GLM-4.5V

Z.AI (formerly Zhipu) announced GLM-4.5V on August 13, 2025, as an open-source vision-language model engineered for multimodal reasoning across images, video, long documents, charts, and GUI screens. The model uses a 106B total / 12B active MoE architecture designed to pair high accuracy with practical latency and deployment cost. It follows the earlier GLM-4.1V-9B-Thinking release and scales that approach to enterprise workloads while keeping developer ergonomics central. Built on the GLM-4.5-Air text base and extending the GLM-4.1V-Thinking lineage, GLM-4.5V delivers SOTA performance among similarly sized open-source VLMs across 41 public multimodal evaluations. Access channels include Hugging Face, GitHub, the Z.AI API platform, and Z.AI Chat, ensuring broad developer reach. Capabilities span image reasoning with localization, video shot segmentation and event recognition, GUI tasks including screen reading and icon detection, complex chart and long-document analysis, and precise spatial grounding with bounding coordinates.

Zhipu AI released and open-sourced GLM-4.5V on August 11, 2025, describing it as the best-performing open-source visual reasoning model at the 100B parameter scale. The model is built on Zhipu's GLM-4.5-Air foundation and follows the technical approach of GLM-4.1V-Thinking, with weights published on both Hugging Face and ModelScope under an MIT-style open license. It achieved state-of-the-art results among open-source models across 41 public visual multimodal benchmarks spanning image reasoning, video understanding, GUI agent tasks, and document parsing. GLM-4.5V uses a Mixture-of-Experts architecture with 106B total parameters and 12B activated parameters, supporting a 66K token context window. A switchable "thinking mode" lets users toggle between fast response and deep reasoning for different latency-versus-quality tradeoffs. Zhipu also open-sourced a desktop assistant application that captures screenshots and screen recordings to feed GLM-4.5V for tasks like code assistance, video analysis, and document interpretation. The API is available on Zhipu's BigModel.cn platform with a free token package for new and existing users.

GLM-4.5V is Z.AI's visual reasoning model built on a Mixture-of-Experts architecture with 106B total parameters and 12B activation parameters, achieving state-of-the-art performance among open-source VLMs across 42 public visual multimodal benchmarks covering image, video, document understanding, and GUI tasks. The model continues the GLM-4.1V-Thinking technology route, extending it on the GLM-4.5-Air text base. Pricing is set at $0.6 per million input tokens and $1.8 per million output tokens, with a maximum output of 16K tokens. GLM-4.5V accepts video, image, text, and file inputs while producing text output, and supports both thinking and non-thinking modes. Documented capabilities include one-click webpage code generation from screenshots, precise visual grounding with bounding coordinates, GUI agent command execution, deep interpretation of multi-page documents, image recognition with built-in reasoning, and long video parsing. The model is accessible through Z.AI's developer API at the documented endpoint for integration into multimodal pipelines.