NVIDIA used its Computex Taipei keynote on June 1, 2026 to launch the Cosmos Coalition, an open ecosystem for world models with six founding members. The lineup includes creator-tool makers Runway, Black Forest Labs, and LTX, signaling that the next round of video and image models will share a common training stack, datasets, and evaluation harness.
Who Joined and Why It Matters for Creators
The six founding members are Agile Robots, Black Forest Labs, Generalist, LTX, Runway, and Skild AI. Three of those ship the tools working creators use daily: Runway for generative video and the new Aleph editing system, Black Forest Labs for the FLUX family that powers most production image pipelines, and LTX for the LTX-Video open-weights generator and LTX Studio production environment. Coalition members commit to sharing weights, datasets, evaluation methods, and training recipes built on Cosmos 3 and NVIDIA DGX Cloud.
Try It Now: Download Cosmos 3 Nano
The fastest creator action this week is to pull Cosmos 3 Nano from Hugging Face, which is 16B parameters and fits on a single 24GB GPU. Run the bundled Diffusers example to generate a 5-second video clip with synced ambient audio. The Nano variant is the entry point for evaluating whether to swap your current text-to-video model for an open-weights base you can fine-tune locally. Cosmos 3 also ships in a higher-fidelity Super tier and a forthcoming Edge tier sized for mobile-class hardware.
Why the Coalition Format Matters
Until now, every flagship video and world model has been a black-box silo: Runway, Pika, Kling, Veo, Sora, and FLUX all train independently, with no shared evaluation harness and no interoperability targets. The Cosmos Coalition is the first cross-vendor commitment to publish weights, datasets, and training recipes under a single ecosystem. For creators, this should narrow the workflow split between "good at lighting" and "good at camera moves," because every coalition member will train against the same physics and continuity benchmarks. Press release language at the NVIDIA blog describes the shared roadmap as targeting "faster innovation, broader interoperability, and more rapid advances in physical AI."
Key Details
Cosmos 3 itself uses a mixture-of-transformers architecture that pairs a reasoning transformer with an expert generation transformer. It natively handles text, image, video, ambient sound, and action prediction in one model. The first joint coalition releases are expected at GTC Washington in October.
What to Do Next
If you use Runway, FLUX, or LTX in production, bookmark each coalition member's news page so you catch Cosmos-3-derived releases the day they ship. The integration roadmap is open, so you can also file feature requests on the Cosmos 3 GitHub repo for specific creator workflows you want supported in the shared evaluation harness.
