This paper introduces a parametric World Knowledge Model (WKM) to enhance agent planning by leveraging global task knowledge and local state knowledge. The authors address the limitations of large language models (LLMs) in understanding the real world, which often lead to brainless trial-and-error and hallucinatory actions. The WKM is trained using expert trajectories and sampled trajectories to synthesize task knowledge and state knowledge. The task knowledge guides global planning, while the state knowledge assists local planning. Experimental results on three complex real-world simulated datasets with three state-of-the-art LLMs (Mistral-7B, Gemma-7B, and Llama-3-8B) demonstrate superior performance compared to various strong baselines. The WKM effectively reduces blind trial-and-error and hallucinatory actions, generalizes well to unseen tasks, and shows potential in unified WKM training. The paper also explores the impact of explicit state knowledge and the weak-guide-strong paradigm, highlighting the importance of implicit knowledge constraints in agent planning.This paper introduces a parametric World Knowledge Model (WKM) to enhance agent planning by leveraging global task knowledge and local state knowledge. The authors address the limitations of large language models (LLMs) in understanding the real world, which often lead to brainless trial-and-error and hallucinatory actions. The WKM is trained using expert trajectories and sampled trajectories to synthesize task knowledge and state knowledge. The task knowledge guides global planning, while the state knowledge assists local planning. Experimental results on three complex real-world simulated datasets with three state-of-the-art LLMs (Mistral-7B, Gemma-7B, and Llama-3-8B) demonstrate superior performance compared to various strong baselines. The WKM effectively reduces blind trial-and-error and hallucinatory actions, generalizes well to unseen tasks, and shows potential in unified WKM training. The paper also explores the impact of explicit state knowledge and the weak-guide-strong paradigm, highlighting the importance of implicit knowledge constraints in agent planning.