Both measure how noisy the chip is. Neither proves the classical cost estimate. The authors call the run the first quantum advantage demonstration for vanilla random-circuit sampling on a broadly accessible commercial processor, but that “first” is their own claim, and nobody outside the team has confirmed it.

Why It Matters to You

Almost nobody benefits directly yet. Random-circuit sampling has no known practical use, and the paper is an unreviewed preprint. As an illustration only, take a data scientist at a small biotech startup wondering whether a quantum approach could someday speed up drug screening: this result doesn’t help her yet. What might change is access. Google’s Sycamore ran on Google’s own hardware, while this ran on a chip IBM offers through its cloud, and ZME Science says outside researchers can reach it there. IBM’s pricing page lists a free Open plan with up to 10 minutes of runtime a month and pay-as-you-go at $96 a minute, billed by the second. At that rate 19 seconds would cost about $30, ignoring the calibration and benchmarking runs behind the result. The page doesn’t say which processors each plan includes, so check with IBM before planning on Nighthawk r2. Anyone who can reach the same hardware can try to reproduce the run, and classical researchers can try to beat the 110-year estimate. We also covered IBM’s error-mitigation work and how IonQ’s error decoder kept pace on a laptop chip, in simulation.

What to watch, and what the numbers don’t show

If classical researchers shrink the 110-year estimate the way they did with Sycamore, the claim weakens. If the gap holds up, verified quantum advantage stops being a once-a-decade headline and starts being something smaller labs plan budgets around. Until then:

  • The paper is an arXiv preprint posted September 23, 2026, and it has not been peer reviewed, according to ZME Science.
  • The 110-year figure is an estimate. ZME says it assumes one tensor-network method and unlimited memory, and better algorithms or approximations could shrink it.
  • The surviving quantum signal at 36 cycles was small. The paper reports an XEB fidelity of about 0.23 percent (2.3 x 10-3). The authors argue this is the point where classical simulation gets very costly while the signal is still detectable.
  • The two checks measure noise on the chip. They do not directly verify the classical cost estimate.
  • IBM’s tracker says advantage claims rely on community review and rapid iteration. A different BlueQubit candidate on that tracker, using peaked circuits, was overtaken by new classical methods in February 2026.

Quick answers

What did the experiment actually do?

It ran random-circuit sampling on 61 qubits of IBM’s Nighthawk r2 for 36 cycles and produced one million samples in 19 seconds. The task has no practical use.

Is the 110-year figure a fact?

No. It is the authors’ estimate for one classical method on Frontier. ZME notes it assumes unlimited memory and that better classical methods could cut it sharply.

Can I run this on IBM’s cloud?

IBM’s pricing page lists a free Open plan with up to 10 minutes of runtime a month and pay-as-you-go at $96 a minute. The page does not say which processors each plan includes.

Sources

Related Reading