Understanding others’ mental states is essential for social interaction, yet it fundamentally involves inference under uncertainty. The brain must transform noisy perceptual input into socially meaningful interpretations. I will present a computational framework that models this process as evidence accumulation under uncertainty. We applied hierarchical drift diffusion modeling to decompose social judgments into mechanistic components across participants spanning schizophrenia-spectrum disorder, bipolar disorder, depression. We found that impairments were specific to evidence accumulation under ambiguity and were shared across diagnoses. Archetypal analysis revealed distinct social decision-making strategies expressed dimensionally across individuals. A profile characterized by efficient evidence accumulation was associated with lower symptoms and better functioning. This work illustrates how computational modeling can bridge social cognition, psychiatric heterogeneity, and functional outcomes within a unified framework.