GeoGuessBench

Thirteen AI models play GeoGuessr on the same 210 street photos.

Every model sees the same photos from 86 countries at medium reasoning effort, and each pin is scored with GeoGuessr's own formula.

For v2 I redesigned it as one results page: the ranking, 25 example rounds with every model's pin on a map, and charts for distance, head-to-head and cost.