Many computational predictors classify amino acid variants as either pathogenic or benign. PON‑P3 is an artificial intelligence‑based tool built on a large training dataset and 1,526 features describing variants, sequence and structural context and protein parameters. The gradient boosting model was benchmarked against 23 existing predictors and showed superior performance. PON‑P3 classifies variants as benign, pathogenic or of uncertain significance, predicts all possible substitutions in MANE proteins and assesses ClinVar variants to assign pathogenicity, benignity or uncertainty.