inclusionAI/VISTA-9B: a GUI grounding vision-language model trained with VISTA
Original title:inclusionAI/VISTA-9B
AI summary
VISTA-9B is a GUI grounding model trained on a Qwen3.5 9B backbone. It takes screenshots and natural language instructions as input and outputs normalized coordinates from 0 to 1000. It uses the VISTA (view-consistent self-verification) method, including view-consistent GRPO and self-verified cross-view anchoring. It scores 69.2, 95.8, 68.1, and 75.5 on SSPro, SSV2, OSWorld-G, and OSWorld-G-R, respectively, surpassing Qwen3.5-9B and GRPO-9B baselines. The model is open-source and can be loaded via HuggingFace.
Key Takeaways
- VISTA-9B is a GUI grounding model trained on a Qwen3.
- 5 9B backbone.
- It takes screenshots and natural language instructions as input and outputs normalized coordinates…
- It uses the VISTA (view-consistent self-verification) method,including view-consistent GRPO and…
- It scores 69.
Quick Facts
- Published
- Updated
- Category
- Model Releases
- AI score
- 71
蚂蚁 inclusionAI 的 VISTA-9B 专攻 GUI 定位,把截图和指令直接变成点击坐标,做界面自动化的开发者可以直接拿来用。训练中的视角一致性约束有点新意,但基准提升不大,更像个实干范本。
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/VISTA-9B: a GUI grounding vision-language model trained with VISTA. June 12, 2026. https://new.nat6.net/en/news/cmqbwmfgh05sdslam91ctu2zd/