GLM-4.5V

GLM-4.5V is Z.ai's (formerly Zhipu AI) open vision-language model, released August 2025. It builds on the 106B-parameter GLM-4.5-Air backbone as a mixture-of-experts with 12B active parameters per token, which keeps inference costs practical for a model this size. It handles images, video, long documents, charts, and GUI screens through one pipeline, with 64K context and support for 4K-resolution images at any aspect ratio. A Thinking Mode toggle switches between fast answers and step-by-step reasoning. Z.ai reports state-of-the-art results among similarly sized open VLMs across 41 public multimodal evaluations, plus strong numbers on MMBench and AI2D; these are team-reported. The MIT license permits commercial use and redistribution, which is unusually permissive for a frontier-class VLM. Access is through Hugging Face, GitHub, the Z.ai API platform, and Z.ai chat. Note that Z.ai was placed on the US entity list in early 2025, which is worth knowing for procurement and compliance reviews.

Team-reported: SOTA among open VLMs on 41 public multimodal evaluations