SpatialBench: Benchmarking Multimodal Large Language Models for Spatial Cognition

Researchers introduce SpatialBench, a comprehensive benchmark for evaluating spatial cognition in multimodal large language models (MLLMs). The framework reveals that while MLLMs excel at perceptual grounding, they struggle with symbolic reasoning, causal inference, and planning compared to humans who demonstrate more goal-directed spatial abstraction.

• →SpatialBench introduces the first systematic framework for measuring hierarchical spatial cognition in MLLMs across five progressive complexity levels.
• →The benchmark covers 15 tasks with a unified evaluation metric to assess spatial reasoning abilities across heterogeneous tasks.
• →Extensive testing reveals MLLMs have strong perceptual capabilities but significant limitations in symbolic reasoning and planning.
• →Human performance shows selective, goal-directed abstraction while MLLMs tend to over-focus on surface details without spatial intent.
• →This research establishes foundational infrastructure for developing future spatially intelligent AI systems.