Background: Early,noninvasively and accurate differentiation of malignant from benign solid pulmonary nodules is essential to avoid both overtreatment and delayed intervention. We aimed to develop a clinically translatable multi‑omics model that integrates plasma ctDNA methylation, CT radiomics, and routine clinical data to improve preoperative diagnosis of indeterminate pure solid pulmonary nodules ≤3 cm. Methods: In a prospective two‑center study, we consecutively enrolled adults with indeterminate pure solid pulmonary nodules (≤3 cm) scheduled for surgical resection. Patients were allocated by enrollment time into a modeling cohort and an independent validation cohort (7:3). We built three single‑modality models (clinical, radiomics, methylation) and a fused multi‑omics model. We standardized features within each modality, removed low‑variance and low‑relevance features, and then applied Elastic Net and SHAP‑based ranking, followed by dimensionality reduction and element‑wise fusion to obtain 130 integrated features. A stacked machine‑learning classifier, using logistic regression and XGBoost as base learners and a support vector machine as the meta‑learner, generated calibrated malignancy probabilities. The primary performance metric was area under the receiver operating characteristic curve (AUC) in the independent validation cohort; secondary metrics included sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). We compared AUCs by DeLong test (two‑sided alpha = .05). Results: A total of 324 patients were included. In the validation cohort, the clinical model achieved an AUC of 0.72 (95% CI, 0.63-0.80; sensitivity, 71.9%; specificity, 73.9%; PPV, 87.2%; NPV, 51.5%). The radiomics model yielded an AUC of 0.80 (95% CI, 0.67-0.90; sensitivity, 78.9%; specificity, 73.9%; PPV, 88.2%; NPV, 58.6%). The methylation model performed better, with an AUC of 0.95 (95% CI, 0.91-1.00; sensitivity, 78.9%; specificity, 100.0%; PPV, 100.0%; NPV, 65.7%). The multi‑omics model showed the highest discrimination, with an AUC of 0.99 (95% CI, 0.98-1.00), sensitivity of 87.7%, specificity of 100.0%, PPV of 100.0%, and NPV of 76.7%. DeLong tests showed that the multi‑omics model significantly outperformed the clinical (p<0.05) and radiomics (p<0.01) models; the difference versus the methylation model was not statistically significant (p =0.076), indicating that ctDNA methylation provided the dominant diagnostic signal, which multimodal fusion further enhanced. Conclusions: In this prospective two‑center study, a multi‑omics machine‑learning model accurately distinguished malignant from benign pure solid pulmonary nodules ≤3 cm. The model achieved near‑perfect specificity and high sensitivity in an independent validation cohort, with ctDNA methylation contributing the largest single‑modality signal.