Can a MUD Evaluate LLMs? $99 Proof‑of‑Concept Demonstrated

Researchers built a simple Multi‑User Dungeon (MUD) as a testbed for LLM evaluation. The entire environment was assembled for roughly $99 in hardware and hosting. The MUD allows language models to interact with a text‑based game world. Evaluation focuses on the models' ability to understand commands and generate appropriate responses. Early results suggest the MUD can surface strengths and weaknesses of LLM behavior. The proof of concept was posted online for community review. It demonstrates that inexpensive setups can support meaningful AI benchmarking. Future work may expand the game scenarios or integrate additional metrics.