Well-designed online educational games can improve students' math knowledge, skills and engagement; however, more research is needed to understand how to formatively assess components of students' mathematical understanding and learning as students solve problems in online educational games. In this study, we examined how 7th-grade students' (N = 760) in-game performance (i.e., strategy efficiency, mathematical validity of the first action, mathematical errors) and help-seeking behaviours (i.e., hint requests) correlated with their algebraic knowledge in an online mathematics game. The k-means cluster analysis identified four groups of students based on their in-game metrics, and some variabilities in their in-game performance were found. Although hint requests were available, only a few students showed a high percentage of hint requests during their gameplay. The regression analysis revealed that students' in-game performance and help-seeking behavioural patterns explained a significant amount of variance in students' algebraic knowledge above and beyond their prior knowledge. Our results suggest that how students interact with the online learning game is related to their later math performance, and qualitative aspects of student behaviours captured using log data provide meaningful information to students' algebraic understanding.