The Think-In-Games (TiG) framework was proposed by Tencent in August 2025, and the relevant research results were published in the paper "Think in Games: Learning to Reason in Games via Reinforcement Learning with Large Language Models".

The TiG framework enables large language models (LLMs) to develop procedural understanding by interacting directly with the game environment, while retaining their inherent reasoning and explanation capabilities. Specifically, TiG reformulates reinforcement learning-based decision making as a language modeling task: LLMs generate language-guided policies and iteratively optimize these policies through online reinforcement learning based on environmental feedback. This framework successfully bridges the gap between declarative and procedural knowledge, achieving competitive performance compared to traditional RL methods while significantly reducing data and computation requirements.