On June 15, Baidu's enterprise-level general AI agent product "Baidu Duma" announced the completion of its core engine upgrade. For the first time in China, a general AI agent has achieved a significant reduction in task consumption through the Harness engine and engineering optimizations.
This upgrade successfully reduced the Token consumption during task execution by 75% while ensuring that the agent's intelligent capabilities and task execution effects remain unaffected. Through continuous optimization at the engineering level and the use of the Harness engine, the corresponding user point consumption also decreased by 75%. As an enterprise-level general AI agent launched by Baidu in March 2026, Duma supports local deployment and aims to provide secure and compliant AI services for enterprise users.
In the current highly competitive landscape of large model application deployment, Token consumption often represents the core cost bottleneck restricting enterprises from large-scale deployment of AI applications. By deeply optimizing the underlying engine and engineering aspects, Baidu Duma has found the best balance between "performance not reduced and costs significantly lowered."
