Grouping similar students together within schools (streaming) or classrooms (tracking) based on past literacy skills (reported by parents), family socioeconomic status (SES) or reading attitudes might affect their reading achievement. Our multilevel analysis of the reading tests of 208,057 fourth-grade students across 40 countries, and their parents’, teachers’, principals’, and their survey responses yielded the following results. Streaming was linked to higher reading achievement (consistent with customized instruction), but tracking was linked to lower reading achievement (consistent with more help opportunities). Students had higher reading achievement when classmates had stronger past literacy skills (suggesting sharing ideas) or extremely poor ones (help opportunities). Also, when classmates have higher family SES, students had higher reading achievement (suggesting sharing resources), with diminishing marginal returns. When classmates’ family SES differed more (more diversity), students with greater past literacy skills had higher reading achievement (Matthew effect). Lastly, when classmates had better reading attitudes, students with lower past literacy skills showed higher reading achievement (modeling, norms). When classmates’ reading attitudes differed more, students had higher reading achievement (contrasting cases), although extreme differences weakened this link (less homophily). These results suggest that streaming across schools and mixing of students within classrooms (by past achievement, family SES and reading attitudes) are linked to overall reading achievement.