The field of multimodal large language models (MLLMs) is rapidly advancing, with a focus on improving spatial understanding and reasoning capabilities. Recent developments have introduced new benchmarks and frameworks to evaluate the fine-grained spatial understanding and reasoning capabilities of MLLMs, such as RoadBench, EventBench, and SpatialBench. These benchmarks have revealed significant shortcomings in existing MLLMs' spatial understanding and reasoning capabilities, particularly in complex urban scenarios and event-based vision. To address these gaps, researchers are exploring new methodologies, including the development of trajectory-focused foundation models and spatio-temporal foundation models. Noteworthy papers in this area include RoadBench, which evaluates MLLMs' fine-grained spatial understanding and reasoning capabilities in urban scenarios, and SpatialBench, which provides a hierarchical spatial cognition framework to assess MLLMs' spatial reasoning abilities. Overall, the field is moving towards more comprehensive and nuanced evaluations of MLLMs' spatial understanding and reasoning capabilities, with a focus on developing more robust and generalizable models.