The cost of large language models is mainly driven by their size and complexity, which require considerable computational power, storage, and resources for both training and deployment. As a result, building, maintaining, and utilizing these models can incur significant expenses, sometimes reaching several dollars per conversation or thousands of dollars per month.