Neuromodulatory systems play a crucial role in the brain, driving rapid, global changes in neuronal computation and learning. While the dopaminergic system is thought to broadcast a scalar reward prediction error, the serotonin system’s function has been difficult to elucidate, partly due to the heterogeneity of its individual neuronal responses [1]. Although serotonergic neurons are known to respond to various salient events [2] and rewards [3], which can be accounted for by ad-hoc extensions to a reinforcement learning framework, our study presents the first comprehensive, normative framework for understanding this system through the lens of unsupervised learning. In this work, we propose and test the idea that serotonin encodes a multidimensional state prediction error (SPE) which underlie the building of world models. We incorporated an attention mechanism to privilege encoding of relevant information in the limited channel capacity of the serotonin system. We formalized this theory in a computational model (‘A-SPE’) that combines unsupervised next-state prediction with reinforcement learning using a value-driven, attention-modulated SPE to guide learning. We tested the ability of this model to capture a wide range of global serotonin response properties by recording serotonin activity in the dorsal raphe nucleus (DRN) of mice learning to navigate a virtual corridor, where pairs of images on the wall are predictive of upcoming rewards. Serotonin transients were triggered by the presentation of novel visual cues, which rapidly adapted before increasing again during the learning process, becoming significantly stronger for unpredictable than for predictable cues and stronger for rewarded than unrewarded cues. The A-SPE model, without additional assumptions, captured the entire evolution of DRN serotonin activity during learning, steady-state behavior, and reversal of image-reward contingencies. This work suggests that serotonin may complement dopamine by orchestrating the unsupervised learning of world models through SPE, explaining the breadth and diversity of serotonin signals.