Researchers used GeoGuessr world champion Radu Casapu, a master's student in Georgia Tech’s School of City and Regional Planning, to test vision-language models (VLMs) on geolocation accuracy. Casapu, who won the third annual GeoGuessr World Championship in September, collaborated with School of Interactive Computing professor James Hays and associate professors Alan Ritter and Wei Xu after Hays reached out at the beginning of the spring 2025 semester to create a dataset evaluating VLMs' geolocation ability and reasoning. Hays noted that "VLMs are surprisingly good at geolocation right out of the box, even when they’re not trained to be good at it." Ritter stated that when OpenAI released GPT 4 Vision, the company claimed the model "wasn’t good at geolocation beyond being able to recognize a city or famous monument," but the researchers "found that wasn’t the case." The team, including Casapu, Joshua Diao, and Tejas Santanam, each received 500 images to geolocate, recording their reasoning. The result was GeoRC, the first benchmark for VLM geolocation performance, consisting of 800 "ground truth" reasoning chains. Hays and Ritter gave the same images to GPT 5, Gemini, Llama, and Qwen.