QUBIKOS

QUBIKOS (QUantum Benchmark wIth Known-Optimal SWAP Counts) is the successor to QUEKO from the same UCLA group (Ping, Lin, Tan & Cong), introduced in February 2025 and published at DAC 2025. It closes QUEKO’s blind spot: QUEKO circuits need zero SWAPs by construction, so they never exercise a compiler’s routing, whereas QUBIKOS circuits are built so that their minimal SWAP count on a target device is nonzero and provably known. Like its predecessor, it benchmarks classical compiler software, not quantum processors.

How it works

The generator constructs circuits whose optimal SWAP insertion on a given coupling graph can be proven, then scores a layout-synthesis tool by the ratio of SWAPs it actually inserts to that known optimum, where 1x is optimal. The released set contains 40 circuits per architecture (ten each at optimal SWAP counts of 5, 10, 15, and 20) for the Rigetti Aspen-4 (16-qubit), IBM Rochester (53), Google Sycamore (54), and IBM Eagle (127) coupling graphs.

Strengths and limitations

QUBIKOS is the first suite to certify routing (SWAP-count) optimality rather than just mapped depth, and its initial results are sobering: average optimality gaps of roughly 63x for LightSabre, 117x for ML-QLS, 250x for QMAP, and 330x for t|ket⟩. Those gaps swing strongly with architecture (LightSabre ranged from about 2x on Sycamore to about 234x on Eagle), and the circuits are deliberately adversarial, so headline gaps need not transfer to typical application workloads. The suite is young: the GitHub repository is minimal and specifies no license, so the arXiv paper is the authoritative reference.

Notable results

QUBIKOS is already being adopted alongside QUEKO and QASMBench in 2025–2026 layout-synthesis evaluations (e.g. the multilevel framework of arXiv:2505.24169). The independent QKNOB suite covers the related regime of provably near-optimal, rather than exactly optimal, transformation costs.