NVIDIA Expands Cosmos 3 with New Open-Weight Models

# New NVIDIA Cosmos 3 Open Models for Interactive World Simulation

## Introduction

On November 5, 2026, NVIDIA announced a significant expansion of its Cosmos 3 model family, introducing four new open-weight checkpoints designed for interactive world simulation and autonomous driving applications. This development is part of NVIDIA's ongoing effort to enhance AI-driven simulation capabilities. The newly unveiled models, Cosmos3-Nano-Transfer-Auto, Cosmos3-Nano-Sim-Auto, Cosmos3-Nano-Sim-Transfer, and Cosmos3-Nano-Sim-Bimanual, each boast 16 billion trainable parameters. These models are post-trained examples of the Cosmos 3 Nano, and they utilize the OpenMDW1.1 license, allowing for both commercial and non-commercial use.

## What Changed

NVIDIA's launch of these four models marks a significant advancement in AI technology, particularly for the fields of robotics and autonomous vehicles. These models are designed to simulate complex environments and interactions, offering functionalities such as multi-sensor data generation and video prediction based on robotic actions. This move aligns with NVIDIA's strategic vision to support the development of more sophisticated AI simulations. By providing open-weight models, NVIDIA is positioning itself as a key player in the AI industry, encouraging innovation and collaboration.

The Cosmos3-Nano-Transfer-Auto model is tailored for generating multi-view driving data under varied conditions such as weather, lighting, and time of day. It supports synchronized video generation across up to 11 camera views, including fisheye lenses, and integrates joint RGB-LiDAR generation. This capability is particularly beneficial for developers looking to expand training datasets and evaluation coverage. By providing options to vary environmental conditions, this model serves as a foundational tool for adapting simulations to specific sensor configurations and driving domains.

The Cosmos3-Nano-Sim-Auto model focuses on streaming synchronized camera and LiDAR observations for driving simulations. It ensures that visual observations align with 3D sensing, which is crucial for developing reliable autonomous driving systems. This model allows researchers to manipulate scene and traffic-signal controls, enabling them to study how different driving policies respond to various road layouts and signal states. The ability to simulate these scenarios is essential for advancing autonomous vehicle technology.

Cosmos3-Nano-Sim-Transfer turns simulator controls and text prompts into photorealistic video outputs. This model bridges the gap between simulation and reality, offering a unique tool for generating realistic video content based on virtual controls. This functionality is particularly useful for creating immersive simulations that can be used for training and evaluation purposes in both robotics and autonomous driving applications.

The Cosmos3-Nano-Sim-Bimanual model is designed to predict video outcomes from robot or camera actions. This capability is crucial for developing interactive simulations where the consequences of actions are visually represented. By providing a tool that links actions to visual outcomes, this model supports the creation of more dynamic and responsive simulated environments.

## Strategic Implications

The introduction of these open-weight models has far-reaching implications for the AI landscape. By offering these models openly, NVIDIA is likely to drive innovation across the industry, enabling researchers and developers to build more advanced simulations. This could potentially shift the competitive dynamics, as companies that rely on proprietary technology may find themselves at a disadvantage. The open nature of these models also opens the door for international collaborations, although it may increase dependency on NVIDIA's ecosystem for AI advancements.

## What Happens Next

In the months following this release, the AI research community is expected to adopt these models widely. Researchers and developers will likely integrate them into existing projects, particularly in the fields of autonomous driving and robotics. By Q3 2027, new applications and breakthroughs may emerge as teams leverage the open-weight nature of these models to customize and enhance their functionality. This could lead to policy discussions around standardizing open AI model practices to ensure ethical use and interoperability.

## Second-Order Effects

The broader availability of these models could lead to significant advancements in AI-driven simulations, impacting supply chains related to autonomous vehicle development. As more companies adopt these models, market dynamics could shift, allowing smaller firms to compete with established players. Regulatory bodies might also need to consider new guidelines to address the implications of widespread use of open AI models, particularly in safety-critical applications.

## Expert Perspective

NVIDIA's expansion of the Cosmos 3 model family reflects a growing trend towards open AI development. Similar to the release of OpenAI's GPT-3, this move democratizes access to powerful AI tools. However, unlike GPT-3, NVIDIA's focus is on simulation and robotics, highlighting a strategic pivot towards these sectors. Analysts suggest that this could position NVIDIA as a leader in providing foundational tools for the next generation of AI research, with a potential increase in global AI autonomy.

In conclusion, the introduction of these four new models represents a significant step forward in AI technology, providing researchers and developers with powerful tools to advance the fields of robotics and autonomous driving. As these models are adopted and integrated into various projects, the potential for innovation and collaboration within the AI community is immense.