Organizer Report: Two Classical Solvers Broke a Quantum-Advantage Challenge
qBitTensor says two independent solvers found the hidden outputs of its 48–56-qubit peaked circuits after a three-month stand. The implementations are not yet public, so the result is a red-team report—not an independently reproduced benchmark.
.jpg){kind=link}
A public quantum-advantage challenge did what a useful benchmark should do: it gave classical researchers a concrete target. On October 5, organizer qBitTensor reported that two independent solvers recovered the hidden outputs of its third peaked-circuit milestone after the instances had resisted attempts for nearly three months. The reported attacks ran for roughly four hours on one offline GPU. That is a company-reported counter-result, not yet an independently reproduced benchmark.[1][2][3]
Long circuits can still preserve a shortcut
Peaked circuits are designed so that one output bit string is much more likely than the rest. A quantum processor can sample that peak, while anyone can verify a proposed answer cheaply. BlueQubit deepened otherwise manageable circuits with mirrored identity-like blocks, hidden permutations and other obfuscation, then invited classical attacks. The important distinction is between making a circuit long and making its computation structurally hard: extra gates can hide a shortcut without removing it.[2][3]
The attack surface was structural
That distinction had already produced one reversal. An April preprint showed that a 56-qubit, 1,917-two-qubit-gate instance could be contracted in about one hour on a single A100 GPU. Its method split the circuit around the mirror, extracted hidden permutations through an “unswapping” step and rewired the remaining tensor network. qBitTensor now says two new solvers broke Milestone 3 independently, using different routes to expose or cancel related structure in 48–56-qubit circuits with more than 1,400 two-qubit gates.[1][3][4]
A public failure can improve the benchmark
QubitWire's analysis: the failure is informative because verification and falsification were built into the challenge. A large qubit or gate count is not itself evidence of classical hardness when those gates mostly disguise an identity operation. Obfuscation raises the search cost, but a public red team can discover which structure still leaks. The useful comparison is therefore not quantum runtime versus yesterday's simulator; it is quantum runtime versus the best attack after the circuit family and its construction have been exposed.[1][2][3][4]
The new attacks still need reproduction
The principal limitation is that qBitTensor and BlueQubit are organizers of the challenge and, at announcement time, the two new solver implementations and full reproducibility package were not public. The four-hour figure therefore remains organizer-reported, and one broken construction does not rule out every peaked-circuit benchmark.[1][2]
Publish the solvers before Milestone 4
The informative next test is to release both solvers, exact Milestone 3 instances and compute logs, then benchmark them against tensor-network and other classical baselines under a preregistered budget before advancing Milestone 4.[1][4]