Vox Deorum: A Hybrid LLM Architecture for 4X / Grand Strategy Game AI – Lessons from Civilization V

Summary

This paper explores the integration of Large Language Models (LLMs) into complex 4X and grand strategy games, focusing on Sid Meier’s Civilization V with the Vox Populi mod. We introduce Vox Deorum, a hybrid LLM+X architecture that addresses the unique challenges of game complexity and computational limitations by separating strategic reasoning from tactical execution.

Through extensive testing with 2,327 complete game simulations, we demonstrate that LLMs can handle macro-strategic reasoning and achieve competitive end-to-end gameplay with distinct play styles. The layered technical design delegates high-level strategic decision-making to LLMs while specialized subsystems manage tactical execution, creating more natural human-AI gameplay interactions.

This research establishes a viable architecture for integrating LLMs in commercial strategy games and opens new opportunities for game design and agentic AI research. By comparing two open-source LLMs against enhanced game AI, we provide insights into how LLMs can transform traditional game AI into more contextually aware and engaging opponents.

Recommended citation: Chen, J., et al. (2025). Vox Deorum: A Hybrid LLM Architecture for 4X / Grand Strategy Game AI -- Lessons from Civilization V. arXiv preprint arXiv:2512.18564.
Download Paper