Quantum X Labs reported progress on its AI-driven quantum error-correction program using NVIDIA accelerated computing and CUDA-Q QEC software libraries. The company benchmarked its transformer-based QECCT decoder against the classical MWPM decoder in simulated toric-code and surface-code environments. QECCT outperformed MWPM in selected simulated regimes. QXL also tested synthetic surface-code configurations modeled on Google’s public surface-code geometry and experiment structure, spanning multiple code distances. Across these scenarios, the QECCT decoder showed stable logical and bit error rates under varying physical error conditions. Future work is expected to focus on extending these evaluations to publicly available experimental datasets and continuing to refine data pipelines and decoder workflows compatible with CUDA-Q QEC frameworks. The broader roadmap includes QXL’s planned work with IQCC, a Quantum Machines company, to generate hardware-derived syndrome data on superconducting quantum processing hardware. QXL is also reviewing where AI-based pre-decoder workflows using NVIDIA Ising and low-latency optimization can add the greatest value.