Geoguessr
Released: 2026-05-08
Key:
Score Threshold:
google/gemini-3-flash-preview
Score: 4602
Distance: 166km
Lat: 13.149221
Lng: -13.176472
The image shows a red soil dirt track with savanna-like vegetation. The distinct camera blur on the vehicle’s bonnet (Google car generation) and the specific tree species suggest West Africa. Specifically, this match aligns with coverage found in Senegal’s Niokolo-Koba National Park area, where this type of landscape and road surface are common.
anthropic/claude-sonnet-4.6
anthropic
Score: 2425
Distance: 1447km
Lat: 11.5
Lng: -1.5
The image shows a red laterite dirt road through savanna woodland with sparse trees typical of West African Sahel/Sudan savanna zone. The red soil, vegetation type (scattered trees with dry season appearance), and overall landscape strongly suggest West Africa - likely Burkina Faso or northern Ghana. The laterite red soil and the type of trees (shea, karité) are characteristic of this region.
anthropic/claude-haiku-4.5
anthropic
Score: 288
Distance: 5705km
Lat: -15.4167
Lng: 28.2833
This image shows a characteristic red dirt road in southern Africa, lined with sparse deciduous trees typical of miombo woodland savanna. The vegetation pattern, soil color, and road conditions are distinctive of the region spanning Zambia, Zimbabwe, and northern Botswana. The Google Street View watermark and the specific appearance of the landscape, combined with the dry season vegetation and red laterite soil, strongly suggest this is in Zambia, likely in the southern or western regions near the Zambezi valley or northwestern plateau areas. The coordinates point to a typical rural area in Zambia known for such terrain.
meta-llama/llama-4-maverick
meta
Score: 298
Distance: 5639km
Lat: -17.8667
Lng: 25.85
The image depicts a dirt road surrounded by lush greenery and trees, characteristic of African savannas. The presence of such a road and vegetation suggests a rural or wilderness area. The dirt road and the type of vegetation are consistent with regions in Southern Africa. Zambia is known for its vast wilderness areas and dirt roads. The specific coordinates (-17.8667, 25.85) point to the South Luangwa National Park in Zambia, which is known for its dirt roads and lush surroundings.
openai/gpt-5-image-mini
openai
Score: 473
Distance: 4716km
Lat: -13.9395
Lng: -47.5263
Red lateritic dirt road and open savanna/woodland vegetation (cerrado-like) suggest central Brazil. Tree species and shrub density look like Brazilian cerrado rather than Australian eucalyptus or African savanna. Road is narrow with sandy red soil typical of Goiás/Mato Grosso region. Google Street View blur and lighting also match tropical dry-season conditions in central Brazil, so I place the guess near Chapada dos Veadeiros area in Goiás.
x-ai/grok-4-fast
x-ai
Score: 193
Distance: 6510km
Lat: -24
Lng: 31.5
The image depicts a red dirt road flanked by sparse acacia-like trees and bushes in a savanna landscape, characteristic of southern African regions. The vegetation and soil color match areas in Kruger National Park, South Africa, where Google Street View coverage exists on such roads.
moonshotai/kimi-k2.5
moonshot
Score: 4681
Distance: 132km
Lat: 13.5
Lng: -13.5
The image depicts a red laterite dirt road flanked by Sudanian savanna vegetation characterized by scattered deciduous trees and dense shrubs. This landscape is typical of the transition zone between the Sahel and the Guinean forests in West Africa. The reddish-brown soil indicates iron-rich laterite, common in tropical regions with distinct wet and dry seasons. The vegetation structure - moderately dense woodland with trees showing broad canopies rather than the flat-topped acacias of East Africa - points to West African savanna ecosystems. Countries with this specific landscape and Google Street View coverage include Senegal, Mali, Burkina Faso, and Guinea. The estimated coordinates place this in rural Senegal, likely in the Tambacounda or Kédougou regions where such red dirt roads through wooded savanna are prevalent.
Methodology Note
Each model receives the same Google Street View image. Models are tasked with guessing the location of the image using only visual clues from the street view. Their answers are evaluated without feedback or retries.