Tianyan Quantum Cloud Platform Overview

• Tianyan Quantum Cloud Platform is a cloud service offering remote access to superconducting quantum processors, featuring systems like Tianyan-287 with up to 105 physical qubits and a 74-qubit benchmark region.
• It integrates HPC-assisted orchestration and a full-stack software environment via the open-source Cqlib SDK, enabling efficient circuit construction, optimization, and calibration.
• The platform demonstrates quantum advantage by executing a 74-qubit, 24-cycle random circuit sampling task in 18.4 minutes, contrasting classical simulation times that span years or millennia.

Tianyan Quantum Cloud Platform is a cloud-accessible quantum computing service operated by China Telecom Quantum Information Technology Group and centered, in the cited demonstration, on the superconducting processor Tianyan-287. In the platform’s own framing, it is a quantum cloud service with “quantum advantage” capability: not merely a remote interface to quantum hardware, but a full-stack environment intended to expose hardware strong enough to run benchmark tasks beyond practical classical reach, while also providing software access through the open-source Cqlib SDK (Group, 11 Dec 2025).

1. Institutional role and platform scope

The platform is presented as a cloud service whose goals are to democratize access to advanced superconducting quantum processors, enable researchers, developers, and enterprises to run large-scale experiments remotely, provide a full-stack software environment through Cqlib, expose hardware with sufficient scale and fidelity to support quantum-advantage-class experiments, and support future exploration of practical NISQ applications and eventually fault-tolerant workflows (Group, 11 Dec 2025).

Within the Tianyan family, the paper names multiple superconducting systems accessible through the cloud and Cqlib: Tianyan-287, Tianyan-176, Tianyan-176 II, Tianyan-504, and Tianyan-24. It highlights a staged development trajectory. In 2023, Tianyan deployed Tianyan-176 and Tianyan-176 II, based on Zuchongzhi 2.0-like processors and integrated with HPC clusters for hybrid quantum-classical computing. In 2024, it launched Tianyan-504, a 504-physical-qubit superconducting system described as a commercially available large-scale engineering milestone. The focal system in the paper is Tianyan-287, described as Zuchongzhi 3.0-like and associated with the platform’s quantum-advantage demonstration (Group, 11 Dec 2025).

The paper also records a stronger institutional positioning: the authors explicitly characterize Tianyan as offering “cloud services with quantum advantage,” as giving the community access to “high quantum computational power,” and as “the first commercial cloud-access to leading quantum computing power.” Those formulations are the paper’s own claims rather than an externally adjudicated industry classification (Group, 11 Dec 2025).

Historically, Tianyan belongs to a broader evolution of quantum cloud services in which remote access, scheduling, simulators, and hybrid classical infrastructure become as important as the underlying qubit technology. Earlier Chinese cloud efforts such as NMRCloudQ already framed quantum cloud access as a way to expose otherwise inaccessible laboratory hardware, but did so on a 4-qubit liquid-state NMR backend with a very different control and measurement model (Xin et al., 2017). Tianyan extends the cloud-service concept to a superconducting system in the same class as Zuchongzhi 3.0-like hardware (Group, 11 Dec 2025).

2. Tianyan-287 hardware and operational characteristics

Tianyan-287 is described as a “Zuchongzhi 3.0-like superconducting quantum processor.” The main text states that it integrates 105 physical qubits and 182 couplers in a square grid lattice. The appendix further notes that one qubit is non-functional, leaving 104 working qubits in typical characterization. The headline quantum-advantage benchmark does not use the entire chip; it uses a tuned 74-qubit subset (Group, 11 Dec 2025).

The reported whole-system coherence values for Tianyan-287 are a mean operating $T_1 = 44.4\,\mu s$ and a mean $T_2^{CPMG} = 41.1\,\mu s$. Appendix data add that, across the 104 working qubits and over a 500 MHz tunable range, 2D $T_1$ measurements yield an average of $45.2\,\mu s$; one qubit is non-functional, and two qubits have suppressed $T_1$ due to two-level systems (TLS). For the 74-qubit subregion used in random circuit sampling, the paper reports that retuning and idle-frequency optimization yield average values of $T_1 = 47.7\,\mu s$ and $T_2^{CPMG} = 44.1\,\mu s$ (Group, 11 Dec 2025).

The appendix describes a calibration and operating pipeline designed to sustain those figures. The 74-qubit subset was optimized by biasing unused qubits to minimum operating frequency, deactivating inter-qubit couplings, dynamically tuning idle frequencies to balance coherence, TLS effects, and frequency collisions, and re-measuring and re-tuning around idle frequencies to stabilize $T_1$. This indicates that the benchmark region is an operationally selected and re-optimized subregion rather than an arbitrary subset (Group, 11 Dec 2025).

The abstract and main text report average fidelities of 99.90% for single-qubit gates, 99.56% for two-qubit gates, and 98.7% for readout. The paper also gives the corresponding average errors as $e_1 = 1\textperthousand$, $e_2 = 4.4\textperthousand$, and $T_2^{CPMG} = 41.1\,\mu s$0. Daily monitoring over 30 consecutive days, from August 29 to September 28, 2025, is said to show long-term stable performance (Group, 11 Dec 2025).

The hardware-level calibration details are unusually explicit. Single-qubit gates use a fixed gate time of 26 ns; drive frequency, amplitude, and DRAG coefficient are jointly optimized by an optimal control theory-based scheme; crosstalk is mitigated by anti-phase compensation cancellation; and verification is carried out by fully parallel XEB. Two-qubit gates are iSWAP-like with a fixed duration of 40 ns; calibration involves scanning optimal swap frequency, accounting for coherence and frequency-collision thresholds, fine calibration using odd-numbered gate sequences, and tuning coupling strength $T_2^{CPMG} = 41.1\,\mu s$1 and detuning frequency. Residual interactions are addressed by dynamic coupling-off technology and idle gate benchmarking and calibration. Readout uses X12-gate-driven 0-2 readout, and correlated readout errors are controlled below 2% using synchronous state preparation and measurement (Group, 11 Dec 2025).

3. Software stack, cloud access, and execution workflow

Access to Tianyan is provided through the cloud using Cqlib, an open-source SDK built on QCIS (Quantum Computing Instruction Set). The paper states that Cqlib supports work at the level of extended quantum circuits, operators, and primitives, and provides capabilities for circuit construction, compilation or transpilation, optimization, simulation, visualization, execution on cloud hardware, and result output or post-processing. Cqlib supports native hardware access to Tianyan and QuantumCTek, and interoperability with Qiskit, Cirq, and PennyLane through adapters (Group, 11 Dec 2025).

The access model described in the paper is a six-stage cloud workflow: construct circuits locally using high-level abstractions; download backend configuration for a chosen machine; transpile logical circuits to hardware-compatible instructions; submit jobs to the Tianyan cloud backend; wait for execution and backend or HPC processing; then retrieve results and post-process them. In the concrete Cqlib workflow, the user authenticates with a login key or API key, downloads machine configuration with download_config(machine='tianyan-287'), maps logical circuits to physical hardware through transpile(circuit, config), submits jobs, and retrieves processed outputs through task identifiers (Group, 11 Dec 2025).

The paper’s architecture description shows a layered service model with user-facing software and Cqlib, a circuit construction layer, a transpilation layer, a task submission and scheduling system, backend execution resources, post-processing and results analysis, and cloud storage with data retrieval. Backend resources include quantum hardware systems, quantum circuit simulators, and HPC servers. The orchestration model includes task queuing, workload distribution over heterogeneous resources, iterative hybrid execution, and local classical collaboration through HPC during execution (Group, 11 Dec 2025).

HPC participation is not incidental. During task execution, HPC servers act as the classical collaborator for circuit dispatch, sample collection, associated data processing, and iterative optimization. After sampling, an automatic optimization module on locally connected HPC refines iSWAP-like or fsim gate parameters. This means that the cloud service is architected as a hybrid quantum-classical platform rather than as a simple remote job relay to superconducting hardware (Group, 11 Dec 2025).

In the broader taxonomy of quantum cloud computing, a platform with Tianyan’s stated characteristics aligns with QCaaS or QaaS models and with the hybrid quantum-classical cloud pattern emphasized in current reviews. A plausible implication is that Tianyan should be analyzed not only as hardware exposure, but also as a scheduling, orchestration, simulator, and software-interface system (Nguyen et al., 2024). In that respect, Tianyan’s described stack also resembles the more general full-stack pattern later articulated in modular platform architectures such as QCI Connect, where SDKs, orchestration, heterogeneous backends, and workflow management are treated as core cloud functions rather than auxiliary tooling (Bertok et al., 12 Jun 2026).

4. Random circuit sampling and the quantum-advantage claim

The platform’s flagship demonstration is based on random circuit sampling (RCS) on a 74-qubit system over 24 cycles. The paper states that the random circuits are designed to be genuinely random, adapted to hardware layout, and structurally optimized to maximize classical simulation cost. The two-qubit patterns follow two principles: adapting the layout to chip architecture and maximizing classical simulation cost (Group, 11 Dec 2025).

The two-qubit gates are iSWAP-like and use four coupling layouts labeled A, B, C, and D. The paper contains a notation inconsistency: the main text gives the within-cycle sequence as ABCD–CDAB, whereas the figure caption gives ABCDCDBA. Both descriptions indicate a structured alternating pattern over the four coupling layouts, but the discrepancy is part of the published record. In each cycle, single-qubit gates are randomly chosen from

$T_2^{CPMG} = 41.1\,\mu s$2

The paper also reports experiments over circuits of 8–24 layers and distinguishes between a four-patch version and a full 74-qubit circuit (Group, 11 Dec 2025).

Because full ideal simulation is intractable at the flagship scale, verification is performed using patch circuits and linear cross-entropy benchmarking (XEB). The paper reports a 4-patch circuit experimental fidelity of 0.056%, a 4-patch circuit estimated fidelity of 0.064%, and an estimated fidelity of 0.054% for the full 74-qubit, 24-cycle circuit. It further reports that approximately $T_2^{CPMG} = 41.1\,\mu s$3 bitstrings were collected, with the figure caption specifying 91.255 million bitstrings (Group, 11 Dec 2025).

The runtime claim that anchors the platform’s quantum-advantage narrative is that Tianyan completes 1,000,000 samples for the 74-qubit, 24-cycle RCS task in 18.4 minutes. The classical comparison uses tensor-network simulation as the state of the art for random quantum circuits. Under a Frontier-like practical memory limit of 9.2 PB, the paper estimates that generating 1 million independent bitstrings at 0.054% fidelity would require $T_2^{CPMG} = 41.1\,\mu s$4 FLOPs; using Frontier’s peak single-precision throughput of $T_2^{CPMG} = 41.1\,\mu s$5 FLOPs/s, an assumed 20% floating-point efficiency, an adjustment for low target fidelity, and the assumption that each single-precision complex FLOP requires 8 machine FLOPs, the estimated time is $T_2^{CPMG} = 41.1\,\mu s$6 years, or about 16,000 years. Even under a more-than-762.2-PB memory assumption, the paper gives a lower-bound cost of $T_2^{CPMG} = 41.1\,\mu s$7 FLOPs and an estimated runtime of 19 years (Group, 11 Dec 2025).

Within the paper’s argument, the quantum-advantage claim rests precisely on that contrast: 18.4 minutes on the cloud-accessible superconducting platform versus years to millennia for tensor-network simulation under the stated classical assumptions (Group, 11 Dec 2025).

5. Users, applications, and service semantics

The conclusion explicitly invites quantum developers, researchers, and enterprises. The software discussion also suggests algorithm developers, experimentalists, application researchers, and industrial users testing workflows in optimization, simulation, or machine learning. Historically, the platform is said already to have supported work in quantum simulation, optimization, and machine learning (Group, 11 Dec 2025).

The immediate use case highlighted by the paper is large-scale RCS benchmarking rather than a practically useful application workload. The paper is explicit on this point: “RCS tasks based on iSWAP-like gates have limited practical application.” That caveat is central to the correct interpretation of the platform. The significance claimed is not that Tianyan already solves a commercially useful task, but that cloud delivery is being demonstrated for large-scale coherent control, stable calibration, high-fidelity operation, HPC-integrated orchestration, and workloads beyond practical classical simulation (Group, 11 Dec 2025).

The paper therefore portrays Tianyan as a platform for validation and exploration as much as for execution. The stated purpose is to enable the community to validate and explore practical quantum advantages, and the software design exposes this through Cqlib’s support for circuit construction, transpilation, simulation, execution, optimization, and result retrieval. A plausible implication is that Tianyan’s user-facing identity is closer to a research and development platform than to a narrowly defined benchmark appliance (Group, 11 Dec 2025).

In the broader literature on quantum cloud computing, this positioning is consistent with a shift from simple remote QPU access toward more elaborate platform models that combine hardware, simulators, workflow tooling, heterogeneous scheduling, and classical co-processing. Reviews of the area emphasize precisely these dimensions—resource management, backend selection, queueing, security, hybrid orchestration, and software interoperability—as the criteria by which serious cloud quantum systems are increasingly assessed (Nguyen et al., 2024). Tianyan’s description fits that pattern in a specifically superconducting, quantum-advantage-oriented form (Group, 11 Dec 2025).

6. Caveats, limitations, and historical significance

Several caveats are explicit in the paper and materially qualify its interpretation. First, the showcased benchmark uses a 74-qubit subregion of a processor whose physical design capacity is 105 qubits and 182 couplers, with one qubit reported non-functional. Second, the full-circuit fidelity is not directly obtained by exact classical simulation of the flagship circuit, but estimated through patch-circuit validation and a discrete error model. Third, the circuit notation for the A/B/C/D coupling pattern is inconsistent between the main text and a figure caption. None of these points invalidates the reported result, but they define the exact evidentiary basis of the claim (Group, 11 Dec 2025).

The platform’s importance is therefore partly historical. Earlier cloud systems such as NMRCloudQ established a model of queue-based remote access to real quantum hardware, with an emphasis on direct experimental access, pulse-aware execution, and educational or diagnostic comparison between simulation and experiment (Xin et al., 2017). Tianyan belongs to a later phase in which cloud access is tied to substantially larger superconducting hardware, explicit quantum-advantage claims, and tightly coupled HPC participation in the cloud workflow (Group, 11 Dec 2025).

A broader operational perspective is also relevant. Work on cloud-accessible photonic systems has emphasized that a quantum cloud platform must ultimately be judged not only by backend physics but by availability, maintenance discipline, monitoring, orchestration, and HPC-hosting readiness (Maring et al., 2023). Tianyan’s paper contributes chiefly on performance, workflow, and benchmark scale, but its architecture—task scheduling, simulators, quantum hardware, cloud storage, and locally connected HPC—places it within the same full-service conception of quantum cloud infrastructure (Group, 11 Dec 2025).

Taken together, the available evidence portrays Tianyan Quantum Cloud Platform as a full-stack superconducting quantum cloud service whose distinctive claim is cloud access to Zuchongzhi 3.0-like computational power. Its defining technical features are the Tianyan-287 backend, the reported fidelities of 99.90% for single-qubit gates, 99.56% for two-qubit gates, and 98.7% for readout, the 74-qubit, 24-cycle RCS benchmark completed in 18.4 minutes, and the Cqlib-centered cloud workflow with integrated scheduling, simulation, and HPC-assisted optimization (Group, 11 Dec 2025).