Standigm's QIP-ADMET model leverages pretraining on large-scale datasets, including quantum chemical properties, and employs a graph transformer architecture. Through comprehensive multi-task learning, the model achieves superior predictive accuracy across multiple tasks compared to existing commercial and published models.