inclusionAI releases VISTA-4B GUI grounding vision-language model
Original title:inclusionAI/VISTA-4B
AI summary
VISTA-4B is a GUI grounding model based on the Qwen3.5-4B backbone. It takes screenshots and natural language instructions as input and outputs normalized 0-1000 coordinates. Training uses view-consistent GRPO and self-verified cross-view anchoring. On GUI grounding benchmarks, it scores 64.2 on SSPro (up 2.0 vs. GRPO-4B), 93.8 on SSV2 (down 0.4), 61.2 on OSWorld-G (up 1.3), and 69.7 on OSWorld-G-R (up 0.5). The model is open-source on HuggingFace, and users are recommended to use prompts and return coordinates in [x,y] format.
Key Takeaways
- VISTA-4B is a GUI grounding model based on the Qwen3.
- 5-4B backbone.
- It takes screenshots and natural language instructions as input and outputs normalized 0-1000…
- Training uses view-consistent GRPO and self-verified cross-view anchoring.
- On GUI grounding benchmarks,it scores 64.
Quick Facts
- Published
- Updated
- Category
- Model Releases
- AI score
- 62
蚂蚁 inclusionAI 开源了一款 GUI 定位模型,基于 Qwen3.5 微调,在接地基准上小幅提升,关键是提供了自验证训练方法,做桌面自动化的可以直接下载用。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: inclusionAI releases VISTA-4B GUI grounding vision-language model. June 12, 2026. https://new.nat6.net/en/news/cmqbwmfgh05seslam0r0r80vz/