YANAMI Multimodal Dataset for Stateful Interaction
November 2, 2026
YANAMI introduces a large-scale dataset and benchmark comprising 400K multi-turn multimodal dialogues designed for stateful interaction. It focuses on a model's ability to maintain and update evolving dialogue states across turns rather than solving independent single-turn tasks.
HOW THIS AFFECTS YOU
●
builderThis provides a better way to benchmark agents requiring long-term state memory.
●
researcherUse this dataset to evaluate how models track information across long multimodal dialogues.