OpenHalDet: A Unified Benchmark for Hallucination Detection across Diverse Generation Scenarios

In three linesOpenHalDet is a unified benchmark for hallucination detection in LLMs. It standardizes evaluation (prompt construction, generation, annotation, scoring) and supports three detector families: black-box (outputs only), gray-box (probability signals), white-box (internal signals). Open-source codebase released.

Summary generated by Claude — human-verified