Skip to content

[GitHub Trending] NVIDIA/cosmos

7.9 relevance
Score Breakdown
technical depth
9
novelty
9
actionability
4
community
8
strategic
9
personal
9

Scored daily by a customisable AI persona to surface the most relevant engineering leadership news.

NVIDIA's open platform for world models is a major strategic move in Physical AI, highly novel.

AI/ML github.com
NVIDIA Cosmos is an open platform of world models, datasets, and tools that enables developers to build Physical AI for robots, autonomous vehicles, smart infrastructure, and more. - NVIDIA/cosmos
Summary

NVIDIA Cosmos 3 is an open platform of omnimodal world models for Physical AI, offering two runtime surfaces: Reasoner (text/vision to text for world understanding and planning) and Generator (multimodal in/out for simulation and synthetic data). Built on a unified Mixture-of-Transformers architecture with shared attention and 3D mRoPE embeddings, it comes in 16B (Nano) and 64B (Super) sizes and supports Diffusers, vLLM-Omni, and Transformers for development and serving.

Author

NVIDIA