The Signal
Quantum X Reports 12.7% Fewer Errors on Google Data
Quantum X Labs says its decoder produced fewer logical errors than PyMatching on Google’s public surface-code data in both a five-round test and a separate streaming evaluation. The company says it plans to test the decoder in a real-time p
Quantum.world Team · Written with AI · fact-checked ·
12.7% fewer errors at distance 3
Quantum X Labs says its decoder produced 12.7% fewer logical errors than PyMatching on Google’s public surface-code data at code distance 3. At distance 5, the reported reduction was 12.1%. The company announced the results on October 21, 2026, saying its quantum error-correction decoder is deployed and running on NVIDIA’s CUDA-Q platform, which supports hybrid quantum-classical computing.
The comparison used 10,000 held-out runs, each covering five error-correction rounds, at the two code distances. Quantum X says it ran each decoder through the same CUDA-Q QEC interface on the same data, making the comparison like-for-like. The release describes company-reported results and does not report an independent replication.
BP+OSD remained close
PyMatching was not the only comparison. Quantum X described its decoder as on par with BP+OSD, reporting about 2% fewer errors at distance 3 and 1.5% fewer at distance 5. In simulations, it beat standard matching in all 108 tested combinations of code size and noise level.
A decoder processes error-correction measurements, or syndromes, and works out corrections as a quantum computation proceeds. Improved decoding may enable more reliable logical operation on the same physical hardware, the company said. These results came from Google’s publicly released surface-code dataset, recorded on a superconducting quantum processor. They do not establish performance across other devices, code configurations or operating conditions.
Quantum X says the CUDA-Q QEC integration is intended to let quantum hardware teams building on NVIDIA’s quantum platform select its decoder alongside existing CUDA-Q decoders without setting up a separate workflow. The company presents the integration as a step from decoder research and offline benchmarking toward accelerated, real-time error correction. That is a stated direction, not a claim that real-time operation alongside quantum hardware has already been demonstrated.
Streaming results used a different test
Quantum X also tested a streaming version that receives error data round by round as a quantum computer runs and returns corrections continuously. In a separate evaluation spanning 25 rounds on held-out shots from the same Google dataset, the company reported observed logical error rates about 6% below PyMatching at distance 3 and 9% below at distance 5.
For that streaming comparison, QXL says its decoder also had lower observed error rates than CUDA-Q QEC’s sliding-window baseline and full-experiment PyMatching, while performing comparably to BP+OSD. Those streaming figures are distinct from the reductions in the five-round evaluation; the release describes separate tests and metrics.
“This milestone demonstrates that QXL’s decoder technology can operate within a leading quantum computing software environment and deliver encouraging results on real experimental data,” said Prof. Nir Sharon, Chief Scientist of Quantum X Labs.
Live hardware operation is still a planned test
The company’s next stated step is to run the decoder in a real-time error-correction pipeline on a system using NVIDIA NVQLink, with the goal of moving it closer to live operation alongside quantum hardware. Quantum X also says it is working with quantum-control-system providers on potential integration directly into their platforms. Those are plans, not evidence that the decoder already operates live alongside a quantum processor; the reported tests used Google’s released dataset.
The decoder tested with CUDA-Q QEC is one implementation within Quantum X’s broader adaptive quantum error-correction platform. The company says its roadmap is designed to support multiple decoder architectures, error-correcting codes and quantum hardware technologies, with the aim of adapting decoding solutions to different systems. The release does not report benchmark results for those other implementations or hardware platforms.