OpenMobile-2 open mobile agent trained on 110-plus apps
Original title: OpenMobile-2: Building Versatile Mobile Agents with Scalable Environments and App-Native Tools
Authors report AndroidWorld 79.9, MobileWorld 50.4, and SPA-Bench rising from 31.9 to 59.6.
OpenMobile-2 is an open mobile agent with public training environments and recipes. MobileGym++ adds 35 commercial-style apps and, with Android emulator apps, a playground of more than 110 apps, plus nearly 12K supervised trajectories and more than 2K verifiable RL tasks. A hybrid setting lets agents mix GUI actions with more than 300 app-native tools across 50-plus apps. The authors say it is competitive on AndroidWorld and MobileWorld and nearly doubles SPA-Bench on real devices.
Unconfirmed: Benchmark scores are the authors' reported results, not an independent evaluation.
- MobileGym++ includes 35 simulated commercial-style apps with cross-app workflows.
- Open RL data covers more than 2K executable tasks with automatic rewards.
- MobileGym++ Bench evaluates GUI-only and hybrid GUI-tool use.
- Resources are posted at the OpenMobile-2 project site.
- AI models
- OpenMobile-2
From the original article
1 min read at huggingface.co
We introduce OpenMobile-2, a near-frontier mobile agent with fully open training environments and recipes. We make three key advances: (1) Diverse environments with simulated commercial apps: We build MobileGym++, featuring 35 realistic, functionally rich commercial-style apps with cross-app workflows, while preserving full controllability for reset and verification.
By Chenyang Yan, Yingying Zhang, Kanzhi Cheng · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.