Abstract Attributing motives to others is a crucial aspect of mentalizing, which is disturbed by prejudice and is also affected by common psychiatric disorders. Thus it is important to understand in depth the neuro-computational functions underpinning mentalizing and social reward. We hypothesized that people quickly infer whether other’s motives are likely beneficial or detrimental, then refine their judgment. Such ‘Classify-refine’, active inference models of mentalizing motives might improve on traditional models, and hence allow testing the hypothesis that serotonergic antidepressant drugs improve function partly by inducing more benign views of others. In a week-long, placebo vs. Citalopram study using an iterated dictator task, ‘Classify-refine’ models accounted for behaviour better than traditional models. Citalopram did not lead to more magnanimous attributions of motives, but we found evidence that it may help refine attributions about others’ motives through learning. With respect to social differences, model comparison clearly indicated that ethnicity-dependent, in-task biases played no role in attributing motives for the large majority of participants. This is a very encouraging result which further research should seek to replicate, and, if replicated, celebrate. Lower subjective socio-economic status was associated with lower attributions of harm intent to others. We discuss how classify-refine social cognition may be highly adaptive. Future research should examine the role of Serotonergic antidepressants in clinical studies over longer time spans. Significance Statement We developed computational models to study how, in social situations, first impressions count a lot, but people also gradually refine their views to do justice to others. In our study, such ‘classify-refine’ models clearly outperformed the ones based on simpler learning. Modeling analyses indicated that refining one’s views was facilitated by serotonergic antidepressants. We expected that those who perceived themselves as socially disadvantaged, and those interacting with people of color, would tend to attribute less benign motives to others. However in our low income, highly educated, young sample, we found evidence against both these biases. Further studies are needed to test how far these encouraging neuropharmacological and psychosocial findings apply to other populations.