The field of large language and vision-language models is rapidly advancing, with a focus on improving spatial reasoning and real-world applications. Recent developments have demonstrated the potential of these models to democratize access to real estate insights, evaluate complex spatial constraints, and enable human-centered AI agents for neighborhood assessments. However, challenges persist, including limited spatial intelligence, overconfidence in certain tasks, and the need for more robust and interpretable reasoning capabilities. To address these challenges, researchers are introducing novel benchmarks, such as LocationReasoner and SIRI-Bench, to evaluate the spatial reasoning abilities of large language and vision-language models. Additionally, new methods, such as GLOBE and GRPO, are being proposed to enhance the locatability assessment and visual clue reasoning of these models. Noteworthy papers in this area include:

• On the Performance of LLMs for Real Estate Appraisal, which highlights the potential of LLMs to improve transparency in real estate appraisal.
• LocationReasoner: Evaluating LLMs on Real-World Site Selection Reasoning, which introduces a benchmark to evaluate LLMs' reasoning abilities in real-world site selection tasks.
• Recognition through Reasoning: Reinforcing Image Geo-localization with Large Vision-Language Models, which proposes a novel pipeline to construct a reasoning-oriented geo-localization dataset and introduces GLOBE for locatability assessment and visual clue reasoning.