This dataset provides neural-network ex-situ classification scores for Gaia DR3 stars with full 6D phase-space information. It is associated with the paper Li et al. (2024), "Exploring the ex-situ components within Gaia DR3", MNRAS, 527, 9767–9781. The catalogue contains Gaia DR3 source identifiers, Galactocentric phase-space coordinates, orbital/dynamical quantities, propagated uncertainties, and neural-network classifier outputs. The original model was applied in Li et al. (2024) to a Gaia DR3 target sample of 27,085,748 stars, from which 160,146 ex-situ candidates were identified using the adopted threshold. The catalogue includes two classifier-score columns: prediction and prediction_refined. The prediction column corresponds to the neural-network model used in the published analysis and is retained for reproducibility. The prediction_refined column is produced by an updated model retrained with a Gaia DR3-based mock catalogue and is recommended for new scientific applications. The classifier-score values are sigmoid neural-network outputs, not explicitly calibrated probabilities. They should be interpreted as ex-situ classification scores rather than ground-truth stellar-origin labels.