NVIDIA Cosmos is the world's first world foundation model.

Judged as of 2025-01-07. True: no model publicly described or released before Cosmos (announced January 2025) was presented as a "world foundation model", that is, a general-purpose pretrained model of physical-world dynamics meant to be adapted for robotics or autonomous vehicles. Mostly true: earlier world models existed (such as Ha & Schmidhuber's World Models, or the GAIA-1 driving world model), but none was a broadly released general foundation model of this kind, so Cosmos could fairly be called the first openly available one. Mixed: earlier models arguably fit the term (such as DeepMind's Genie 2, announced December 2024, or GAIA-1 described as a world foundation model) but are debatable. Mostly false: one or more earlier models were clearly described and released as world foundation models, and the claim holds only for a narrow sense such as "first platform". False: an earlier model of the same type, explicitly described as a world foundation model, was publicly released before Cosmos.

Set before the research began.

1. E1 NVIDIA Makes Cosmos World Foundation Models Openly Available to Physical AI Developer Community NVIDIA Blog · 7 January 2025 Own record

   “NVIDIA Cosmos, a platform for accelerating physical AI development, introduces a family of world foundation models — neural networks that can predict and generate physics-aware videos of the future state of a virtual environment — to help developers build next-generation robots and autonomous vehicles (AVs).”

2. E2 Genie 2: A large-scale foundation world model Google DeepMind · 4 December 2024

   “Today we introduce Genie 2, a foundation world model capable of generating an endless variety of action-controllable, playable 3D environments for training and evaluating embodied agents.”

3. E3 Wayve releases GAIA-1 Technical Report to advance world models for autonomy Wayve · 3 October 2023

   “GAIA-1 is the first generative world model designed specifically for self-driving systems. It was introduced earlier this year and builds on Wayve’s research in future prediction, dreaming about driving, predicting in a bird’s eye view and learning a world model.”

4. E4 Scaling GAIA-1: 9-billion parameter generative world model for autonomous driving Wayve · 3 October 2023

   “In June 2023, we unveiled GAIA-1 as the first proof of concept of a cutting-edge generative model for autonomous driving.”

E1 is NVIDIA's own launch post. It presents Cosmos as an openly available family of world foundation models for robots and autonomous vehicles, but it does not itself say Cosmos is the first. A month before Cosmos, Google DeepMind introduced Genie 2 as "a foundation world model" for training and evaluating embodied agents (E2). Wayve's GAIA-1 was presented in 2023 as a generative world model for self-driving (E3, E4); neither quote calls it a "world foundation model". Genie 2 arguably fits the term, but the evidence does not show it was openly released or built for robotics or autonomous vehicles, so the criteria's "mixed" level applies.