88Z AI, 26 June — Tencent Hunyuan has officially open-sourced PhoneBuddy, a mobile agent model with 4 billion parameters. Alongside the release, the team published five academic papers, including PhoneWorld and PhoneHarness, establishing a comprehensive technical framework covering environment construction, training, execution, and privacy security. The model's core innovation lies in its Real+Mock hybrid reinforcement learning strategy, which combines operations in real apps with a self-built PhoneWorld simulati
At a glance
- Tencent Hunyuan open-sources the 4B-parameter mobile agent model PhoneBuddy, accompanied by five papers covering the full technical pipeline.
- Utilizes Real+Mock hybrid RL training, surpassing GPT-5.4 in multiple metrics across 150 real-device evaluations.
The update
Tencent Hunyuan recently announced the open-source release of PhoneBuddy, a 4-billion-parameter agent model designed specifically for mobile devices. To comprehensively showcase its technical architecture, the team simultaneously released five related papers, with PhoneWorld and PhoneHarness being particularly notable. These research outputs systematically cover the entire pipeline from environment setup and model training to actual execution and privacy security protection. In terms of training methodology, PhoneBuddy innovatively introduces a Real+Mock hybrid reinforcement learning mechanism. This approach cleverly combines interaction data from real applications with a self-built PhoneWorld simulation environment, effectively addressing the contradictions between scarce training data and environmental complexity. Empirical data shows that in real-device evaluations comprising 150 test items, the model outperformed GPT-5.4 in multiple core metrics, demonstrating robust task processing capabilities on mobile platforms.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 量子位Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.