A wide range of RNA modifications is increasingly recognised for crucial roles in RNA function and pathology. While substantial work has been done to study individual RNA modifications, their combined impact and interplay within individual RNA molecules remain elusive. To address this, we developed SWARM, a method which leverages nanopore direct RNA sequencing signals for detecting N6-methyladenosine, pseudouridine, and 5-methylcytosine, providing per-base readouts of multiple modifications within each long read. Through comprehensive benchmarking, SWARM demonstrates high accuracy in identifying modification sites and estimating modification rates, showing strong concordance with orthogonal validation datasets, and outperforming previous nanopore-based approaches. Critically, SWARM is trained with a focus on accurate separation between different RNA modifications in the same molecule, which is overlooked by previous methods and is crucial for avoiding false positives stemming from the effect of nearby modifications in the nanopore signal. Leveraging this new capability, we applied our models across transcriptomes from numerous species and tissues, uncovering conserved patterns of modification co-occurrence within individual RNA molecules. Our flexible workflow is adaptable to detect other RNA modifications with sufficient training data and supports both SQK-RNA002 and SQK-RNA004 nanopore sequencing chemistries, enabling broad application to both current and legacy datasets. SWARM expands the ability to explore RNA modifications in diverse biological contexts, paving the way for deeper insights into the complexity and function of the epitranscriptome.