CCR-Bench

CCR-Bench is designed to assess LLMs’ ability to follow complex instructions through a progressive and multi-dimensional lens. The construction of CCR-Bench follows a logical progression from simple to complex, and from foundational to application-level scenarios. It contains 174 test cases and comprises three core components: Complex Content-Format Constraints, Logical Workflow Control and Industrial Scenario Application. The goal is to evaluate the practical utility and robustness of LLMs under conditions that approximate real-world industrial deployments.

CCR-Bench 是一款专为评估大语言模型（Large Language Model，LLM）遵循复杂指令的能力而打造的基准数据集，其评估视角兼具渐进性与多维度性。该数据集的构建遵循从简单到复杂、从基础场景到应用级场景的逻辑递进脉络，共包含174个测试用例，由三大核心模块构成：复杂内容格式约束（Complex Content-Format Constraints）、逻辑工作流控制（Logical Workflow Control）以及工业场景应用（Industrial Scenario Application）。其核心设计目标为在贴近真实工业部署的环境中，评估大语言模型的实际应用效用与鲁棒性。