ShowUI-2B

ShowUI is a compact vision-language-action model from Show Lab, a collaboration spanning the National University of Singapore and Microsoft, described in arXiv 2411.17465 and presented at CVPR 2025. The 2B-parameter checkpoint was released on Hugging Face in November 2024, and the design centers on UI-guided token selection plus interleaved action history, letting a small model perceive screenshots and emit the next GUI action.

ShowUI targets the computer-use agent stack: mobile, web and desktop automation driven purely by screenshots rather than DOM or accessibility trees. One third-party listing reports an MIT license, but confirm on the model card before commercial use. It remains one of the lightest credible open GUI-agent models, and it integrates with the project's Computer Use OOTB runtime for driving a real PC.