Newsletter Subscribe
Enter your email address below and subscribe to our newsletter
[forminator_form id="25163"]

worldlabs+1aiweeklyaiweekly+1World Labs on Tuesday introduced Atlas, a multimodal AI model designed to generate, reconstruct, and simulate three-dimensional environments — a release the startup says marks a new frontier in what it calls "spatial intelligence."
"Introducing Atlas — a first of its kind multimodal world model trained from scratch," co-founder Fei-Fei Li wrote on X, calling it "the best camera conditioned world model ever."x
Atlas is what World Labs describes as a "multimodal autoregressive diffusion transformer" — a single architecture capable of processing text, images, video, and 3D data within a shared spatial framework. The model can generate camera-controlled video at resolutions up to 1440p for durations of up to one minute, using as few as one to six reference images and a user-defined camera trajectory.diariobitcoin+2
Beyond generation, Atlas handles spatial reconstruction from sparse inputs. According to the company's blog post, the model can reconstruct scenes from as few as two or three images, producing novel views, depth maps, point clouds, and 3D Gaussian splats. World Labs published benchmark results claiming Atlas outperforms reproduced open-source specialist models across DTU, ETH3D, ScanNet, and other evaluation sets. In human evaluations of camera-controlled generation, the company reported that third-party raters preferred Atlas over competitors including MiniMax H3, Gemini Omni Flash, and Seedance 2.5 by wide margins.aiweekly+2
Atlas builds on Marble, World Labs' first commercial model, which launched in November 2025 and enabled 3D world creation from multimodal inputs. The company followed that with a public World API in January 2026 for generating explorable environments, and in July 2026 acquired SceniX to bolster its robotics-oriented spatial intelligence capabilities.cryptobriefing+2
World Labs has raised $1.23 billion in total funding, including a $1 billion round in February 2026 with a $200 million contribution from Autodesk . AMD Advanced Micro Devices, Inc. , Nvidia , and Fidelity have also backed the company.cryptobriefing
World Labs positions Atlas for use in visual effects, video games, design, and robotics. For robotics applications, the model can generate simulated sensor observations along a robot's planned path, a capability the company frames as useful for training embodied agents through real-to-sim pipelines. However, the company's claims about Atlas's performance have not been independently validated, and no pricing, latency details, or compute requirements have been disclosed.worldlabs+2
Atlas is entering early access with select partners and will power future versions of Marble and other World Labs products.aiweekly+1