Nvidia introduced the “Cosmos Reason” artificial intelligence model with 2 billion parameters

Nvidia, the world’s semiconductor giant, unveiled its latest generation of World Models at the Siggraph 2025 conference. The most prominent product of this collection is Cosmos Reason; A 5 billion parameter vision-based model developed for “physical artificial intelligence” and robotics applications and is capable of predicting and planning a smart factor (such as a robot) based on the memory and understanding of physics laws.

In addition, the second version of the Cosmos Transfer was also introduced, which could accelerate the production of synthetic data from 3D simulations and space inputs. There is also a more compact version of the model that has a faster speed. These models are designed to build artificial text, visual and video databases in the tutorials on robots and artificial intelligence agencies.

Nvidia has also introduced new neural regeneration libraries that allow developers to simulate the real environment using sensor data. This feature has also been added to the Carla Simulator and a new version of the Omniverse Development Kit has been released.

In the hardware section, the RTX Pro Blackwell server was unveiled as a single architecture for all the robotic development processes and DGX Cloud cloud platforms to manage these processes.

Nvidia’s new move shows that the company is seeking to expand its graphic processor territory beyond artificial intelligence data centers and more serious entry into the raising market.