Overview
WebVIA-Agent is a vision-language web exploration model in the WebVIA framework, built on GLM-4.1V-9B-Base. It navigates websites and collects screenshots of multiple UI states to support interactive UI-to-code workflows. (huggingface.co)
Capabilities and uses
- Uses page screenshots and DOM information to identify interactive components and guide exploration.
- Supports collecting UI states and interaction traces for reconstructing interactive web interfaces. In the broader framework, WebVIA-UI2Code generates HTML, CSS and JavaScript, while a separate validation module checks behavior; those functions should not be attributed to WebVIA-Agent alone. (huggingface.co)
- The published model card lists downloadable weights under the MIT license and provides local inference examples. (huggingface.co)
Limitations
The research paper describes the exploration pipeline as limited to click, enter and select actions, rather than drag or draw. Training primarily on synthetic webpages may also limit performance on specialized real-world interfaces, such as calculators and function-plotting tools. Results should be checked before use. (arxiv.org)
Release date
November 12, 2025 is the date of the supplied official repository commit; the commit alone does not establish a separate announcement date. (huggingface.co)
Pricing
Official input and output API pricing for this specific model could not be verified on either the domestic BigModel pricing page or the international Z.ai pricing page, checked November 17, 2026. The pages do not list WebVIA-Agent; prices for other GLM models are not assumed to apply. Self-hosting may incur infrastructure costs. (docs.bigmodel.cn)