On March 3, HONOR introduced MagicAgent, the first foundational model in the industry specifically crafted for intelligent agents. This model is engineered to facilitate general-purpose planning across a wide array of scenarios and is adept at managing heterogeneous task orchestration. The collaborative effort between HONOR and Fudan University has yielded a model boasting 30 billion parameters, which surpasses the performance of commercial models with hundreds of billions of parameters, such as GPT-5.2, across six distinct task benchmarks. MagicAgent empowers the YOYO intelligent agent to autonomously handle task planning and orchestration.

As a significant advancement within HONOR's 'Alpha Strategy,' MagicAgent tackles the challenges associated with heterogeneous task conflicts and limited data availability. It achieves this by implementing a streamlined synthetic data framework and an 'SFT+RL' training methodology. This approach strikes a balance between exploration and exploitation efficiency in dynamic settings. The model has already been integrated into commercial products like the HONOR Magic8 and Magic V6, where it supports intricate business processes such as schedule management and tool utilization. By doing so, it lowers the threshold for AI adoption while simultaneously boosting the productivity of developers.