CogVideo Models by Zhipu AI
Last refreshed 2026-07-11. Next refresh: weekly.
Capabilities
About
CogVideo is Zhipu AI's family of video generation models. The original CogVideo was released as an open, large Transformer-based text-to-video model. Its successor line, CogVideoX, includes open 2B and 5B models built with a 3D causal VAE that cover text-to-video, video continuation, and image-to-video, and CogVideoX1.5 added 10-second, higher-resolution clips. On Zhipu's BigModel API, CogVideoX-3 makes 5- or 10-second videos at up to 4K from text, images, or a first and last frame, while CogVideoX-Flash is a free video generation model.
Decision facts
- Best fit
- video generationvision and multimodal work
- Capability starting point
- CogVideoX-3 with multimodal inputs
- Lowest tracked input
- Not tracked
- Closest related family
- AutoGLM
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs video generation, multimodal inputs, and video.
Use when the workload needs video generation, multimodal inputs, and video.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| CogVideo | Use when the workload needs 9.4 parameters. | 2022-12 | 9.4 parameters | Current |
| CogVideoX-3 | Use when the workload needs video generation, multimodal inputs, and video. | Unknown release | video generationmultimodal inputsvideo | Current |
| CogVideoX-Flash | Use when the workload needs video generation, multimodal inputs, and video. | Unknown release | video generationmultimodal inputsvideo | Current |
Release Timeline
2 release groupsSpecifications(3 models)
| Model | Released | Parameters | Vision | Multimodal |
|---|---|---|---|---|
| CogVideo | 2022-12 | 9.4 | No | No |
| CogVideoX-3 | — | — | Yes | Yes |
| CogVideoX-Flash | — | — | Yes | Yes |



