Financial inclusion requires equitable access to financial services, yet linguistic barriers hinder non-native Japanese speakers. This study examines using large language models (LLMs) to generate Easy Japanese for financial literacy education and proposes a reference-free evaluation framework. The study assesses the effectiveness of LLMs through manual, automated, and machine-based evaluations using Zero-shot and One-shot prompting. The findings demonstrate that LLM-generated Easy Japanese enhances accessibility while preserving accuracy. Additionally, SHAP analysis of linguistic features across Japanese, Chinese, and Korean speakers reveals that hiragana, kanji, and particle usage influence text difficulty. This study provides a scalable solution for financial education and contributes to advancing financial inclusion.