The vacuum degree within the arcing chamber of a vacuum circuit breaker directly impacts its breaking performance and is pivotal for the operational safety of power systems. The lack of a reliable online vacuum degree detection method currently constrains the wider adoption of vacuum circuit breakers at higher voltage levels. This study proposes an online vacuum degree detection method integrating fiber optic LIBS and machine learning algorithms. By constructing a FO-LIBS system and vacuum chamber setup, elemental spectra were collected across the range of 10–4 Pa to 105 Pa. The original spectra were divided into training and testing sets at an 8:2 ratios, and four machine learning algorithms—KNN, SVM, Random Forest, and CNN—were employed to establish vacuum degree classification models. The hyperparameters of each algorithm were fine-tuned to ensure optimal performance. Experimental results demonstrate that this method achieves a detection limit of 10–4 Pa and a maximum accuracy rate of 94%. The flexible structure of fiber optics renders it suitable for on-site applications. Thus, this method represents a reliable approach for online vacuum degree detection in vacuum circuit breakers, holding significant promise for practical applications.