- NVIDIA launched Cosmos 3, a new open world foundation model built on a mixture-of-transformers architecture, and announced the Cosmos Coalition to coordinate open-model development around it.
- LTX is one of six founding members, alongside Agile Robots, Black Forest Labs, Generalist, Runway, and Skild AI, contributing models, research, and evaluation methods while building on Cosmos 3 and NVIDIA DGX Cloud infrastructure.
- This extends LTX's existing position as an open world model with a pretrained checkpoint built for physical AI adaptation, rather than marking a pivot into new territory.
NVIDIA used its keynote at GTC Taipei to launch Cosmos 3, its newest open world foundation model for physical AI, and to announce a new collaboration it's calling the Cosmos Coalition.
LTX is one of six founding members named in NVIDIA's own announcement, alongside Agile Robots, Black Forest Labs, Generalist, Runway, and Skild AI.
What Is NVIDIA Cosmos 3?
Cosmos 3 is NVIDIA's latest open world foundation model, built on a mixture-of-transformers architecture that pairs a reasoning transformer with a separate generation transformer, letting the model understand object interactions, motion, and spatial relationships before it generates video or action trajectories.
NVIDIA describes Cosmos 3 as the first fully open omnimodel that natively understands and generates across text, image, video, ambient sound, and action, and reports that it leads open models on several physical AI and vision benchmarks.
The lineup ships in three sizes: Cosmos 3 Super for high-accuracy post-training work, Cosmos 3 Nano for fast video and action reasoning, and Cosmos 3 Edge, coming soon, for real-time inference at the edge.
Cosmos 3 Super and Nano are available through Hugging Face, and NVIDIA NIM microservices.
Announcing the launch, NVIDIA CEO Jensen Huang framed it as a turning point for physical AI, driven by recent progress across multimodal reasoning, vision, and world models.
What Is the Cosmos Coalition, and Why Does It Matter?
The Cosmos Coalition is NVIDIA's attempt to organize the open side of the world model race into something more coordinated than a shared architecture.
Founding members, including LTX, Agile Robots, Black Forest Labs, Generalist, Runway, and Skild AI, contribute models, research, and evaluation methods while building on Cosmos 3 and NVIDIA DGX Cloud infrastructure.
The stated goal is faster innovation and broader interoperability across the group, rather than every company solving the same foundational problems in isolation.
That's one direction the world model category is taking shape around.
Google DeepMind's Genie 3 world model, for example, points toward interactive, generated environments, currently available through a closed research preview with no released weights and no public API.
The Cosmos Coalition is organizing around a different approach: open models, shared research, shared evaluation, and shared infrastructure.
Why This Matters for LTX
LTX-2.5 is already an open world model, released as open weights with a pretrained checkpoint built for deep adaptation toward physical AI domains: egocentric robotics and manipulation, action-conditioned world prediction, autonomous vehicle and drone data, industrial digital twins, and private domain-specific models.
Joining a coalition organized around exactly that kind of open, adaptable foundation is a natural extension of that positioning, not a pivot into new territory.
It also puts LTX in direct company with other builders working on adjacent pieces of the same problem: Black Forest Labs and Runway on generative media, Agile Robots, Generalist, and Skild AI on the robotics and physical AI side.
Coalition members can contribute models, research, and evaluation methods while using Cosmos 3 technologies, training tools, and NVIDIA DGX Cloud infrastructure for large-scale training.
What This Means for Builders
For teams already working with LTX, this doesn't change what LTX-2.5 does today. What it signals is direction: LTX is positioning itself inside the open world model ecosystem NVIDIA is trying to organize at scale, alongside the same infrastructure and research-sharing that's meant to accelerate physical AI development industry-wide.
The next generation of world models won't be defined only by what they can generate. It'll be defined by what people can build on top of them.
