Psychotic and psychotic-like experiences are thought to emerge from various patterns of disrupted belief updating. These include belief rigidity, overestimating the reliability of sensory information, and misjudging task volatility. Yet, these substrates have never been jointly addressed under one computational framework and it is not clear to what degree they reflect trait-like computational patterns. Here, we introduce a computational model that describes interindividual differences in how individuals update their beliefs about a volatile environment through noisy observations and use this model in a series of studies. In a test-retest study with healthy participants (N = 45, 4 sessions), we find that interclass correlations were moderate to high for session-level model parameters and excellent for averaged belief trajectories and learning rates estimated through hierarchical Bayesian inference. Across three studies (total N=590) we then demonstrate that two distinct computational patterns describe two different transdiagnostic categories. Higher uncertainty about the task volatility is related to schizotypal traits (N = 45, d = 0.687, P = 0.02, N = 437, d = 0.14, P = 0.032) and to positive symptoms in a sample of patients with schizophrenia (N = 108, d = 0.187, P = 0.039), when learning to gain rewards. In contrast, depressive-anxious traits were associated with more rigid beliefs about the underlying mean (N = 437, d = −0.125, P = 0.006) and outcome variance (N = 437, d = −0.111, P = 0.013), as were negative symptoms in patients with schizophrenia (d = - 0.223, P = 0.026, d = −0.298, P = 0.003), when learning to avoid losses. These findings suggest that individuals high on schizotypal traits across the psychosis continuum are less likely to learn or utilize higher-order statistical regularities of the environment and showcase the potential of clinically relevant computational phenotypes for differentiating symptom groups in a transdiagnostic manner.