MQT Bench

MQT Bench is a benchmark circuit library from the Chair for Design Automation at the Technical University of Munich, part of the Munich Quantum Toolkit (MQT). Introduced by Quetschlich, Burgholzer & Wille in 2022, it supplies the standard workloads on which quantum software (compilers, simulators, verifiers) is evaluated. Its organizing idea is that a benchmark suite for software tools must provide the same circuit at whichever abstraction level each tool actually consumes.

How it works

The library holds 70,000+ benchmark circuits spanning 2 to 130 qubits, each provided at four abstraction levels: algorithmic (the abstract circuit), target-independent (compiled to a generic gate set), target-dependent native gates (expressed in a specific device’s gate set), and target-dependent mapped (routed to a device’s connectivity). A high-level simulator takes algorithmic-level input, a mapper is judged at the native-gates level, and a verifier can compare any two levels of the same benchmark. Circuits are served through a Python package and a web interface with filter-and-download access, so tool papers can cite a reproducible workload set instead of hand-rolled circuits.

Strengths and limitations

The four-level structure distinguishes it from single-level circuit suites such as QASMBench: each stage of the design-automation stack can be isolated and tested on identical workloads (Quetschlich, Burgholzer & Wille 2022). The authors position RevLib, the reversible-circuit-era library, as its inadequate predecessor for quantum tooling.

The flip side: MQT Bench supplies workloads, not scores. It defines no timing harness, pass criterion, or headline number; timed harnesses such as Benchpress exist precisely to add that layer. And as a software-stack suite it says nothing directly about quantum hardware performance.

Notable results

The paper was published in Quantum 7, 1062 (2023), and the library remains actively maintained and extended as part of the Munich Quantum Toolkit.