A new multimodal artificial intelligence model, NEVA, improves neuroblastoma diagnosis and prognostic stratification by predicting histological and molecular features from routine pathology data, according to a multi-institutional study involving 1,238 patients. Researchers developed NEVA using a pathologist-inspired hierarchical workflow with end-to-end optimisation to better replicate clinical decision-making. Compared with 10 representative foundation models across 11 clinical tasks, NEVA outperformed comparators on most tasks. The model achieved area under the receiver operating characteristic curve values of 0.916 for subtype classification, 0.823 for Shimada classification, and 0.806 for risk group stratification. For molecular alterations, predictive performance reached 0.924 for NMYC amplification and 0.830 for 1p36 deletion. NEVA also enabled prognostic stratification for progression free survival and overall survival across multiple test cohorts and incorporated interpretable attention maps to localise histologically relevant regions. The findings establish a scalable framework for neuroblastoma risk stratification and clinical decision support using routinely available pathology data.