GLM-4.5V

GLM-4.5V is a vision-language model built upon ZhipuAI's flagship GLM-4.5-Air text foundation model, featuring 106 billion total parameters with 12 billion active parameters during inference. The model continues the technical approach established by GLM-4.1V-Thinking, designed specifically to move beyond basic multimodal perception toward enhanced reasoning capabilities. This architecture enables complex problem solving, long-context understanding, and the development of capable multimodal agents, positioning it as a versatile platform for demanding visual-language tasks.

Training leverages scalable reinforcement learning techniques to cultivate versatile multimodal reasoning abilities, which contributed to the model achieving state-of-the-art performance among similarly scaled models on 42 public vision-language benchmarks. As an open-source release from the GLM-V team, GLM-4.5V supports both thinking and non-thinking modes, allowing developers to trade off depth and speed depending on task requirements. The model is available through Hugging Face with a desktop assistant demo application and an online chat interface, making it accessible for both research exploration and practical application development.