Primary use stated or inferred cautiously from official documentation.
N
NVIDIA / nvidia/Cosmos3-Nano-Sim-Transfer
Cosmos3-Nano-Sim-Transfer
NVIDIA distilled causal model published on October 20, 2026, to convert visual simulation controls into RGB video. Geometry, physics, and ground truth still belong to the classic simulator; the network provides visual appearance. It accepts text and depth, segmentation, edge, or blur controls. It is not a policy that produces robot actions nor a certified physics engine.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | nvidia/Cosmos3-Nano-Sim-Transfer |
|---|---|
| Model type | Robotics · Video · Multimodal |
| Access model | Open weights |
| License | OpenMDW 1.1 |
| Deployment | Self-hosted |
| Release | 20 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Video |
| Exit | Video |
| Context window | Depth, segmentation, edge, or blur controls |
| maximum output | Up to 600 frames at 480p; playback at 30 fps |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · PyTorch |
Visual variation for reinforcement learning and simulation-to-real transfer. The model card describes integration with Isaac Lab and up to 600 frames at 480p. The example of eight parallel environments depends on hardware and configuration: it is not a universal limit or performance. Check alignment between images and simulator annotations.
Photographic appearance may hide temporal errors or control inconsistencies. Evaluate stability throughout the sequence and preserve seeds, prompts, steps, and checkpoint revision. Figures from other Cosmos3 models do not transfer to this variant. Its visual output does not demonstrate that a policy trained with it is safe on real hardware.
03 / CAPABILITIES
What it can do
Orientation
Visual variation for reinforcement learning and simulation-to-real transfer. The model card describes integration with Isaac Lab and up to 600 frames at 480p. The example of eight parallel environments depends on hardware and configuration: it is not a universal limit or performance. Check alignment between images and simulator annotations.
Context
Depth, segmentation, edge, or blur controls · Up to 600 frames at 480p; playback at 30 fps
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
TextVideo
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
Public 16B parameter checkpoint for self-hosted infrastructure. We have not verified a commercial endpoint or token rates: numerical costs are left blank. Downloadable weights do not make inference free. Budget for GPU, memory, video encoding, and operation; review the license of each variant and its dependencies. Playback frames per second do not equate to generation performance.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
NVIDIA publishes TransferBench within PAI-Bench-C. DOVER and segmentation mIoU improve as they increase; depth SI-RMSE improves as it decreases. These are different metrics, not interchangeable percentages. We recorded values for this checkpoint with a link to the exact revision. They are not independent Inferama results nor a certification of physical consistency.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| PAI-Bench-C TransferBench DOVER ↑ | 10,175942 | Score | View source ↗ |
| PAI-Bench-C TransferBench segmentation mIoU ↑ | 0,73776 | Score | View source ↗ |
| PAI-Bench-C TransferBench depth SI-RMSE ↓ | 0,619426 | Score | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Cosmos3-Nano-Sim-Transfer
Version added to Inferama's verified catalog.
07 / SOURCES