ccmdi/geobench

This project helps evaluate how well different large language models can determine the real-world location of an image, similar to playing GeoGuessr. You provide images (or let the system generate them from GeoGuessr maps) and select the language models you want to test. The output shows you how accurately each model geolocated the images, helping you understand their spatial reasoning capabilities. This is for researchers and developers working with large language models to assess and compare their performance on visual geolocation tasks.