Q-CTRL reported a 100-qubit Quantum Fourier Transform (QFT) experiment on IBM Quantum hardware on August 10, 2026. The team used a convolutional compilation method on a 156-qubit IBM Heron r3 processor and identified the target frequency in every test circuit up to 100 qubits.
The result sets a new width benchmark for an experimental QFT, according to Q-CTRL and its technical manuscript. The result also needs careful wording. A 100-qubit QFT subroutine is not a 100-qubit run of Shor's algorithm, and the experiment did not factor a large number.
What the experiment measured
The team prepared periodic signals with target frequencies encoded across registers from 5 to 100 qubits. After applying the QFT and measuring the register, the correct target bitstring appeared as the most frequent result in every tested width.
At 50 qubits, the target appeared 8.4 times more often than the most frequent wrong output in raw data. The reported unitary process fidelity was 11.4 percent. At 80 qubits, the target led by a factor of 7.5, while process fidelity fell to 1.8 percent.
At 100 qubits, the target still led the measurement distribution. Q-CTRL did not report a comparable process-fidelity figure for 100 qubits. The 100-qubit claim therefore rests on target selectivity rather than high overall circuit fidelity.
How convolutional compilation reduces circuit cost
A standard QFT creates many controlled rotations between distant qubits. Hardware with limited connectivity must add routing operations, increasing two-qubit gate count and exposure to noise.
Q-CTRL's Convolutional QFT moves a compact circuit kernel along a linear chain. The construction uses one ancilla qubit and keeps each qubit's causal history shorter. The hardware implementation uses n² - n + 2 two-qubit gates for an n-qubit register, close to the ideal all-to-all gate count.
The team also selected a device-aware layout, scheduled gates to keep operations local in time, inserted dynamical decoupling during idle periods, and applied statistical readout-error mitigation. These choices matter because gate count alone does not predict performance on a real processor.
Why this matters for quantum algorithms
The Quantum Fourier Transform supports phase estimation and quantum factoring. It also appears in several algorithms for periodicity and spectral analysis. Running a meaningful QFT subroutine across 100 physical qubits gives compiler designers a larger test case for hardware-aware circuit construction.
The result does not show a cryptographic threat. Shor's algorithm needs a complete factoring circuit, far higher fidelity, error correction, and a much larger fault-tolerant machine. A noisy QFT benchmark measures progress toward those requirements, not completion of them.
The result also differs from the Mitsui and Mitsubishi Electric Helios benchmark. The Helios study compared physical and error-detected QFT circuits on Quantinuum hardware. Q-CTRL focused on compilation and error suppression for a wider pre-fault-tolerant circuit.
Scope and next tests
The experiment used an approximate QFT. The manuscript drops small-angle entangling rotations below a chosen threshold, reducing circuit cost while adding synthesis error. Q-CTRL also used readout mitigation when estimating process fidelity.
Those choices fit a hardware benchmark, but they set the result's scope. A full QFT at 100 qubits would require more entangling gates and face a harder noise problem. Future work needs full-transform tests, deeper algorithms built from QFT blocks, and independent runs across different processor calibrations.
For now, the result shows a useful engineering pattern: circuit structure, device topology, scheduling, and error suppression work together. Qubit count alone did not produce the benchmark.