GLM-4.5V
Release date: August 11, 2025. (huggingface.co)
Model positioning
GLM-4.5V is a vision-language model for multimodal understanding and visual reasoning. Based on GLM-4.5-Air, it uses a mixture-of-experts (MoE) architecture, with 106 billion total parameters and 12 billion active parameters. The model weights have been officially released, and the model card lists the license as MIT. (docs.z.ai)
Capabilities and uses
- Accepts text, image, video, and file inputs and outputs text; the international API documentation lists a maximum output of 16K tokens.
- Supports image and multi-image reasoning, video event understanding, visual object localization, and chart and long-document analysis.
- Can be used for screen understanding and GUI agent-assisted operations, generating web page code from page screenshots, document information extraction, and image-text question answering.
- Supports switching between thinking and non-thinking modes to balance response speed and reasoning depth. (docs.z.ai)
Limitations
The official sources note that pure text question answering still has room for improvement; complex prompts may lead to overthinking, repeated reasoning, or restating answers; and visual perception tasks such as counting and identifying specific people also have limitations. Important results and automated operations should be verified. (github.com)
API pricing
Z.AI international platform pay-as-you-go pricing, US dollars (USD): $0.60 per million input tokens and $1.80 per million output tokens; cache-hit input is $0.11 per million tokens, and cache storage is free for a limited time. Checked on: November 18, 2026. Official international pricing page. (docs.z.ai)
The official Zhipu domestic pricing page was also checked. The page requires JavaScript, and the current retrieved result did not display prices, so its pricing could not be verified. (open.bigmodel.cn)