Cosmos 3 Super Reasoner
Cosmos 3 Super Reasoner is the language-generating 32B reasoning tower of NVIDIA’s Cosmos 3 Super world model, adapted from the Qwen3-VL 32B architecture. It accepts text, images, and video and produces textual reasoning about physical scenes, spatial relationships, trajectories, and visual events with a documented 256K context. NVIDIA’s downloadable Cosmos3-Super checkpoint packages this autoregressive tower with a separate 32B diffusion generator, making the full omnimodal checkpoint 64B rather than a 64B language reasoner. NIM exposes the reasoning tower as nvidia/cosmos3-super-reasoner; Nano’s 8B reasoner is a smaller family variant. OpenMDW 1.1 permits commercial and non-commercial model use.
2026-05-31
32B reasoner; 64B full Super checkpoint
Autoregressive multimodal Transformer; reasoning tower of Mixture-of-Transformers model
OpenMDW-1.1
Specifications
- Parameters
- 32B reasoner; 64B full Super checkpoint
- Architecture
- Autoregressive multimodal Transformer; reasoning tower of Mixture-of-Transformers model
- License
- OpenMDW-1.1
- Context Window
- 262,144 tokens
- Type
- multimodal
- Modalities
- textimagevideo
Benchmark Scores
Advanced Specifications
- Model Family
- Cosmos
- API Access
- Not Available
- Chat Interface
- Not Available
- Variants
- nvidia/cosmos3-super-reasonerCosmos3-SuperCosmos3-Nano 8B Reasoner
Capabilities & Limitations
- Capabilities
- physical-world reasoningvision-language understandingvideo understandingspatial reasoningtext generation
- Known Limitations
- Physical reasoning does not establish reliable autonomous controlFull Super checkpoint includes a separate generator and larger memory footprintSpatial outputs require task-specific validation
- Notable Use Cases
- physical scene analysisrobotics research assistantsvisual simulation planning