Primates can use various objects as tools. How they use the objects varies with the species. This provides a measure of primates' intelligence. This study focuses on the following issues on the ability of cotton-top tamarin's tool-use behavior shown by Santos et al.; 1) Whether does tamarin use an inverse model or a forward model as the internal model on the tool-use behavior? 2) How many steps of tools' movement can tamarin predict using the internal model when multiple tools are used? We present computational models that are expected to reproduce tamarin's behavioral data. The models are implemented with an actor-critic model of reinforcement learning and the internal model. We conducted computer simulations in which five agents (virtual actors) execute the same tool-use tasks as the tamarins did. The only one agent that uses the inverse model and predicts one tool's movement replicated Santos' results. This result suggests that the tamarins used the inverse model and predicted one tool's movement in the experiments by Santos et al.