WorldMemArena is a large-scale multimodal memory benchmark designed to evaluate how well AI systems retain, update, and recall information across extended multi-session interactions grounded in real-world scenarios.
Split
Samples
Sessions
Turns
Images
Memory Points
QA Pairs
Agent / GUI
203
4,074
16,252
5,755
13,907
10,382
Agent / Embodied
220
3,075
12,087
6,437
12,186
11,788
Lifelong / Project
18
540
14,580
1,601
6,980… See the full description on the dataset page:
https://huggingface.co/datasets/LCZZZZ/WorldMemArena.
