MemLens: Benchmarking Multimodal Long-Term Memory in Large Vision-Language Models

TL;DR

MemLens introduces a comprehensive benchmark for evaluating multimodal long-term memory in large vision-language models, highlighting the strengths and limitations of current approaches and suggesting hybrid solutions.

Contribution

We present MEMLENS, a new benchmark for multimodal memory in multi-session conversations, and evaluate 27 LVLMs and 7 memory-augmented agents on it.

Findings

Long-context LVLMs excel in short-term accuracy but decline with longer conversations.

Memory-augmented agents maintain length stability but lose visual fidelity over time.

Most systems struggle with multi-session reasoning, achieving below 30% accuracy.

Abstract

Memory is essential for large vision-language models (LVLMs) to handle long, multimodal interactions, with two method directions providing this capability: long-context LVLMs and memory-augmented agents. However, no existing benchmark conducts a systematic comparison of the two on questions that genuinely require multimodal evidence. To close this gap, we introduce MEMLENS, a comprehensive benchmark for memory in multimodal multi-session conversations, comprising 789 questions across five memory abilities (information extraction, multi-session reasoning, temporal reasoning, knowledge update, and answer refusal) at four standard context lengths (32K-256K tokens) under a cross-modal token-counting scheme. An image-ablation study confirms that solving MEMLENS requires visual evidence: removing evidence images drops two frontier LVLMs below 2% accuracy on the 80.4% of questions whose evidence includes images. Evaluating 27 LVLMs and 7 memory-augmented agents, we find that long-context LVLMs achieve high short-context accuracy through direct visual grounding but degrade as conversations grow, whereas memory agents are length-stable but lose visual fidelity under storage-time compression. Multi-session reasoning caps most systems below 30%, and neither approach alone solves the task. These results motivate hybrid architectures that combine long-context attention with structured multimodal retrieval. Our code is available at https://github.com/xrenaf/MEMLENS.