The realm of personal agents is currently witnessing a significant upswing, with tech enterprises across the globe vying to launch associated products. Concurrently, the models are undergoing rapid iteration to bolster the practical applications of these personal agents. Meta's unveiling of the Muse model signaled the commencement of this trend. Subsequently, the introduction of the Muse Wonder series models, boasting proactive memory and focus functionalities, effectively tackled challenges related to the long-term retention and retrieval of information, as well as the sustained progression of goals for personal agents.

Alibaba's Qianwen has also made strides by introducing the online reinforcement learning framework, QwenGyre, tailored for ultra-long-term agent tasks. This framework addresses issues of computational inefficiency and redundant sample calculations during prolonged training sessions, thereby substantially boosting training efficiency. It has already been implemented in the training of the Qwen 3.8 series models.

The industry widely anticipates that personal agents will steer adjustments in the trajectory of model iteration. This shift will not only emphasize the preservation of general capabilities but also place a stronger emphasis on comprehending user needs and persistently advancing tasks assigned by users. The ultimate goal is to achieve a user experience that minimizes the necessity for repeated information input and eliminates the requirement for constant supervision.