VideoMemBench: Evaluating In-Context Memory in Video MLLMs
November 2, 2026
VideoMemBench is a new benchmark designed to measure in-context video memory in MLLMs through long, multi-turn dialogues. Unlike existing benchmarks, it requires models to retrieve and reuse information established across multiple turns of a video conversation.
HOW THIS AFFECTS YOU
●
researcherYou can use this benchmark to test whether your MLLM truly maintains long-term context in video dialogues.