← All models

CogVideoX-3

Published Publisher Zhipu AIcogvideox-3

Model Overview

CogVideoX-3 is Zhipu AI's video generation model. It takes text or images as input and produces video. The official launch was announced on July 15, 2025. (docs.bigmodel.cn)

Capabilities and Uses

  • Supports text-to-video and image-to-video generation, as well as generating transition videos using first- and last-frame images.
  • The company says this version improves visual stability, clarity, large-motion performance, instruction following, and physical simulation, as well as performance in photorealistic and 3D scenes. These are official claims and do not guarantee the results of every generation.
  • AI sound effects can optionally be generated and are disabled by default.
  • Can be used to produce e-commerce advertisements, cultural tourism short films, animation, film and television content, and short-video assets. (docs.bigmodel.cn)

Specifications and Limitations

  • Video duration can be set to 5 or 10 seconds, with a default of 5 seconds. Multiple resolutions are supported, up to 3840×2160, and the frame rate can be set to 30 or 60 fps.
  • Prompts must not exceed 512 characters; the prompt and image cannot both be empty. Images can be PNG, JPEG, or JPG, with a maximum size of 5 MB each; first- and last-frame mode requires two input images.
  • The video generation API is asynchronous, and the task result must be queried. Generated content may contain inaccurate or unexpected visuals and should be reviewed before use. (docs.bigmodel.cn)

Pricing and Billing

Date checked: November 18, 2026. The following prices are published separately by different official platforms and should not be regarded as the same plan convertible at an exchange rate:

Official Sources