Cosmos answers commercial use in its own text
The tag is the NVIDIA Open Model License, written by the publisher. What sets the row apart is that the card says the models are commercially usable in its own text, rather than leaving that to a document elsewhere. As of 2026-09-22.
| Column | What the pages state |
|---|---|
| Licence, in the vendor's own words | NVIDIA Open Model License |
| The detail published beside it | The card states that models are commercially usable |
| Size, from the same pages | 2,059,174,912 parameters |
Inclusion rule. The value this page is about, the detail the vendor attached to it, and the two columns that change how it should be read. A column the vendor left empty keeps its row and says so. Order. This column first, then its detail, then the two columns that qualify it.
1Bespoke does not have to mean opaque
A licence written by a vendor usually exists because the vendor wanted to add something a standard document does not say. Sometimes that addition is a restriction; here the card uses it to state plainly that commercial use is allowed.
For a reader deciding whether to spend an afternoon on a licence file, that sentence on the card is the difference between reading it now and reading it later.
2The rest of the card is equally exact
The same page publishes a parameter count to the unit and a memory figure to two decimal places, with the GPU architectures named. A card that precise about numbers being precise about terms is consistent rather than surprising.
Precision is not correctness. A figure quoted to two decimals still describes one configuration, and the register keeps the configuration attached to it.
3What is still outside the entry
The sentence on the card is not the licence. The document has its own conditions, and a plan that redistributes weights or ships them inside a product has to be checked against the text rather than against the summary.
The register records the name, the card's own statement about commercial use, and the date it was read. It does not read the conditions on anyone's behalf.
- LicenceNVIDIA Open Model License, with the card stating that models are commercially usablea vendor-written licence answering the commercial question in its own text
- Model sizeNumber of model parameters: 2,059,174,912published as an exact count
4Sources
The licence name was read from huggingface.co/nvidia on 2026-09-22 and is quoted rather than summarised. The rest of this release is on its family entry; how names in this column are handled is on licence. Nearby: Wan, licensed and unlicensed, A hedged parameter count.