World Labs opens its Atlas world model to everyone with full developer APIs: worlds from one image, reconstruction from up to 64 inputs, splat export

Why it matters

【Key points】

· Launch: On Nov 12 (18:09 UTC) World Labs opened Atlas to everyone, with apps and developer API endpoints. Its blog lists "Atlas: Releasing our next generation world model" the same day, linking to the Atlas product page.

· Viewfinder: generate a world from a single image, or reconstruct a real environment from up to 64 inputs. Export spatially consistent images, video or 3D Gaussian splats.

· Chisel: agentic workflows for explicit 3D geometry, from text or reference images, with renders or splat export. Splats to Mesh and Relight also launched.

· API: every endpoint behind the apps, from text-to-image and pose/depth estimation to view generation, Chisel, splat reconstruction and GLB meshes, plus an MCP server, a Blender integration and Spark.js.

• · Pricing: 1,000 credits per US dollar, and credits do not expire.
• Per-task rates appear only in the platform.
• A Frameboy session costs 500 credits ($0.50).
• Rate limits are "to be announced."

【Our read】

Background

• Atlas debuted Sep 1 as an omni autoregressive diffusion transformer (video up to 1 min at 1440p).
• World Labs said Sep 28 it is joining AMD.
• This step is about access and APIs.

Comparison

• DeepMind's Genie 4 remains a research preview with no public pricing.
• Viggle's JST-2 likewise uses explicit 3D and up to 64 views.
• Atlas ranked #1 of 19 models in Rapidata's camera-movement benchmark, which measures one task only.

Impact

Splat and mesh export plus Blender and MCP hooks target real 3D pipelines, but per-task prices and rate limits are not public yet.

【Verifiable】

• · Atlas page
• · Launch post
• · Viewfinder post
• · Docs
• · Billing