Cosmos 3 Nano
Cosmos 3 Nano is an omnimodal world model developed by NVIDIA for Physical AI applications like robotics and autonomous systems. It generates video, image, audio, and action commands from combinations of text, image, video, and action trajectory inputs.
Input
Output
Context
262K
Max Output
4K
Parameters
15.8B
Technical Specifications
Model TypeVision
Context Window262,144 tokens
Max Output Tokens4,096 tokens
Parameters15.8B
Release DateMar 10, 2026
Training CutoffNot available
Licenseother
Open SourceYes
Input Modalities
TextImageVideoAudio
Output Modalities
TextImageVideoAudio
Capabilities
Resources & Links
Estimated GPU Requirements for Cosmos 3 Nano
Estimates based on INT8 quantization. Actual requirements vary by framework and configuration.
More Models by NVIDIA
Llama Nemotron Embed Vl 1B V210K context · 1.7B · Open weightsNvidia Nemotron Nano 12B V2 VL BF16131K context · 13.2B · Open weightsLlama Nemotron Rerank 1B V2131K context · 1.2B · Open weightsLlama Nemotron Embed 1B V28K context · 1.2B · Open weightsNvidia Nemotron 3 Nano 4B BF16262K context · 4B · Open weightsParakeet Tdt 0.6B V30K context · 627.1M · Open weights
Related Tools
Data sourced from official provider APIs and documentation
Last updated: Aug 8, 2026
Your AI stack shouldn't stand still.
Every month new models become cheaper, faster, and more capable. Inferbase ensures your application automatically benefits without changing a single API call.