

Unknown · od 1. September 2026 · 2× · naposledy 14. 11. 2026
18
Momentum
Atlas is an "omni" world model from World Labs (Fei-Fei Li), which according to the vendor was pretrained from scratch on text, images, video and 3D. It is described as a multimodal autoregressive diffusion transformer and generates camera-controlled videos of up to one minute at 1440p as well as 3D reconstructions from a few images. The performance figures come from the vendor and have not been independently verified.
Vývoj momenta
18.08.16.11.
Vlastnosti
| Key Benchmark (%) | Human raters preferred Atlas in camera-control tests 75% (vs. MiniMax H3) to 94% (vs. Seedance 2.5); vendor's own test |
| Multimodality | Text, images, video and 3D in one model; input of 1 to 6 reference images |
| Platform | Web apps and public API (api.atlas-beta.worldlabs.ai/api/v2), per report dated November 12, 2026; previously early access via request form only |
| Price per 1M Tokens | Not published |
| Release Date | September 1, 2026 (early access) |