Go further · Chapter 14 · 14b · A step deeper
Has any quantum computer really beaten an ordinary one?
The short answer (tap to open)
Yes, on test tasks chosen to suit the machine, and some of those wins later fell to better ordinary methods. Since 2025 claims have moved towards checkable physics results and a few practical-sounding tasks, but none yet shows a clear advantage on a useful problem that independent experts accept. This page grades each major claim on three questions.
What are the three questions?
Last checked: 2 October 2026.
- Useful or made-up? Does anyone need the answer, or was the task picked to show off the machine?
- Ahead of the best ordinary method? Not just ahead of an old or slow one.
- Checked by others? Has an independent team tested or tried to beat it?
How do the claims score?
| Claim | Task | Ahead? | Checked? | Grade |
|---|---|---|---|---|
| Google Sycamore, 2019 | Made-up sampling | Matched by 2022 and 2024 ordinary methods | Yes | Overturned |
| China's Jiuzhang, 2020 to 2023 | Made-up sampling with light | Matched in 2024 | Yes | Overturned |
| Google Willow, 2024 | Made-up sampling | Company estimate: 10²⁵ years | Not overturned yet | Stands, made-up task |
| D-Wave, March 2025 | Simulating a magnetic material | Partly matched by Flatiron (2026) | Yes, disputed | Contested |
| JPMorganChase and Quantinuum, March 2025 | Certified random numbers | A task ordinary computers can't do the same way | Peer reviewed | Shown, narrow use |
| Google Quantum Echoes, October 2025 | Measuring a physical effect | 13,000 times faster than Frontier | Peer reviewed; not yet reproduced classically | Stands, physics task |
| HSBC and IBM, September 2025 | Bond-trade prediction | Disputed; gain may come from noise | Experts challenged it | Preprint, disputed |
| China's Jiuzhang 4.0, May 2026 | Made-up sampling | Claims 10⁴² years | No rebuttal yet | Too early |
| IBM and partners' tracker, 2026 | Three physics and maths problems | Claimed; one of the three matched on ordinary GPUs (preprint, Aug 2026) | Partly | Contested |
Made-up sampling
The pattern: made-up tasks came first and some fell. The newest claims aim at tasks that are checkable and closer to useful. A 2026 expert view argued advantage has been shown, with scaling and checking still open.
Why do claims fall?
A quantum advantage claim is a bet that no one will find a cleverer ordinary method. Ordinary-computing experts love that challenge: tensor networks, better sampling tricks and bigger GPU clusters have overturned several claims. That's science working, not a scandal. The claims that survive teach the most.
Check
Two quick ones
Collect
Your aha card
Three questions, every claim.
Useful? Ahead of the best? Checked by others?