<p>Efficient information sampling is crucial for human inference and decision-making even for young children. It is also closely associated with the core symptoms of autism spectrum disorder (ASD), since both the social interaction difficulties and repetitive behaviors suggest that autistic people may sample information from the environment distinctively. However, the specific ways in which autistic children sample information, especially when facing explicit costs and adapting to environmental changes, remain unclear. Thirty-two autistic and 41 IQ-matched neurotypical children aged five to eight participated in a computerized bead task, where children decided to gather samples sequentially from an unknown target to infer which of the two options was the target. Autistic children showed lower sampling efficiency under costly conditions compared to neurotypical peers, resulting from increased variability in sample numbers across trials, rather than solely systematic sampling bias. Computational models indicated that while both groups shared a similar decision process, autistic children’s sampling decisions were less influenced by dynamic changes and more by recently gathered evidence. This led to higher sampling variation and lowered the efficiency of autistic children. These findings offer valuable insights into the cognitive mechanisms underlying fundamental behaviors in autistic children.</p>