Add model card for OpenMobile

README.md

@@ -0,0 +1,30 @@
+---
+library_name: transformers
+pipeline_tag: image-text-to-text
+---
+
+# OpenMobile Qwen3-VL
+
+This repository contains the fine-tuned Qwen3-VL model presented in the paper [OpenMobile: Building Open Mobile Agents with Task and Trajectory Synthesis](https://huggingface.co/papers/2604.15093).
+
+OpenMobile is an open-source framework designed to synthesize high-quality task instructions and agent trajectories for mobile agents. It addresses the data gap in the field by introducing:
+1. **Scalable Task Synthesis**: A pipeline that constructs global environment memory from exploration to generate diverse and grounded instructions.
+2. **Policy-Switching Strategy**: A trajectory rollout method that captures essential error-recovery data by alternating between learner and expert models.
+
+## Resources
+- **Paper:** [OpenMobile: Building Open Mobile Agents with Task and Trajectory Synthesis](https://huggingface.co/papers/2604.15093)
+- **Project Page:** [https://njucckevin.github.io/openmobile/](https://njucckevin.github.io/openmobile/)
+- **Code:** [https://github.com/njucckevin/OpenMobile-Code](https://github.com/njucckevin/OpenMobile-Code)
+
+## Performance
+Agents trained using the OpenMobile framework achieve competitive results across dynamic mobile agent benchmarks. Notably, this fine-tuned Qwen3-VL checkpoint reaches **64.7%** success on AndroidWorld, significantly surpassing existing open-data approaches.
+
+## Citation
+```bibtex
+@article{openmobile2025,
+  title={OpenMobile: Building Open Mobile Agents with Task and Trajectory Synthesis},
+  author={Wu, Zhiyong and others},
+  journal={arXiv preprint arXiv:2604.15093},
+  year={2025}
+}
+```