NodeValid

Published video-model weights, licences and hardware notes

Open-Sora 2.0 carries 11B in its name

This release is named as an 11B model, and the name is where the count lives. Every other figure on the page, including the memory table and the resolution pair, belongs to that same model. As of 2026-09-22.

Open-Sora in the size column, as publishedPutting the count in the release name makes the whole page unambiguous: there is one model, so there is no question which variant a figure describes.Fixed by the published countLeft to the runThe countAn 11B model—What it belongs toOpen-Sora 2.0, the current release—Licence on that releaseApache 2.0—Memory for a run—60.3GB peak at 768px on one GPUOutput size—256px and 768pxNo variant ambiguity to resolve
Fig. 1 One named size, so every figure on the page has a known owner.
Open-Sora in the size column, with the columns that qualify it. Read from github.com/hpcaitech on 2026-09-22.
ColumnWhat the pages state
Size, in the vendor's own wordsAn 11B model
The detail published beside itOpen-Sora 2.0, the current release
Licence, from the same pagesApache 2.0
Hardware, from the same pages60.3GB peak at 768px on one GPU

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.

1One model removes the commonest ambiguity

Where a family publishes several sizes, a memory figure or a resolution has to say which one it belongs to, and often does not. A single named release cannot have that problem.

It is the quiet advantage of a narrow release. The register has rows where a figure is stated for a family and the variant has to be inferred from context; this is not one of them.

2Eleven billion, documented at two scales of machine

The memory table gives 60.3GB on one GPU and 44.3GB across four or more, on H100 or H800 cards. Both belong to the same 11B weights, which is what makes them comparable.

That pair is the useful shape for planning. It says what sharding buys, in the vendor's own measurements, rather than leaving a reader to guess at the scaling.

3What the count does not cover

The resolution pair of 256px and 768px is stated as what the model supports, and the frame constraint is written as a rule the code enforces rather than as a ceiling in seconds.

So the size names the object and the other columns describe what the object was measured doing. Neither column is derivable from the other.

4Sources

The count comes from github.com/hpcaitech, read 2026-09-22, exactly as the vendor wrote it. Everything else this release publishes is on its family entry; why the count sits in its own column is on size. Nearby: 2B with no floor, 13B against 6GB.