CogVideoX-3

CogVideoX-3 is Z.ai's hosted video generator, the third generation of a line that started as one of the most widely used open-weight video models on Hugging Face. That heritage is worth stating plainly, because it ended here. CogVideoX-2b and CogVideoX-5b are still downloadable and still pulled tens of thousands of times a month; the third generation exists only behind the API. Anyone who built on the open version cannot follow the line forward. What the closed generation buys is reach. The output ceiling is 3840x2160 — genuine 4K, where the open predecessors topped out at 720p — with a documented list of shapes covering landscape, portrait, square and cinematic ratios. Clips run 5 or 10 seconds at 30 or 60 frames per second, and the model can generate a soundtrack alongside the picture rather than returning silent video. The input side is more interesting than the resolution. Besides text prompts and single-image animation, the model accepts a first frame and a last frame together and generates the transition between them — a control that turns the generator from a slot machine into something closer to a tool, because the beginning and the end are fixed by the user. The documentation notes the catch: that mode works only in the faster of the two quality settings. Pricing is flat and unusually simple for this class of model: $0.20 per video, regardless of length, resolution or frame rate. A 10-second 4K clip at 60 fps costs exactly the same as a 5-second 720p one.

▸Videos

No videos yet.

▸Reviews