cosmos-rl

Important

🚀 Cosmos 3 Has Arrived

Cosmos 3 is NVIDIA's next-generation foundation model platform for Physical AI. Compared with Cosmos-RL, Cosmos 3 unifies reasoning, world prediction, simulation, transfer, and action generation within a single model family and ecosystem.

Rather than relying on separate models for reasoning, prediction, transfer, and policy learning, a single Cosmos 3 model can understand the world, reason about physical interactions, predict future outcomes, transform observations across domains, and generate actions for embodied agents. This unified architecture enables stronger performance across a broad range of Physical AI applications, including robotics, autonomous vehicles, and smart spaces.

This repository is no longer under active development and will receive only limited maintenance updates. Future model releases, features, documentation, and community support will be focused on Cosmos 3.

👉 Visit the new Cosmos home: https://github.com/nvidia/cosmos

There you will find the latest Cosmos 3 models, technical reports, tutorials, benchmarks, and ecosystem updates.

Thank you for your support of Cosmos-RL. We encourage all users to migrate to Cosmos 3 for the latest state-of-the-art Physical AI capabilities.

NVIDIA Cosmos Header

Getting Started

Cosmos-RL is a flexible and scalable Reinforcement Learning framework specialized for Physical AI applications.

Documentation.

System Architecture

Cosmos-RL provides toolchain to enable large scale RL training workload with following features:

Parallelism
- Tensor Parallelism
- Sequence Parallelism
- Context Parallelism
- FSDP Parallelism
- Pipeline Parallelism
Fully asynchronous (replicas specialization)
- Policy (Consumer): Replicas of training instances
- Rollout (Producer): Replicas of generation engines
- Low-precision training (FP8) and rollout (FP8 & FP4) support
Single-Controller Architecture
- Efficient messaging system (e.g., weight-sync, rollout, evaluate) to coordinate policy and rollout replicas
- Dynamic NCCL Process Groups for on-the-fly GPU [un]registration to enable fault-tolerant and elastic large-scale RL training

Policy-Rollout-Controller Decoupled Architecture

License and Contact

This project will download and install additional third-party open source software projects. Review the license terms of these open source projects before use.

NVIDIA Cosmos source code is released under the Apache 2 License.

NVIDIA Cosmos models are released under the NVIDIA Open Model License. For a custom license, please contact cosmos-license@nvidia.com.

Metadata

Release files for cosmos-rl 0.4.6

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for cosmos-rl 0.4.6
File	Size	Uploaded
cosmos_rl-0.4.6.tar.gz	6.1 MB	Sep 16, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for cosmos-rl 0.4.6
File	Interpreter	ABI	Platform	Reset
cosmos_rl-0.4.6-py3-none-any.whl 1.9 MB Sep 16, 2026	Python 3	none	any	Details

Total release size: 8.0 MB

Release files / cosmos_rl-0.4.6.tar.gz

Download URL	cosmos_rl-0.4.6.tar.gz
Size	6.1 MB
Tags	Source
SHA-256 checksum How to use checksums	`1ab5cf2195d47c541c3235f484677d127a6baa89922afa69b258a8116a6fa257`
BLAKE2b-256 checksum How to use checksums	`e4bc7997905c5d9ffc081c4f2059dfe27cc4411b48132b52a59efde6eb833799`
Upload date	Sep 16, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Release files / cosmos_rl-0.4.6-py3-none-any.whl

Download URL	cosmos_rl-0.4.6-py3-none-any.whl
Size	1.9 MB
Tags	Python 3
SHA-256 checksum How to use checksums	`17231a54062d799c3734707f1a828982784ff4ded596991adf4dc1e5443a09fa`
BLAKE2b-256 checksum How to use checksums	`1fea561451ae1c5e3157e7824a54ffa5cc8327edf73ba8c942aa077f5a1f7328`
Upload date	Sep 16, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	No
Uploaded via	`twine/7.0.0 CPython/3.13.14`