CogView4

CogView4 is Zhipu AI's 6-billion-parameter text-to-image model, released in March 2025 as the company's first open-source image model of that year. Its headline capability is generating legible Chinese characters inside images, a long-standing weak point of diffusion models, alongside strong bilingual Chinese-English prompt understanding.

Zhipu reports that CogView4 topped the DPG-Bench benchmark among open-source text-to-image models at release, with support for arbitrary resolutions and complex semantic alignment aimed at advertising and short-video production. The weights are released under Apache 2.0 with diffusers integration, an HF Space demo, and Replicate availability. It is a genuine open-weights release, not an API-only product, which distinguishes it from most frontier image models.