---
title: "NVIDIA Cosmos | World Foundation Model | world-models.io"
canonical: "https://world-models.io/en/models/nvidia-cosmos/"
description: "NVIDIA Cosmos is a world foundation model platform for physical AI, providing pre-trained models for autonomous driving, robotics, and embodied AI."
author: "Tyler D."
generated: "2026-10-30"
---

New: the Timeline is live. Track world model releases, papers, and benchmark updates in real time.

The Knowledge Hub for AI World Models

# NVIDIA Cosmos

NVIDIA Cosmos is a world foundation model platform for physical AI, using autoregressive and diffusion-based architectures trained on massive video datasets.

robotics

model-based-rl

simulation

embodied-ai

## What is a world model?

A world model is an AI system that learns a predictive representation of how an environment changes over time. It can use observations, actions, or language to estimate future states and possible outcomes before an agent acts. World models support planning in model-based reinforcement learning, robotics, embodied AI, autonomous driving, and interactive simulation. This site catalogs these systems with model records, research papers, benchmark context, and source links. Start with the [world models database](https://world-models.io/en/world-models-database/), compare systems on the [leaderboard](https://world-models.io/en/leaderboard/), or read the [Performance Index methodology](https://world-models.io/en/methodology/performance-index/).

## Key Attributes

Static HTML snapshot of the model record for crawlers and no-JS readers.

| Attribute | Value |
| --- | --- |
| Model | NVIDIA Cosmos |
| Lab / Organization | NVIDIA |
| Category | Foundation World Model |
| Subtype | Video Generation World Model |
| World Model Type | Video world foundation model |
| Primary Domain | Physical AI |
| Architecture | Autoregressive and diffusion-based transformer models |
| Modality | Video + 3D |
| Training Method | Large-scale video pre-training with physics-aware objectives |
| Status | active |
| Year | 2025 |
| Performance Index | 87/100 (medium confidence, v1.1) |

## About NVIDIA Cosmos

Main editorial body preserved directly in static HTML.

NVIDIA Cosmos is a comprehensive platform providing world foundation models that understand physics, spatial reasoning, and temporal dynamics. It combines autoregressive and diffusion-based transformer models trained on large-scale video data to generate physically consistent simulations for autonomous vehicles, robots, and embodied agents.

NVIDIA Cosmos is a video world foundation model developed by NVIDIA in 2025 for physical ai.

## Editorial Snapshot

Short extractable facts for answer engines and no-JS readers.

| Signal | Value |
| --- | --- |
| Definition | NVIDIA Cosmos is a video world foundation model developed by NVIDIA in 2025 for physical ai. |
| Short Description | A platform of state-of-the-art generative world foundation models for physical AI development. |
| Benchmark Rows | 2 |
| FAQ Entries | 2 |
| Related Models | 2 |
| Related Guides | 2 |
| Related Research Topics | 6 |
| Last Updated | 2026-03-12 |

## Notable Features

Key capabilities associated with this model.

- • Physics-aware video generation
- • Comprehensive platform for physical AI
- • Both autoregressive and diffusion modes
- • Industrial-scale deployment

## Use Cases

Representative applications attached to this model record.

Autonomous driving simulation

Robotics training

Physical AI development

Synthetic data generation

## Strengths and Limitations

Balanced assessment surfaced in static HTML.

### Strengths

- • Physics-aware generation
- • Massive scale training
- • Industrial applicability
- • Comprehensive platform

### Limitations

- • Proprietary ecosystem
- • Extreme compute requirements
- • Limited open-source components

## Benchmarks

Published benchmark evidence attached to this model record.

| Benchmark | Metric | Result | Source |
| --- | --- | --- | --- |
| Video Generation Quality | FVD 42.3 FVD↓ | State-of-the-art | [Source](https://arxiv.org/abs/2501.03575) |
| Physics Consistency | Physics Score 94.2 % | Industry leading | [Source](https://arxiv.org/abs/2501.03575) |

## References and Citations

Primary references preserved in static HTML for citation extraction.

| Reference | Link |
| --- | --- |
| NVIDIA, 2024. Cosmos: A Platform for World Foundation Models. | [Open source](https://arxiv.org/abs/2501.03575) |

## Related Models

Nearby models linked from the current editorial record.

| Model | Category | World Model Type | Index v1.1 |
| --- | --- | --- | --- |
| [Genie 2](https://world-models.io/en/models/genie-2/) | Generative World Model | Generative environment model | 79/100 |
| [UniSim](https://world-models.io/en/models/unisim/) | Generative World Model | Action-conditioned video simulator | 72/100 |

## Direct Comparisons

Side-by-side comparisons already connected to this model.

| Comparison | Matchup | Summary |
| --- | --- | --- |
| [World Models vs LLMs](https://world-models.io/en/compare/world-models-vs-llms/) | World Models vs Large Language Models | World models and LLMs represent fundamentally different approaches to AI. World models learn causal dynamics of physical environments; LLMs learn statistical patterns over text. Both are essential for the future of AI. |
| [NVIDIA Cosmos vs DreamerV3](https://world-models.io/en/compare/nvidia-cosmos-vs-dreamer/) | NVIDIA Cosmos vs DreamerV3 | Cosmos and DreamerV3 represent two different scales and approaches to world modeling: Cosmos is a foundation-scale video world model platform for physical AI, while DreamerV3 is a sample-efficient RL agent with learned dynamics. |
| [Sora vs Genie 2](https://world-models.io/en/compare/sora-vs-genie-2/) | Sora (OpenAI) vs Genie 2 (DeepMind) | Sora and Genie 2 both generate video from prompts, but they approach world simulation very differently. Sora generates passive, high-fidelity videos from text; Genie 2 generates interactive, controllable 3D environments from images. |
| [V-JEPA vs Video Generation Models](https://world-models.io/en/compare/v-jepa-vs-video-generation/) | V-JEPA (Meta) vs Video Generation Models (Sora, Cosmos) | V-JEPA and video generation models like Sora both learn from video, but follow opposite philosophies: V-JEPA predicts in abstract representation space without generating pixels, while video generation models focus on producing realistic pixel outputs. |
| [GAIA-1 vs Copilot4D](https://world-models.io/en/compare/gaia-1-vs-copilot4d/) | GAIA-1 (Wayve) vs Copilot4D (Waabi) | Both are world models designed for autonomous driving, but they operate on different sensor modalities: GAIA-1 generates camera video, while Copilot4D predicts LiDAR point clouds in 4D. |
| [Genie 2 vs UniSim](https://world-models.io/en/compare/genie-2-vs-unisim/) | Genie 2 vs UniSim | Both are generative world models that create interactive environments, but Genie 2 generates 3D worlds from single images while UniSim learns a universal action-conditioned simulator from diverse real-world data. |
| [NVIDIA Cosmos vs Genie 2](https://world-models.io/en/compare/nvidia-cosmos-vs-genie-2/) | NVIDIA Cosmos vs Genie 2 | Two foundation-scale world models with different strategies: Cosmos is an open industrial platform for physical AI training, while Genie 2 is a DeepMind research system that generates interactive 3D environments from images. |
| [GAIA-1 vs NVIDIA Cosmos](https://world-models.io/en/compare/gaia-1-vs-nvidia-cosmos/) | GAIA-1 vs NVIDIA Cosmos | Both are video-based world models for autonomous driving and physical AI, but GAIA-1 is a domain-specific driving world model from Wayve while Cosmos is a general-purpose foundation platform from NVIDIA. |

## Guides Referencing This Model

Crawler-readable guide links tied to this model.

| Guide | Summary |
| --- | --- |
| [How to Apply World Models in Robotics](https://world-models.io/en/guides/world-models-for-robotics/) | How to use world models for robot learning: from simulation-based training to real-world deployment and sim-to-real transfer. |
| [World Models vs Large Language Models: A Practitioner's Guide](https://world-models.io/en/guides/world-models-vs-llms-guide/) | How world models differ from LLMs in objective, architecture and capability, and why both paradigms are likely to converge on the path to general-purpose AI. |

## Research Topics Referencing This Model

Connected research areas surfaced directly in static HTML.

| Topic | Summary |
| --- | --- |
| [World Models for Robotics](https://world-models.io/en/research/world-models-robotics/) | How world models improve robot learning, learned simulation, safe exploration, and sim-to-real transfer across manipulation, navigation, and control. |
| [World Models vs LLMs](https://world-models.io/en/research/world-models-vs-llms/) | The key differences between world models and LLMs across objective, architecture, planning, physical reasoning, and embodied AI use cases. |
| [Foundation World Models](https://world-models.io/en/research/foundation-world-models/) | How foundation world models such as Cosmos and Genie 2 bring large-scale learned simulation to robotics, autonomous driving, and physical AI. |
| [AI Simulation Systems](https://world-models.io/en/research/ai-simulation-systems/) | How AI simulation systems and learned simulators reduce the reality gap and extend or replace hand-crafted engines for autonomous agents. |
| [World Models: A Comprehensive Survey](https://world-models.io/en/research/world-models-survey/) | A survey of AI world models covering taxonomy, leading architectures, landmark systems, open challenges, and future research directions. |
| [Video World Models](https://world-models.io/en/research/video-world-models/) | How video world models and video generation transformers learn physics, temporal consistency, and interactive simulation from large-scale video, from OpenAI Sora and Google DeepMind Genie to NVIDIA Cosmos and V-JEPA. |

## Timeline Mentions

Recent timeline events connected to this model.

| Event | Published | Source | Summary |
| --- | --- | --- | --- |
| [NVIDIA Cosmos-Predict2 released: improved physical consistency for robotics simulation](https://world-models.io/en/timeline/nvidia-cosmos-predict2/) | 2026-04-18 | NVIDIA Research | NVIDIA Research releases Cosmos-Predict2, a second-generation diffusion world model with stronger physical consistency for robot policy. |
| [NVIDIA Cosmos World Foundation Models open-sourced on Hugging Face](https://world-models.io/en/timeline/nvidia-cosmos-open-source/) | 2026-03-12 | NVIDIA Research | NVIDIA released the Cosmos family of world foundation models under an open license. |
| [Leaderboard update: Performance Index recalculated with March 2026 benchmark data](https://world-models.io/en/timeline/leaderboard-march-2026-update/) | 2026-03-02 | world-models.io Editorial | The world-models. io Performance Index has been recalculated using the latest benchmark data. |

## Frequently Asked Questions

FAQ answers rendered directly into static HTML for extractable responses.

### What is NVIDIA Cosmos used for?

Cosmos generates realistic simulated environments for training autonomous vehicles, robots, and other embodied agents with physics-aware world models.

### Is NVIDIA Cosmos open source?

NVIDIA has released some Cosmos model weights as open-source, while the full platform includes proprietary components.

## Quick Answer

Short extractable summary preserved directly in static HTML.

- NVIDIA Cosmos is a video world foundation model developed by NVIDIA in 2025 for physical ai.
- Use this page when you need a fast read on how NVIDIA Cosmos fits into the foundation world model landscape, then validate the details in the benchmarks, citations, and related pages.
- A key strength surfaced in the editorial record is physics-aware generation.

## Editorial Trust Signals

Editorial provenance and refresh policy preserved directly in static HTML.

**Published by** world-models.io editorial board.

**Lead editor** Tyler D. - Technical editor, methodology and benchmark analysis.

This model page synthesizes primary papers, official model pages, benchmark evidence, and related world-models.io context into a reference resource.

Each editorial page is assembled from primary sources, normalized into extractable summaries, checked for factual drift, and reviewed before publication or major refreshes. Last reviewed: 2026-03-12.

Pages are refreshed when a new paper, benchmark, release, architecture update, or stronger primary source materially changes the answer a reader or AI system should retrieve.

Each page links back to relevant primary sources and keeps a stable canonical URL so readers can verify claims, trace context, and reference the most up-to-date version. [See the editorial policy.](https://world-models.io/en/editorial-policy/)

Primary sources only

Last reviewed date visible

Methodology documented

Source links included

Correction channel published

Review date visible

## External Sources

Primary model and lab sources embedded in static HTML.

- [NVIDIA Cosmos paper](https://arxiv.org/abs/2501.03575) NVIDIA primary research source
- [NVIDIA Cosmos official page](https://www.nvidia.com/en-us/ai/cosmos/) NVIDIA official page
- [NVIDIA Research official website](https://research.nvidia.com) Related lab

## References

1. [[1] NVIDIA, 2024. Cosmos: A Platform for World Foundation Models.](https://arxiv.org/abs/2501.03575)