As of: 2026-11-13 13:49 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/cosmos-3/

# Cosmos 3 (Nano / Super)

NVIDIA, Cosmos family. Status: current. Type: world model. Released: 2026-06-01.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Hugging Face | `nvidia/Cosmos3-Nano` |  | https://huggingface.co/nvidia/Cosmos3-Nano |
| Hugging Face | `nvidia/Cosmos3-Super` |  | https://huggingface.co/nvidia/Cosmos3-Super |
| GitHub |  |  | https://research.nvidia.com/labs/cosmos-lab/cosmos3/ https://github.com/nvidia-cosmos |

## Pricing

- Price not published.
- Last checked: 2026-09-29

## Specs

- Input: text, image, video, audio, action
- Output: text, image, video, audio, action
- Context window: not published
- Max output: not published
- Open weights: yes
- Licence: OpenMDW-1.1 (commercial and non-commercial use)

## Where it stands

- Newer world models in the registry since its release: 2:
  - Genie 4 (2026-10-27) https://postcutoff.com/m/genie-4/
  - Xiaomi-Robotics-U0 (38B) / U0-4B (2026-07-13) https://postcutoff.com/m/xiaomi-robotics-u0/
- Before it from NVIDIA: Cosmos Predict 2.5 / Transfer 2.5 (2025-10-06) https://postcutoff.com/m/cosmos-predict-2-5/
- Weights are public. Licence: OpenMDW-1.1. See the licence for commercial terms.

## What stands out

- Unified omni world model (generation + reasoning + action) (first): One Mixture-of-Transformers model (autoregressive + diffusion) replaces separate Cosmos Predict, Transfer, Reason and Policy models: world generation, physical reasoning, forward/inverse dynamics and action/policy generation. Source: https://nvidianews.nvidia.com/news/nvidia-and-global-robotics-leaders-take-physical-ai-to-the-real-world
- Open omnimodal I/O: Inputs text, images, short video, audio and action trajectories (16-400 frames); outputs text, images, video (5-400 frames), 48 kHz stereo audio and actions (JSON). Source: https://huggingface.co/nvidia/Cosmos3-Nano
- Leaderboard results (found after launch): NVIDIA cites best open text-to-image and image-to-video models on Artificial Analysis and best policy model on RoboArena. Source: https://www.nvidia.com/en-us/ai/cosmos/

## Notes

Announced at GTC 2026-03-16 ('the first world foundation model unifying synthetic world generation, vision reasoning and action simulation' - NVIDIA claim); weights published 2026-05-31/06-01 (HF blog 'The First Open Omni-model for Physical AI Reasoning and Action'). Sizes: Nano 16B, Super 64B. Linux + Ampere/Hopper/Blackwell GPUs, BF16. Technical report dated 2026-06-22.

Sources: [HF blog](https://huggingface.co/blog/nvidia/cosmos-3-for-physical-ai), [Cosmos3-Nano](https://huggingface.co/nvidia/Cosmos3-Nano), [Cosmos3-Super](https://huggingface.co/nvidia/Cosmos3-Super), [technical report](https://research.nvidia.com/labs/cosmos-lab/cosmos3/technical-report.pdf).

Model page: https://postcutoff.com/m/cosmos-3/