BlueQubit claims quantum advantage in random circuit sampling on IBM hardware

Started by PixelTea26, Oct 31, 2026, 08:59 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Topic: BlueQubit claims quantum advantage in random circuit sampling on IBM hardware   Views(Read 69 times)
Active members in this topic:
PixelTea26(1) Bob92(1) GlassyCandle(1) Blue Sasha(1)

PixelTea26

Another quantum advantage claim has arrived, this time on IBM's newest processor. Quantum Computing Report says BlueQubit, working with EPFL and the XPRIZE Foundation, claims quantum computational advantage in random circuit sampling on IBM's 120 qubit Nighthawk r2 chip

The experiment used 61 active qubits and 102 couplers, run through IBM's standard cloud queues with no special calibration. At the main depth of 36 cycles, the circuits contained 918 two qubit gates. The team collected 106 samples in 19 seconds of processor time, and the circuits, scripts and data are published on GitHub

The key claim is about classical cost. The team estimates that reproducing the result at 36 cycles would take Oak Ridge's Frontier supercomputer around 110 years, compared with seven days at 28 cycles and 44 seconds at 20 cycles. Attempts to fake the result with classical shortcuts scored at least ten times lower than the real experiment

Per cycle fidelity improved from 0.836 on the earlier Nighthawk r1 to 0.872 on r2, showing how fast IBM's hardware is improving. Still, the absolute fidelity of the whole circuit is tiny, around 0.0023, which is normal for this kind of benchmark

Random circuit sampling has no practical use, and previous advantage claims by Google and Chinese teams were later undercut by better classical methods. The classical estimates here assume exact simulation with unlimited memory, which leaves room for clever shortcuts. The work is a preprint and not yet peer reviewed

Does random circuit sampling prove anything useful? And how long before classical researchers find a faster way to simulate it?


Bob92

One hundred and ten years on Frontier sounds dramatic, but those estimates assume brute force methods. Tensor network tricks could cut that enormously. Wait for the experts to try. Patience is wise here

GlassyCandle

XPRIZE being involved is a nice touch. Their quantum prize is pushing teams towards useful applications. Benchmarks like this are a stepping stone. Real uses will come later

Benchmarks show the hardware is ready to try. Progress is clear
Cashback on everything or it didn't happen

Blue Sasha

Running it on standard cloud queues without special calibration is the interesting part. That means anyone could reproduce it with IBM access

Open data on GitHub helps too. Transparency like that deserves credit. Others can now try to break it. That is good science