Cosmos-Reason2

An open vision language model from NVIDIA for physical AI and robotics, letting robots and vision agents reason about space, time and physics.

Licenses

Code: Apache-2.0 (opens in a new tab)
Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit a3b4a1d (opens in a new tab), on 5 Nov 2026.
Weights: NVIDIA Open Model LicenseTerms of their own, not a standard license
Read on huggingface.co/nvidia/Cosmos-Reason2-2B (opens in a new tab).
Commercial use: Unknown

A license Robotheca cannot classify by itself: read it before any commercial use.

Custom weights termsThe weights come under their maker's own terms (NVIDIA Open Model License), not a standard license: read their conditions before any use.

Reasons about physical common sense and embodied scenarios, using chain-of-thought to plan the next steps an agent might take.
Comes in 2B, 8B and 32B parameter versions, all published on Hugging Face.
Runs through the Transformers library and is recommended for deployment and batch inference with vLLM.
Supports online serving and offline inference for captioning videos and answering questions about images.
Can be post-trained with TRL or Cosmos-RL, and quantized with llmcompressor.
The repository now receives only limited maintenance; new development moved to Cosmos 3.

Written by Robotheca from the project's README (opens in a new tab) at commit a3b4a1d.

Checked on GitHub on 5 Nov 2026.

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.