World Labs

AI · Spatial Computing · Robotics

Launch

Fei-Fei Li's World Labs launches Atlas, an omni world model for spatial AI

The model fuses image, video and 3D generation into one system, betting spatial reasoning is the next frontier beyond language AI.

• World Labs unveiled Atlas on September 1, 2026, a model that generates and reconstructs pixel-perfect, camera-controlled images and video at up to 1440p for durations up to one minute.
• World Labs was co-founded by Stanford's Fei-Fei Li, ImageNet creator and former Stanford AI Lab director, alongside Justin Johnson, NeRF co-creator Ben Mildenhall, and Christoph Lassner, ex-Meta Reality Labs.
• Atlas is a multimodal autoregressive diffusion transformer pretrained from scratch on text, images, video and 3D data, unifying them into a single shared spatial context rather than separate pipelines.
• Atlas builds on World Labs' Marble model launched in November 2025, its World API released in January 2026, and its July 2026 acquisition of SceniX to bolster robotics-focused spatial intelligence.
• World Labs has raised $1.23 billion total, including a $230 million seed at a roughly $1 billion valuation in 2024 and a $1 billion round in February 2026 with $200 million from Autodesk, alongside backing from AMD, NVIDIA and Fidelity.
• Spatial AI still has no agreed-upon recipe or evaluation standard the way language models converged on transformers, so Atlas's real test is whether its results hold up in independent and partner use, not just company demos.