We propose Multimodal Contrastive Unsupervised RL (M2CURL), a method that learns efficient representations from visuotactile data, improving RL convergence speed and performance. M2CURL is versatile, compatible with various RL algorithms, and enhances learning efficiency in manipulation tasks in the Tactile Gym 2 simulator.