QEC Model Zoo: Democratizing AI-enhanced Quantum Error Correction

Quantum error correction (QEC) is essential for scalable quantum computing. AI-enhanced QEC is gaining increasing attention, with learned full decoders and pre-decoders being explored across code families. However, progress is difficult to assess: existing studies use different datasets, input representations, decoder roles, preprocessing pipelines, and evaluation protocols, making it unclear whether a reported gain comes from the model, the data interface, or the benchmark setup. We present the QEC model zoo, a data-centric benchmark and model-zoo framework for AI-enhanced QEC. The QEC model zoo curates real and simulated datasets for repetition- and surface-code decoding, supports both full decoders and pre-decoders, and records the metadata needed for reproducible comparison. The framework provides common task definitions, model-ready processing pipelines, and evaluation interfaces for heterogeneous QEC data. It further enables systematic comparison across neural, tree-based, and modern tabular models. Our goal is not to advocate a single decoder architecture, but to make AI-enhanced QEC easier to compare, reproduce, and extend.