A new framework called LGMT has been developed to evaluate the reasoning reliability of Large Language Models (LLMs) by using logic-based testing methods. This approach reveals hidden defects in LLMs that traditional evaluations miss, suggesting a need for more robust assessment techniques in AI development.