AI TRENDS | Baidu’s DuMate Cuts Token Use 75% After Core Engine Upgrade

Baidu’s DuMate said it completed a core engine upgrade that reduced token consumption during task execution by 75%, with users’ points consumption also falling by 75%. According to Jin10, DuMate said the reduction was achieved through its Harness engine and ongoing engineering-level tuning, while maintaining the agent’s intelligence capabilities and task execution performance. DuMate said this marked the first time a domestic general-purpose intelligent agent product has delivered such a large drop in task-related consumption through the Harness engine and engineering optimization.