RESCUE-Bench Evaluates Multi-Party Emotional Support in LLMs
September 8, 2026
RESCUE is a new benchmark consisting of 7,079 annotated turns from real family and couple interviews to evaluate relation-aware emotional support. It tests whether LLMs can capture evolving interpersonal dynamics rather than just individual emotional states.
HOW THIS AFFECTS YOU
●
researcherThis provides a dataset for studying complex, multi-agent socio-emotional interactions in conversational AI.