Original sourceThe Decoder
Summary
OpenMobile-2 has released an open-source smartphone agent, including two agent models (Qwen3.5-9B and Qwen3.6-27B) plus a full training environment, data and operating instructions for reproducing results. The training environment simulates more than 110 apps, and the research found that app divers…
Key points
- Teams researching phone agents get a reproducible training environment and data, and insight into how app diversity affects generalization.
- It shows open-source phone agents can approach proprietary models when trained on diverse apps, with the full training pipeline released for reproduction.
- Developers can use its training environment and data to test agent performance on real Android tasks, lowering the cost of building simulation environments themselves.
Editorial note
This page is Code & Chain's editorial summary of public sources. It may be prepared with AI assistance and published through an automated workflow. Refer to the original sources; this content is not investment, legal, or tax advice.