Holo1 GUI automation VLMs power Surfer-H web agent — open-source models on Hugging Face

AI Impact Summary

Holo1 introduces open-weight GUI automation VLMs (Holo1-3B, Holo1-7B) and Surfer-H, a browser-native agent that orchestrates reading, thinking, clicking, scrolling, typing, and validation through a modular Policy/Localizer/Validator architecture. By releasing these models and the WebClick benchmark on Hugging Face, the solution offers cost-efficient, real-world GUI automation with 92.2% accuracy at ~$0.13 per task, reducing reliance on proprietary APIs. This enables teams to deploy scalable web automation within browsers using open-source weights and standard transformer tooling (Qwen2.5-VL+Transformers compatibility).