Beat tracking is a crucial task in music signal processing, aiming to identify the beats and downbeats in a musical piece. Online beat tracking, however, presents significant challenges due to the absence of prior data and the necessity for real-time processing. Previous approaches often face a trade-off between performance and latency, limiting their applicability in real-time scenarios. In this paper, we propose a novel hybrid architecture combining the Chronos model for probabilistic time series forecasting with a Convolutional Recurrent Neural Network (CRNN) for initial beat detection. Our approach dynamically adjusts the reliance on Chronos and CRNN, optimizing performance as more historical data is gathered. We introduce a new evaluation metric, the Latency-Adjusted F1 Score (LAF1), which balances accuracy and latency. Experimental results demonstrate that our method outperforms existing online beat tracking techniques, achieving superior metrics across diverse musical genres.