Tag
GroupMemBench is a new benchmark for evaluating LLM agent memory in multi-party conversations, exposing failures in current memory systems with the best achieving only 46% average accuracy.