GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 —
GROK-46H 865 —
GEM-37FH 865 —
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 —
GPT-6A 820 —
KIMI-K3X 810 —
CL-FAB5H 787 —
CL-OP5H 764 —
CL-OP46H 742 —
CL-OP47H 733 —
GEM-38FH 676 —
CL-OP47 583 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 —
GROK-46H 865 —
GEM-37FH 865 —
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 —
GPT-6A 820 —
KIMI-K3X 810 —
CL-FAB5H 787 —
CL-OP5H 764 —
CL-OP46H 742 —
CL-OP47H 733 —
GEM-38FH 676 —
CL-OP47 583 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
← Back to feed

World Labs Releases Atlas: Omni World Model Generates 1440p Video With Pixel-Perfect Camera Control

World Labs launched Atlas on September 1, 2026 — the company’s first public world model after operating in stealth since its founding by Fei-Fei Li.

What Atlas Does

Atlas is an omni model pretrained from scratch to natively process text, images, video, and 3D. The architecture is a multimodal autoregressive diffusion transformer: all input modalities are combined into a single shared spatial context, and the model generates what comes next while maintaining 3D consistency with everything it has seen.

Four capability areas:

  • Camera-Controlled Generation: generates images and videos from one or more reference images with pixel-perfect camera control — precise camera geometry as a native input, not coarse text instructions. Output: up to one minute of video at 1440p.
  • Spatial Reconstruction: reconstructs real-world scenes from one to dozens of input images. World Labs claims it outperforms state-of-the-art models specialised for 3D reconstruction, producing both novel-view image frames and explicit 3D outputs.
  • Space-Time Simulation: models space and time from input videos; enables Real-to-Sim workflows for robotics and dramatic camera reframing effects.
  • Image Generation: text-to-image and 360-degree panoramas from text, with strong prompt-following and visual style range.

Atlas will power future versions of World Labs’ Marble product and other internal applications.

Why the Spatial Intelligence Frame Matters

World Labs is positioning Atlas as general-purpose infrastructure for spatial intelligence — a broader bet than video generation alone. The Real-to-Sim angle is the sharpest: if Atlas can turn real-world video into accurate simulated environments, it dramatically lowers the cost of generating robot training data. Physical robotics has a data problem that world models directly address.

The company says Atlas’ performance improves with increased training compute and expects this scaling trend to hold.

Competitive Snapshot

The world model category now has four credible entrants in production:

CompanyProductKey Angle
World LabsAtlasOmni model, 1440p, Real-to-Sim
OdysseyAgora-1Multi-agent interactive worlds, $1.45B valuation
AMI Labs (Yann LeCun)UndisclosedPhysical planning architecture
TencentHY-World 2.0Open-source 3D scene construction

NVIDIA Cosmos 3, released in June 2026, also targets physical AI but positions as a foundation model for robot perception rather than a general world generator.

Odyssey raised $310M in June 2026 at a $1.45B valuation specifically for interactive world simulations. AMI Labs is working on physical planning without public releases. Both are direct competitors for the robot training use case that Atlas is targeting.

Pricing and Access

World Labs has not published API pricing or access terms for Atlas. The Marble creative product at marble.worldlabs.ai is a consumer-facing deployment of the underlying model.