Aims: Hypertension is a major risk factor for cardiovascular disease, and scalable, non-cuff-based approaches for its detection may offer substantial population health benefits. This study aimed to evaluate whether combining electrocardiogram (ECG) and photoplethysmogram (PPG) features improves hypertension detection compared with either modality alone in a large population-based British cohort. Methods: UK Biobank participants without previous cardiovascular events and with same-day ECG and PPG recordings were included. PPG recordings consisted of single-beat waveforms, while 15-second ECGs were acquired either in the sitting position (n=52,824; visit 1; hypertension prevalence 49%) or supine position (n=25,861; visit 2; prevalence 53%). TabPFN v2.5, a foundation model for tabular data, was used as the primary classification model. Model performance was evaluated using the AUROC on held-out test data using an 80:20 train:test split, with 95% confidence intervals estimated via 1,000-fold bootstrapping. Models incorporated demographic variables (age, sex, BMI, and smoking status), 13 ECG-derived features (including resting heart rate, ultra-short-term heart rate and repolarization variability, and signal amplitude), 43 PPG morphological features, and their combinations. Results: In visit 1, the TabPFN v2.5 model based on demographic variables achieved an AUC of 0.72 (95% CI 0.71–0.73). Adding PPG-derived features substantially improved performance to an AUC of 0.80 (95% CI 0.79–0.81), while further inclusion of ECG-derived features increased AUC to 0.82 (95% CI 0.81–0.83). In visit 2, similar performance was observed, with the full model achieving an AUC of 0.83 (95% CI 0.81–0.85). Models based on PPG morphological features consistently outperformed those based on ECG features alone. Conclusion: The combination of ECG- and PPG-derived features enables accurate, cuffless detection of hypertension when integrated into a tabular foundation model (TabPFNv2.5). As wearable devices increasingly capture both ECG and PPG signals, this approach may have important implications for scalable population-level hypertension screening.