ChLogic: Evaluating Robustness of Logical Reasoning in Chinese Expressions

In three linesChLogic is an English-Chinese aligned benchmark evaluating the robustness of logical reasoning in LLMs. Built from formal logical templates, it contains 100 aligned propositions and 15 Chinese-specific phenomena. Experiments on Qwen3, Ministral, and GLM reveal a persistent English-Chinese performance gap, with back-translation producing mixed effects.

Summary generated by Claude — human-verified