The application of large language models (LLMs) to Pali written in the Thai script remains limited in accuracy due to its low-resource setting, complex grammar, and the absence of standardized evaluation benchmarks. In this paper, we introduce ThaiPaliBench, a comprehensive benchmark dataset designed to systematically evaluate LLM proficiency in Thai-script Pali. Our proposed dataset is an LLM-assisted benchmark derived from the Royal Pali Examination corpus and spans Parian Tham levels 1-9. ThaiPaliBench, comprises multiple-choice and sentence-pair tasks that evaluate grammatical knowledge, reading comprehension, natural language inference, and translation understanding. To verify the effectiveness of our benchmark, we evaluate over 10 model variants across 3 commercial and open-source LLM families, including ChatGPT, Claude, Kimi, Grok, Typhoon, GPT-OSS-120B, ThaiLLM, and SEA-LION. Our experiments show that even the best-performing model, Kimi K2.5 (1T), achieves only 68.3% overall accuracy, with performance dropping substantially on grammatical reasoning and semantic inference tasks. These results highlight the unique challenges posed by Thai-script Pali and confirm that ThaiPaliBench serves as a reliable, linguistically grounded benchmark for evaluating LLM proficiency in this low-resource language. Our proposed dataset is publicly available at: https://github.com/ainrg-llm-dev/ThaiPaliBench