This study examines the effect of individual differences on insight problem-solving performance. This study identified individual differences in learning properties estimated through a Q-learning model in a reinforcement learning framework and evaluated their effects on insight problem-solving in two tasks, that is, 8-coin and 9-dot problems, both of which are classified as a broad category of spatial insight problems. Although the learning properties of the two problems were different, the results revealed that learning transferred between them. That is, the performance in the insight task improved with the prior experience of another task. Moreover, it was found that loss-taking, as opposed to loss aversion, significantly impacted performance in both problems with prior experience. Loss-taking behavior is hypothesized to facilitate analogical transfer across both tasks and improve performance. Furthermore, this is one of the few studies to attempt to apply a computational approach to insight problem solving. This approach enables the identification of underlying learning parameters for insight problem solving.