NodeValid

Published video-model weights, licences and hardware notes

CogVideoX puts the size in the name

The sizes here are not a line in a table; they are part of the names. CogVideoX-2B and CogVideoX-5B carry their own counts, and in this family the size also selects the licence and the memory floor. As of 2026-09-12.

CogVideoX in the size column, as publishedWhen the size is in the name, every other statement about the release can be indexed by it, which is why this family has three values where most have one.Fixed by the published countLeft to the runThe count2B and 5B, carried in the model names—What it belongs toThe size decides which licence applies aswell—Licence on that releaseSplit by model size—Memory for a run—From 4GB in FP16 on the 2BOutput size—1360x768 on 1.5-5BOne name, three consequences
Fig. 1 A naming convention doing the work a table would otherwise have to do.
CogVideoX in the size column, with the columns that qualify it. Read from github.com/THUDM on 2026-09-12.
ColumnWhat the pages state
Size, in the vendor's own words2B and 5B, carried in the model names
The detail published beside itThe size decides which licence applies as well
Licence, from the same pagesSplit by model size
Hardware, from the same pagesFrom 4GB in FP16 on the 2B

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.

1A naming convention with real consequences

Putting the count in the name means no reader has to look it up, and it makes every other figure addressable: the 4GB floor belongs to the 2B, the 5GB floor to the 5B, the 10GB floor to the 1.5-5B.

It also removes the most common ambiguity in this column, which is a count published without saying which variant it describes. Here that cannot happen.

2The size is also the licence key

This is the one family in the register where checking the count is a legal step as well as a capacity step. The 2B release is Apache 2.0; the larger one carries the project's own terms.

Two lookups collapse into one. That is convenient and it is also a trap for anyone who reads the badge on the repository and assumes it covers both.

3What it does not fix

A name cannot carry a configuration. The memory figures still depend on precision, and the 4GB figure is stated for FP16 rather than for the model in the abstract.

The output figures move too: 1360x768 with up to 161 frames on one release, 720x480 on another. The name indexes the values; it does not compress them.

  • Hardware
    From 4GB in FP16 for the 2B model, from 5GB in BF16 for the 5B, and from 10GB for CogVideoX1.5-5Bthe lowest figures the repository statesTHUDM, CogVideo repository / recorded 2026-09-12
  • Length and resolution
    CogVideoX1.5-5B states 1360x768 and up to 161 frames, described as 5 or 10 secondsper modelTHUDM, CogVideo repository / recorded 2026-09-12

4Sources

The count comes from github.com/THUDM, read 2026-09-12, 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: Counted in two parts, VAE and DiT apart.