On September 1, 2026, World Labs introduced its new product – a spatial AI model named Atlas, designed to create three-dimensional scenes based on more than 100 images. This approach significantly reduces the degree of imagination and allows for high accuracy in reconstructing the environment. According to World Labs, the average error of Atlas in scene reconstruction was 25.3, which is a significant improvement compared to the 28.7-47.7 range of five other specialized models.
Professor Fei-Fei Li from Stanford University, the creator of ImageNet and one of the leading experts in AI, emphasized the importance of spatial intelligence, which reconstructs a volumetric picture of the environment. Software solutions based on such models can be successfully used not only in entertainment – for example, in games – but also in more serious scenarios, such as autonomous transport and robotics. For instance, on February 6, 2026, Waymo introduced its own world model based on Genie 3 technologies, which allows for the output of not only images but also lidar data.
Since its launch in November 2025, World Labs' first commercial product, Marble, has opened new horizons for developers by providing an accessible platform for creating world models. At the same time, the demand for such technologies is growing every year, and investors are confident in their future success – in February 2026, World Labs raised $1 billion. The contract signed with leading technology companies such as AMD and Nvidia only strengthens World Labs' position in the spatial AI market. Atlas, with its improved algorithms, is capable of changing the approach to AI training and actively expands the boundaries of human-technology interaction in the real world.