The Gap That Would Not Close
One puzzle has resisted resolution for years: why certain computations a quantum system performs almost instantly remain, in principle, beyond the reach of any classical machine. This is not a question of engineering patience. It is a question about the shape of physical law itself. Classical computers shuffle bits; quantum systems use superposition and entanglement to explore a space of possibilities that grows exponentially with every added qubit.
Random circuit sampling became the standard arena for testing that gap. The task is deliberately stripped of practical use: apply a sequence of random quantum gates to a register of qubits, then sample the resulting bitstrings. The ideal output distribution is exponentially hard to compute classically, so any machine that produces the right distribution has demonstrated quantum computational advantage. The frontier has moved steadily since: larger registers, deeper circuits, better fidelities.
The most recent benchmark before the current work, as reported by the team, came from Google’s Willow chip, which the team says ran a large-scale circuit at significant depth. That surpassed the earlier Zuchongzhi 3.0 result, which the team cites as involving fewer qubits and lower depth. Each step widened the classical simulation cost, but the experiments remained laboratory-bound. Access was limited to the teams that built the hardware. The community could read the papers but could not run the machines.
The platform is operated by a quantum group at a Chinese telecommunications company in Hefei. Its stated aim is not to win a benchmark contest but to democratize access to high-performance quantum hardware — to let researchers outside the builder’s walls validate and explore practical quantum advantages for themselves.
The instrument that made this possible is the Tianyan-287 system, a superconducting processor modeled on the Zuchongzhi 3.0 design. It integrates 105 physical qubits. The team reports high operational fidelities across the processor.
The fidelities are the headline. Single-qubit gates reach 99.90 percent. Two-qubit gates reach 99.56 percent. Readout fidelity reaches 98.7 percent.
Reliability is the second half of the story, and it is the half that cloud users care about most. The team reports that the processor sustains high operational fidelities for single-qubit gates, two-qubit gates, and readout.
Eighteen Minutes Against Sixteen Thousand Years
The benchmark task itself is where the numbers become difficult to hold in the mind. The team ran random circuit sampling on a 74-qubit region of the processor across 24 cycles. The two-qubit gates were applied in predefined patterns within each cycle, and single-qubit gates were drawn at random.
The platform completed one million samples in 18.4 minutes. To replicate that same sampling endeavor, a state-of-the-art classical supercomputer would need approximately 16,000 years. The estimate comes from a tensor network algorithm, currently the state of the art in classical simulation of random quantum circuits, applied under two memory-constrained scenarios.
In the first scenario, memory is capped at the capacity of a state-of-the-art supercomputer. Generating one million independent bitstrings at a low target fidelity would require a very large number of floating-point operations, and the simulation lands at approximately 16,000 years.
What if memory were effectively unlimited? The team set the limit above a very large memory threshold, including Frontier’s total memory and storage — a configuration that is not practically feasible but serves as a lower bound on sampling cost. Even then, generating the samples at that fidelity would require a substantially reduced but still enormous number of floating-point operations, corresponding to an estimated classical simulation time of many years.
Verifying the fidelity of a full random quantum circuit is its own difficulty, because the ideal probability distribution cannot be efficiently simulated classically. The team worked around this with patch circuits — circuits constructed by selectively removing two-qubit gates between regions, which can be validated independently. They implemented a four-patch version and a full 74-qubit circuit, each with multiple layers, and measured linear cross-entropy benchmarking fidelities for both. The four-patch circuit showed experimental and estimated fidelities consistent with the error model, indicating that the discrete error model reliably predicts fidelity even at large scale.
For the largest configuration — the full 74-qubit, 24-cycle circuit — the team collected a large number of bitstrings.
A Tool for the Community, Not a Trophy
The cloud service is the part of this work that changes who can participate. Access runs through Cqlib, an open-source software development kit designed for working with quantum systems at the level of extended quantum circuits, operators, and primitives. It provides an end-to-end toolkit for scientific experiments and software development, and it outlines a four-step workflow for executing tasks on the platform. The platform’s cloud-accessible systems include Tianyan-287, alongside earlier Tianyan processors, all reachable through the same SDK.
Earlier Tianyan systems, built on a previous-generation processor similar to Zuchongzhi 2.0 and integrated with high-performance computing clusters, enabled quantum-classical hybrid computing. Later Tianyan systems extended that line and served as engineering explorations of system scaling limits.
The stated ambition is to accelerate the arrival of the era of quantum computing utility. The field’s center of gravity has shifted, in the team’s own framing, from “how to compute faster” to “how to compute usefully” — from raw speed to the question of which practical scientific challenges quantum computing can genuinely solve. Possessing computational advantage is described as a prerequisite for exploring that question, not the answer to it. A cloud-accessible superconducting machine with advantage capabilities lets a broader range of researchers and engineers join the journey toward practical computing.
The experiment’s contribution to fundamental research is precise and deliberately narrow. It does not claim to simulate the universe, solve cosmology’s deepest questions, or replace classical computation for any general task. It demonstrates that a 74-qubit, 24-cycle random circuit can be sampled at a fidelity where the classical cost is measured in millennia under realistic memory constraints and in many years even under physically impossible ones. It shows that the error model predicts large-scale fidelity from small-scale patches. And it puts that machine behind an open SDK, so that the next validation does not have to come from the team that built it.
That last point is the quiet one, and it may outlast the headline number. What persists is the arrangement: high-fidelity hardware, a documented classical cost under two memory scenarios, and a software kit that lets an outside researcher check the arithmetic independently.
Sources
- Yahoo Tech — Portal copy — Original: Forbes.com (https