NVIDIA · Image Generation
Cosmos 3 Super — Text to Image
Generate high-resolution images from text prompts. Supports aspect ratios 1:1, 16:9, 9:16, 4:3, and 3:4.
- 64B parameters
- MoT architecture
- #1 open on Text-to-Image eval
- JPEG / PNG output
Model Index
AI Models on cosmos3ai
cosmos3ai is powered by NVIDIA Cosmos 3 Super — a 64B Mixture-of-Transformers model trained on 20 trillion multimodal tokens. Below are the capabilities currently available on the platform.
Image Models
Try it
Video Models
NVIDIA · Video Generation
Cosmos 3 Super — Image to Video
Animate a still image into a short cinematic video clip. Supports 3, 5, and 8 second output durations.
- 64B parameters
- MoT architecture
- #1 open on R-Bench
- MP4 output up to 8s
Benchmark Results
All results are from official NVIDIA launch materials published on 2026-05-31. Source: NVIDIA Developer Blog.
#1 openVANTAGE-BenchVision understanding
#1 openPhysics-IQPhysical reasoning
#1 openR-BenchRobotic video world models
#1 openArtificial Analysis Text-to-ImageImage generation
Architecture
- Family: Mixture-of-Transformers (MoT)
- Total parameters: 64B
- Training tokens: 20 trillion multimodal tokens
- 32B Reasoner
- 32B Generator
- Long-context multimodal reasoning
- Hosted browser workflow