Machine unlearning is a service offered to customers to withdraw their privacy from trained models, but its value is yet to be thoroughly evaluated. In addition, the free unlearning service is insufficient due to unaffordable computational cost and degradation on model utility on the part of the service provider. To address these problems, we design OPMUS, a pricing model for machine unlearning service with the Stackelberg game, where both customers’ benefit and the service provider’s profit are taken into account. The former is considered in terms of the satisfaction of the service and the payment, and the latter in terms of the cost of unlearning service, the degradation on model utility, and the reward and income from providing service. The result of the game model proposes a price on which both customers and the service provider agreed, and the corresponding number of unlearning requests that can be responded with the optimal price is obtained. The theoretical analysis confirms the existence of the optimal result. The numerical simulation suggests that the proposed model is effective in finding optimal results, which offers unlearning services at a reasonable price, and helps handle the requests that are difficult to respond for free. The influence of the parameters on optimal results is evaluated.