Benchmarks

Measured on 2026-10-02 by the documentation build, with Qiskit 2.5.2 and tket 2.18.4, so these numbers always match the published compiler.

bench/compare.py compiles the same circuits with Qiskit (transpile at optimization_level=3), tket (FullPeepholeOptimise, then a rebase and squash) and qirc, all to rz, sx, x and cx. The circuits are the straight line programs in tests/corpus and examples with measurements removed, the QFT, Grover search for the all ones state, and the CDKM, VBE and Draper adders from Qiskit. Each cell is two qubit gates / total gates / depth, and the input column counts each Toffoli and SWAP as one gate. Every result up to 12 qubits is checked against the input by exact operator comparison, allowing for the qubit permutation each tool reports.

All to all connectivity

compile time: Qiskit 0.09 s, tket 4.85 s, qirc -O3 –gates rz-sx-cx –resynth 4 –relabel 0.16 s, qirc -O3 –gates rz-sx-cx –resynth 4 –relabel –cost cx 0.24 s

A line of qubits

Routed with --coupling line:n for qirc, CouplingMap.from_line for Qiskit and DefaultMappingPass for tket.

compile time: Qiskit 0.13 s, tket 4.83 s, qirc -O3 –gates rz-sx-cx –resynth 4 –relabel 0.40 s, qirc -O3 –gates rz-sx-cx –resynth 4 –relabel –cost cx 1.05 s

To run it yourself, install qiskit and pytket, build qirc in release mode and run python bench/compare.py, adding --line for the routed table.