For years, quantum physicists have dreamed of running a Bell test on a remote quantum computer without ever touching the hardware. The challenge: photonic processors are exquisitely sensitive to loss, alignment, and the order of measurements. Until now, no external user had coaxed a cloud-accessible photonic chip into violating the CHSH inequality—the canonical benchmark of quantum correlations—using encoded logical qubits. In August 2026, researchers accessed Quandela’s commercial Belenos photonic processor via its public cloud interface and ran a Bell test that had never been attempted remotely with this encoding. Photonic qubits encoded in spatial modes suffer from mode mismatch and photon loss; the required postselected controlled-Z gate succeeds only one-ninth of the time. On top of that, the cloud platform fixed the measurement order and hid the compiled circuit mapping, leaving the team with an operational reference rather than a loophole-free entanglement witness. [arXiv:2608.18153]
The Core Finding
The team encoded two logical qubits in the zero-sum parity subspace of eight spatial modes—a constraint that acts as a built-in error-detection code. They then applied a postselected linear-optical controlled-Z gate coupling 16 of 24 modes and measured CHSH correlators in four fixed settings. Over eight sequential same-day passes, every raw score exceeded the classical bound of 2. The session means were 2.40 and 2.58, with standard deviations of 0.15 and 0.03. When all counts were pooled, the CHSH score reached Scount = 2.485 ± 0.019. A reweighted model scenario, which accounts for efficiency imbalances, gave Srwcount = 2.380 ± 0.021. Think of it like trying to prove a coin is biased by flipping it on a wobbly table you can’t see, using a remote-controlled arm that only moves in a preset sequence. Despite the constraints, the statistical deviation from fair-coin behavior is unmistakable.
“The primary quantity is the operational CHSH score S on accepted logical coincidences, with each complete four-setting pass as the experimental unit.”
The count-pooled score of 2.485 exceeds the classical limit by more than 25 standard deviations, a clear signal of quantum correlations emerging from a parity-constrained logical encoding on a real-world photonic chip.
The State of the Field
Prior cloud-based Bell tests have been performed on superconducting and trapped-ion processors, but photonic platforms introduce unique hurdles. In 2022, a team led by Jianwei Wang demonstrated a CHSH violation on a programmable silicon photonic chip in a lab, but that experiment used direct access and active switching. The present work is the first to achieve a CHSH score above 2 on a commercial cloud photonic processor with spatial-mode encoding, where the user has no control over the physical implementation. The broader quantum computing landscape is racing toward fault tolerance, and photonics offers advantages in room-temperature operation and networking. Cloud access to such processors is still nascent; Quandela’s Belenos is one of only a handful of publicly available photonic quantum computers. This result shows that even with limited control, external users can extract meaningful quantum correlation benchmarks that probe the quality of encoded logical qubits.
From Lab to Reality
For scientists, the protocol provides a standardized reference acquisition method for characterizing remote photonic processors. By fixing the measurement order and using postselection, researchers can gauge the intrinsic quality of the encoded logical qubits without needing full device transparency. For engineers, the parity-constrained encoding is a primitive for error detection; integrating it with active feed-forward could lead to error-corrected photonic qubits. The quantum error correction market, projected to reach $1.8 billion by 2030 according to Grand View Research, stands to benefit from any platform that demonstrates robust logical qubit operations. For investors, the demonstration validates the commercial viability of cloud-accessible photonic quantum computing, a segment of the quantum cloud services market estimated at $2.2 billion by 2028 by MarketsandMarkets. While this experiment is a reference acquisition, not a full error-corrected logical qubit, it shows that parity-based encodings can survive the noise of a real-world photonic chip and deliver correlation scores that exceed classical limits.
What Still Needs to Happen
The most glaring limitation is that the CHSH score does not constitute a loophole-free Bell test. The measurement settings were applied in a fixed order, and the compiled circuit mapping was not returned, so the team could not verify that the implemented operations matched their intentions. Residual correlations in the remote-setting marginals further prevent an entanglement-witness claim. To close these loopholes, researchers would need fast, random basis switching and high-efficiency single-photon detectors—areas where groups like Anton Zeilinger’s at the University of Vienna and Jian-Wei Pan’s at USTC are making progress. A second challenge is the low success probability of the postselected gate. At 1/9, the coincidence rate is too low for scalable quantum computing. Improving photon sources and using multiplexing could raise the rate, as pursued by PsiQuantum and Xanadu. Finally, the lack of returned mapping means that independent verification of the circuit is impossible; cloud providers will need to offer transparency features for rigorous benchmarking.
In short: a cloud-accessible photonic processor produced a CHSH score of 2.485 using parity-encoded logical qubits, establishing a new operational reference for quantum error correction research on remote platforms.