CogView Models by Zhipu AI
Last refreshed 2026-07-11. Next refresh: weekly.
About
CogView is Zhipu AI's family of text-to-image models. CogView3 uses cascaded relay diffusion, while CogView-3Plus moved to a Diffusion Transformer design. CogView4 is a 6B open-weight model that replaces the English-only T5 text encoder with Zhipu's bilingual GLM-4 encoder, so it understands Chinese and English prompts and can write Chinese characters inside images; Zhipu calls it its first open-source text-to-image model able to do so. It generates images at many resolutions, from 512 to 2048 pixels per side. On Zhipu's BigModel platform, the family is offered as the CogView-4 API and the free CogView-3-Flash model.
Decision facts
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 image generation and vision.
Use when the workload needs image generation and vision.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| CogView 2 | Use when the workload needs 6B parameters. | 2022-10 | 6B parameters | Current |
| CogView-3-Flash | Use when the workload needs image generation and vision. | Unknown release | image generationvision | Current |
| CogView-4 | Use when the workload needs image generation and vision. | Unknown release | image generationvision | Current |
Release Timeline
2 release groupsSpecifications(3 models)
| Model | Released | Parameters |
|---|---|---|
| CogView 2 | 2022-10 | 6B |
| CogView-3-Flash | — | — |
| CogView-4 | — | — |



