Weylattice · measured compilation
100-qubit QAOA compilation for IBM Fez
Measured 9 November 2026 using the live configuration of IBM Fez, a 156-qubit processor. The benchmark compares Weylattice with standard Qiskit using the same source circuit, target snapshot and optimisation level.
Compilation only. No physical QPU execution or solver run was submitted.
Download circuits, results and reproduction scriptThe workload
- 100 logical qubits and a connected, unweighted, 3-regular MaxCut graph with 150 edges.
- Five QAOA layers: 750 RZZ rotations, 500 RX rotations and 100 Hadamard gates.
- Graph seed 42, generated independently of the hardware connectivity and before observing compiler results.
- Fixed, non-optimised angles. This is a compilation benchmark, not a solution to the MaxCut instance.
- Spinorio SDK 1.1.1,
client.qcompile, optimisation level 3.
Complete measured results
| Metric | Standard Qiskit | Weylattice | Change |
|---|---|---|---|
| Native two-qubit gates | 7,245 | 7,013 | -3.20% |
| Two-qubit depth | 711 | 586 | -17.58% |
| Total circuit depth | 1,376 | 1,662 | +20.78% |
| Total gates | 20,537 | 29,225 | +42.30% |
| Compilation time | 1.284 s | 155.350 s | +154.066 s |
Weylattice reduced native two-qubit gate count by 3.20% and two-qubit depth by 17.58%. Total circuit depth increased by 20.78%, and total gate count increased by 42.30%. Compilation took 155.350 seconds versus 1.284 seconds for standard Qiskit.
Compiled gate activity on the hardware topology
Method and validation
The standard baseline is linked to the Weylattice run and uses the same saved source circuit version, target fingerprint and optimisation level. Both downloaded QPY artifacts load successfully. Their independently recomputed gate counts and depths match the server metrics. Every output gate belongs to the target gate set, and every two-qubit operation uses a permitted directed connection.
The target gate set is cz, id, rx, rz, rzz, sx, x. Final two-qubit gate totals combine CZ and RZZ operations. The two-qubit-depth metric filters the circuit to two-qubit operations; it is not calibrated wall-clock execution time. SWAP count after native-gate conversion is not used as a routing-overhead measure.
The benchmark compares Weylattice with standard Qiskit transpilation using the same circuit, hardware target and optimisation level.
No formal circuit-equivalence proof, full 100-qubit state-vector simulation, fidelity measurement or QPU execution was performed. One run per compiler was measured, with no confidence intervals. Results shown are for this 100-qubit QAOA benchmark on IBM Fez.
Reproduce the workload
The download includes the exact graph, angles, source QPY, both compiled QPY outputs, measured metrics, hardware snapshot, checksums, package versions and a Python reproduction script. The script requests your token through a hidden prompt or SPINORIO_API_TOKEN. It submits compilation jobs only. The live target configuration can change between runs.