A new arXiv paper presents MindView, a benchmark designed to assess generative multi-view spatial intelligence. The authors note that spatial intelligence has recently been evaluated through generative tasks such as spatially grounded image editing, but they identify a gap: language-guided generation of spatially consistent views from partial multi-view observations remains underexplored.
MindView appears to target that gap directly, providing a framework for testing models on tasks that require both language understanding and spatial reasoning. The benchmark is positioned as a step beyond existing evaluation methods, which have focused on editing single images rather than generating multiple coherent views from incomplete input.
Because the abstract is truncated, details about the benchmark's construction, evaluation metrics, or baseline results are not yet available. The announcement only outlines the motivation and scope, leaving the full methodology to the paper itself.