CCTU: Tool Use under Complex Constraints (2026)

CCTU (Complex Constraints for Tool Use) is a benchmark designed to evaluate large language models' ability to use tools while adhering to explicit complex constraints. It introduces a taxonomy of 12 constraint categories across four dimensions: resource, behavior, toolset, and response. The benchmark comprises 200 carefully curated and challenging test cases, each involving an average of seven constraint types and an average prompt length exceeding 4,700 tokens. To enable reliable evaluation, CCTU provides an executable constraint validation module that performs step-level validation during multi-turn interactions between models and their environments. The paper evaluates nine state-of-the-art LLMs in both thinking and non-thinking modes and finds that no model achieves a task completion rate above 20% when strict adherence is required. Models violate constraints in over 50% of cases, particularly in the resource and response dimensions, and demonstrate limited capacity for self-refinement even after receiving detailed feedback. CCTU is released as open-source data and code to facilitate further research.