This study presents a comparative evaluation of Convolutional Neural Networks (CNNs), Liquid Neural Networks (LNNs), and Multi-Layer Perceptrons (MLPs) for fluid level estimation using Fiber Bragg Grating (FBG) sensor data. Experiments were conducted with water and oil in a controlled tank environment, generating approximately 8,300 samples comprising amplitude and wavelength measurements from six sensors per fiber. Each model underwent hyperparameter optimization and was assessed under noise-free conditions as well as with Gaussian and 1/f colored noise at varying signal-to-noise ratios (10, 30, 50 dB). Results indicate that while all models achieved high accuracy on clean data, CNNs and MLPs were highly sensitive to strong noise. In contrast, LNNs demonstrated superior robustness and stability, maintaining low RMSE and high R2 across all noise levels. These findings highlight the LNN’s suitability for real-world fluid sensing applications where measurements are subject to environmental variability.