Zhipu AI Releases Open Vision Model GLM-4.5V

Chinese AI firm Zhipu AI has released GLM-4.5V, a new open-source vision-language model (VLM). The model, which uses a Mixture-of-Experts (MoE) architecture, is designed for sophisticated tasks that require understanding and reasoning about both text and images simultaneously.

According to the release notes, GLM-4.5V is built upon the company's GLM-4.5-Air-Base model. The key advancement is its capacity for what Zhipu AI describes as strong multimodal reasoning. This makes it suitable for complex applications like detailed image analysis, visual question answering, and generating text grounded in visual information. The model weights and code are available now on Hugging Face.

Why it matters

The release is significant for two main reasons. First, it adds a powerful, openly accessible VLM to the ecosystem, a domain where proprietary models have often dominated. Second, its release under the permissive MIT license removes significant barriers for both commercial and research applications, allowing developers to freely build upon and integrate the technology.

The MoE architecture also suggests an efficient design, capable of activating only the necessary expert sub-networks during inference. This can lead to faster performance and lower computational costs compared to dense models of a similar capability level, making advanced multimodal AI more accessible to a wider range of developers and organizations.