Tag
MINTEval is a new benchmark for evaluating LLM agents and memory systems in continuously updated environments with frequent context changes. It shows that current systems perform poorly, with an average accuracy of 27.9% across representative systems.