Amap ABot-Earth 0.7: How Its 3D-Native City Model Works

Announced on September 10, 2026, ABot-Earth 0.7 is Amap’s model for generating interactive 3D urban scenes. Amap calls it the world’s first 3D-native urban world model—a priority claim that has not been independently established in the provided sources. 17

Generation instead of reconstruction

Conventional digital-earth workflows often build views from captured satellite imagery, aerial photography and point-cloud reconstructions. What users can see depends in part on the material collected and how it is assembled. Amap’s proposed difference is to train on spatial and temporal data, then generate a city scene end to end in 3D Gaussian Splatting (3DGS) format from a satellite image or text description. The model is intended to fill out a continuous space that users can enter and explore, rather than limit them to predetermined viewing routes. 735

That distinction matters for accuracy. A generated street or building detail may look plausible without being a verified record of that location. Calling the result a digital twin should not, by itself, be taken as evidence that every generated feature matches the physical world. 1735

What Amap claims about speed, reach and detail

Amap says ABot-Earth 0.7 can generate a kilometre-scale 3D city scene in about 10 minutes on a single consumer-grade GPU. It describes the process as 1,000 times more efficient than traditional methods, but the supplied reports do not establish a shared benchmark or specify the comparison conditions. 17

The company also claims coverage of more than 196 countries and regions and says one model can generate views spanning planetary, city and street-landmark scales. Its stated goal is to retain convincing roads, building forms and visual detail as a user moves closer to a scene. Geographic coverage, however, does not mean equally complete or accurate street-level results everywhere. 717

Where it fits in Amap’s app

ABot-Earth 0.7 powers Flying Street View 2.0 in Amap Street Stars. Amap presents the feature as a way to explore a destination in 3D before a trip—for example, to preview a venue, shopping centre or scenic area. It places that preview between Street Stars’ destination recommendations and assistance during the visit, as part of a broader spatial-intelligence experience. 17

What independent testing still needs to show

The published claims leave practical questions open: which GPU and settings produce the 10-minute result; what methods underpin the 1,000× comparison; how responsive the scenes are during exploration; and how well visual detail holds up across scales and locations. Most importantly for a mapping product, testing should separate a convincing generated view from an accurate representation of a real place, particularly when the input is only text or a single satellite image. 71735