BlueQubit claims random-circuit sampling advantage on IBM's cloud Nighthawk r2 chip
A BlueQubit preprint reports 61-qubit random-circuit sampling on IBM's ibm_phoenix in 19 seconds, estimating about 110 years on the Frontier supercomputer.
Researchers at BlueQubit, working with collaborators affiliated with EPFL and the XPRIZE Foundation, report a quantum computational advantage in random-circuit sampling on a commercial IBM processor that anyone with cloud access can use. Their preprint says a 61-qubit experiment on IBM's 120-qubit ibm_phoenix chip, built on the second-generation Nighthawk r2 architecture, produced in 19 seconds a sample set that would take an estimated 110 years to reproduce on the Frontier supercomputer.
What was announced
The paper, Quantum computational advantage in random-circuit sampling on IBM superconducting quantum computers, was posted to arXiv and covered by Quantum Computing Report on October 28. The authors stress that the circuits were compiled and run through the standard Qiskit cloud workflow with no calibration tailored to the benchmark. To their knowledge, they write, it is the first such demonstration on a commercially and broadly accessible processor that most non-expert users could replicate. The circuits, scripts and the full set of one million samples are published on GitHub.
The technical details
Random-circuit sampling runs layers of random gates and asks the machine to output bitstrings; the challenge for a classical computer is to produce samples of comparable quality. The team selected 61 qubits and 102 couplers on the square-lattice chip using the latest calibration data, then applied native CZ gates in a four-color brickwork pattern.
Fidelity was estimated two independent ways, with mirror benchmarking and with patched cross-entropy benchmarking (XEB), and the methods agreed across depths. At the anchor depth of 36 cycles, which contains 918 CZ gates, the estimated fidelity was about 2.3 × 10⁻³. That depth is where the cost of tensor-network simulation stops growing with depth for this system size. Under the authors' cost model, sampling one million bitstrings classically would take about 1.2 × 10²⁷ machine operations, roughly 110 years on Frontier. The whole experiment used about 11 minutes of processor time.
The paper also measures progress between IBM chip generations: per-cycle fidelity rose from 0.836 on the first-generation Nighthawk r1 device (ibm_miami) to 0.872 on r2, which adds up to more than an order of magnitude at deep circuits. The authors tested "severing" attacks that try to fake the XEB score by cutting the circuit into pieces, and found that none reached the experimental score at 36 cycles; the strongest fell about a factor of ten short.
Why it matters
Because the experiment ran through ordinary cloud access and its circuits, data and code are public, other users can try to reproduce it rather than taking the claim on trust. The paper also gives a concrete yardstick for how much IBM's superconducting hardware improved between two chip generations.
What to watch
The authors are careful about the limits of the claim. Their classical estimate assumes unlimited memory and single-amplitude contractions, and they note that approximate, multi-amplitude, slicing and matrix-product-state methods could lower classical costs, as happened with earlier advantage experiments. The work is a preprint, so peer review and attempts by classical-simulation groups to narrow the gap are the next tests.
Sources
Facts on this page come from the sources below, retrieved by DesignQuantum.
- Quantum computational advantage in random-circuit sampling on IBM superconducting quantum computers (arXiv preprint) arxiv.org
- rcs-nighthawk: circuits, data and code (GitHub) github.com
- BlueQubit Benchmarks Quantum Computational Advantage in Random-Circuit Sampling on IBM Nighthawk r2 Hardware (Quantum Computing Report) quantumcomputingreport.com