480p for now against 720P at 16FPS
Mochi 1 states 480p and describes it as the current capability. Cosmos Predict 2.5 states 720P video with 16FPS as a five-second clip. A hedge in the present tense against a complete specification. As of 2026-09-22.
| Point | Mochi 1 | Cosmos Predict 2.5 |
|---|---|---|
| Resolution, as each vendor states it | 480p | 720P |
| The detail published beside it | Described as the current capability | Stated with a 16FPS frame rate |
| Hardware, from the same pages | About 60GB VRAM on a single GPU | 32.54 GB of GPU VRAM |
| Length, from the same pages | No length recorded | A 5-second clip |
Inclusion rule. Two families whose published values in this column lead to opposite decisions. A pair that happens to agree is not listed, because agreement is already visible on the column summary. Order. The column itself first, then its detail, then the columns that qualify it.
1The tense is the information
Current capability tells a reader that 480p is what the release does and that the maintainers did not consider it a property of the design. Both halves matter to somebody planning around it.
The register keeps the phrase. Recording a bare 480p would drop the second half and turn a provisional figure into a specification.
2Complete does not mean generous
720P at 16FPS in five seconds is fully specified and short. Sixteen frames a second is lower than most releases here, so the clip is about eighty frames.
Precision and ambition are separate axes. The exactly specified row is not the one making the larger claim.
3The memory figures invert the expectation
Mochi asks for about 60GB on a single GPU for 480p output. Cosmos states 32.54 GB for 720P. The release with the lower stated resolution needs the larger card.
That is the clearest illustration in this column that resolution does not drive memory on its own, and that the columns cannot be read as one ranking.
- Resolution480p, described as the current capabilitystated as a present limit
- Resolution and lengthProduces 720P video with 16FPS, as a 5-second clipboth stated on the card
4Sources
Read from github.com/genmoai and huggingface.co/nvidia on 2026-09-22. Full entries: Mochi 1 and Cosmos Predict 2.5. What this column is quoting is on resolution. Nearby pairs: Borrowed or published, 6GB against 78GB.