Imagine stepping into a bustling fantasy marketplace where the merchants aren't just automatons with fixed prices. Instead, they haggle, they barter, they try to get the best deal – just like in the real world. This engaging scenario could soon become a reality thanks to the power of large language models (LLMs). Researchers are exploring how LLMs can transform the typically static merchant NPCs in games into dynamic, interactive characters capable of appraising items and negotiating prices. This new framework, known as MART, uses two key modules: an appraiser and a negotiator. The appraiser analyzes item descriptions to estimate a fair price, moving beyond pre-defined values. The negotiator then engages in a back-and-forth with the player, employing various persuasion tactics to strike a deal. Experiments with different LLM sizes and training methods show promising results. Fine-tuning smaller models through techniques like knowledge distillation can create surprisingly effective negotiators, balancing performance and efficiency. However, there are some quirks to iron out. LLMs sometimes offer 'giveaways' not typically found in games, or even 'hallucinate' non-existent items. Smaller models can also struggle with basic arithmetic, occasionally miscalculating prices. Despite these challenges, LLMs offer a compelling path toward more immersive and engaging in-game economies. Future research could explore how different LLMs perform in this context and how human players adapt their tactics when faced with a haggling AI.