GLM-4.5V

Z.ai (Zhipu AI) released GLM-4.5V alongside GLM-4.5 and GLM-4.5-Air as a new open-source lineup for reasoning, coding, and agent tasks. The multimodal GLM-4.5V adds image and video analysis, can reconstruct websites from screenshots, and performs screen operations for autonomous agents. Users can test these capabilities for free at chat.z.ai after logging in. GLM-4.5V supports a hybrid approach with two modes: a "think mode" for complex reasoning and a "quick response mode" for faster answers. Zhipu AI claims GLM-4.5V delivers the strongest performance among open-source models of a similar size. The release is part of Zhipu's broader push highlighted by its billion-dollar valuation as a rising Chinese open-source AI lab.

Z.ai officially announced GLM-4.5V on August 14, 2025 as an open-source vision-language model designed for robust multimodal reasoning across images, video, long documents, charts, and GUI screens. The 100B-class architecture delivers 106B total parameters with 12B active, pairing high accuracy with practical latency and deployment cost for enterprise workloads. The launch followed July's GLM-4.1V-9B-Thinking, which surpassed 130,000 downloads and hit #1 on Hugging Face Trending. GLM-4.5V scales that recipe with SOTA performance across 41 public multimodal evaluations and is accessible through Hugging Face, GitHub, the Z.ai API platform, and Z.ai Chat. Key capabilities include visual grounding with bounding-box coordinates, GUI task automation, and complex chart and document analysis.

GLM-4.5V is part of Z.ai's GLM-V multimodal family and is based on the GLM-4.5-Air foundation model, targeting image, video, and document understanding as well as GUI agent operations. It uses a Mixture-of-Experts Transformer architecture with 106B total parameters and 12B activated parameters. The model was created on August 13, 2025, and is released under an MIT license. The model supports a 66K context length with FP8 precision and is designed for multimodal reasoning applications such as intelligent GUI automation, deep document and chart analysis, and visual QA with anomaly detection. It is described as achieving state-of-the-art performance on its target tasks. The Hugging Face model card lists Z.ai as the model provider.

GLM-4.5V was released on August 11, 2025, as part of Z.ai's (formerly Zhipu AI) model lineup, described as a 100B-scale open vision reasoning model. The release timeline places it after the July 28, 2025 debut of GLM-4.5 and GLM-4.5-Air, and before the September 30, 2025 release of GLM-4.6. It predates the December 8, 2025 launch of the successor GLM-4.6V vision model. Unlike the flagship GLM-4.5/4.5-Air which shipped under an MIT license, GLM-4.5V is listed as Open weights on this third-party timeline. The model is positioned as Z.ai's vision-language entry in the broader GLM family that has seen roughly bi-monthly flagship releases throughout 2026, establishing GLM-4.5V as the first major vision model in the GLM-4.5 generation.

GLM-4.5V is ZhipuAI's open-source vision-language model built on the GLM-4.5-Air text foundation, featuring 106B total parameters with 12B active in a Mixture-of-Experts architecture. According to the official model card on Hugging Face, it extends the GLM-4.1V-Thinking lineage and achieves state-of-the-art performance among same-scale models on 42 public vision-language benchmarks. The model targets full-spectrum visual reasoning including image analysis, video understanding, GUI agent tasks, long-document parsing, and precise visual grounding. The card notes efficient hybrid training enabling handling of diverse visual content and real-world usability on noisy, high-resolution, and extreme-aspect-ratio inputs, with links to the ZhipuAI Open Platform, the GitHub GLM-V repository, and the technical report.