The expanding field of epitranscriptomics has been setting to rival the epigenome in the diversity of biological process involvement, prominently linking biochemical modifications of the RNA to development and disease onset. However, the identification of modifications in individual RNA molecules remains challenging. Here we describe CHEUI, a new computational approach to identify N6-methyladenosine (m6A) and 5-methylcytidine (m5C) using signals from Nanopore direct RNA sequencing reads at single-nucleotide and single-molecule resolution. CHEUI uses a two-stage neural network to accurately predict methylation in individual reads and transcriptomic sites in a single condition, as well as differential methylation between any two conditions. Using extensive benchmarking with Nanopore data derived from in vitro modified and non-modified transcripts as well as cells with or without methylation enzymes, CHEUI showed higher accuracy than other existing methods in the prediction of m6A and m5C sites and their stoichiometry levels, while maintaining a lower number of false positives. CHEUI’s ability to detect RNA modifications can be expanded to other modifications to unveil the full span of the epitranscriptome in normal and disease conditions.