This article discusses methods for balancing the cost and performance of GPT-3.5 and GPT-4 in game development. The author presents several key factors and recommendations, including selecting the appropriate model based on requirements, using GPT-4 for generating formatted outputs, choosing models based on generation speed, playing animations or game events during content generation, and using GPT-3.5 for handling infinite contexts. By making judicious model selections and adjusting parameters, it is possible to maintain high quality while reducing costs.