{"id":23671,"date":"2026-09-03T00:01:05","date_gmt":"2026-09-03T00:01:05","guid":{"rendered":"https:\/\/scannn.com\/atlas-a-world-model-for-spatial-intelligence\/"},"modified":"2026-09-03T00:01:05","modified_gmt":"2026-09-03T00:01:05","slug":"atlas-a-world-model-for-spatial-intelligence","status":"publish","type":"post","link":"https:\/\/scannn.com\/lv\/atlas-a-world-model-for-spatial-intelligence\/","title":{"rendered":"Atlas: A World Model\u00a0for Spatial Intelligence"},"content":{"rendered":"\n
\n<\/p>\n

World models generate, reconstruct, and simulate any possible world.
\nThey understand how worlds appear, behave, and evolve
\nso that we can render imagined worlds for creative users,
\nsimulate the real world in high fidelity, and help robots plan actions.
\nAt World Labs, we build these general purpose world models in pursuit of spatial intelligence.<\/p>\n

Today we are introducing Atlas, our next-generation world model.
\nAtlas is an omni model that we pretrained from scratch to natively operate on
\ntext, images, video, and 3D.
\nIt is a multimodal autoregressive diffusion transformer:
\nall inputs are combined into a shared spatial context.
\nAtlas uses that context to generate what comes next,
\nstaying consistent in 3D with everything it has seen and imagining what lies beyond it.
\nAtlas is built to scale: its performance improves with increased training compute,
\nand we expect this trend to hold as we continue scaling.<\/p>\n

Atlas can perform a broad range of tasks
\nspanning world generation, reconstruction, and simulation:<\/p>\n

<\/p>\n